docs: document structure-element extraction and TextItem.mcid

* chore(release): unify package versions

* fix(release): harden version synchronization

.github/workflows/ci.yml

@@ -14,44 +14,57 @@ jobs:
     name: Test
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
 
       - name: Cache cargo
-        uses: Swatinem/rust-cache@v2
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
+
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
 
       - name: Run tests
         run: cargo test --verbose
 
+      - name: Test developer scripts
+        run: python3 -m unittest discover -s scripts/tests
+
   fmt:
     name: Format
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
         with:
+          toolchain: stable
           components: rustfmt
 
       - name: Check formatting
         run: cargo fmt --all -- --check
 
+      - name: Check WASM formatting
+        run: cargo fmt --manifest-path wasm/Cargo.toml -- --check
+
   clippy:
     name: Clippy
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
         with:
+          toolchain: stable
           components: clippy
 
       - name: Cache cargo
-        uses: Swatinem/rust-cache@v2
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
         with:
           key: clippy
 
@@ -65,15 +78,52 @@ jobs:
       matrix:
         os: [ubuntu-latest, macos-latest]
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
 
       - name: Cache cargo
-        uses: Swatinem/rust-cache@v2
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
         with:
           key: build
 
       - name: Build
         run: cargo build --release --verbose
+
+  wasm:
+    name: WebAssembly
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
+          targets: wasm32-unknown-unknown
+          components: clippy
+
+      - name: Cache cargo
+        uses: Swatinem/rust-cache@c19371144df3bb44fab255c43d04cbc2ab54d1c4 # v2.9.1
+        with:
+          workspaces: |
+            wasm -> target
+          key: wasm
+
+      - name: Check WebAssembly bindings
+        run: cargo check --manifest-path wasm/Cargo.toml --target wasm32-unknown-unknown
+
+      - name: Check root package for WebAssembly
+        run: cargo check --target wasm32-unknown-unknown
+
+      - name: Lint WebAssembly bindings
+        run: cargo clippy --manifest-path wasm/Cargo.toml --target wasm32-unknown-unknown -- -D warnings
+
+      - name: Install wasm-pack
+        run: cargo install wasm-pack --version 0.15.0 --locked
+
+      - name: Test WebAssembly package
+        run: wasm-pack test --node --release wasm

.github/workflows/pages.yml

@@ -0,0 +1,36 @@
+name: Deploy landing page
+
+on:
+  push:
+    branches: [main]
+    paths: ['site/**', '.github/workflows/pages.yml']
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+# Allow one concurrent deployment; don't cancel an in-progress production deploy.
+concurrency:
+  group: pages
+  cancel-in-progress: false
+
+jobs:
+  deploy:
+    name: Build & deploy to GitHub Pages
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deploy.outputs.page_url }}
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Upload site artifact
+        uses: actions/upload-pages-artifact@fc324d3547104276b827a68afc52ff2a11cc49c9 # v5.0.0
+        with:
+          path: site
+
+      - name: Deploy to GitHub Pages
+        id: deploy
+        uses: actions/deploy-pages@cd2ce8fcbc39b97be8ca5fce6e763baed58fa128 # v5.0.0

.github/workflows/publish-crate.yml

@@ -0,0 +1,108 @@
+name: Publish Rust crate
+
+on:
+  push:
+    branches: [main]
+    paths: ['Cargo.toml']
+  # Manual fallback: retry a publish that failed after the version was
+  # already merged (a plain re-push won't register as a version change).
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  check-version:
+    name: Check version change
+    # Guard manual dispatches: crates.io trusted publishing matches
+    # repo+workflow+environment but NOT branch, so without this a
+    # workflow_dispatch from any branch could publish unmerged code.
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    outputs:
+      changed: ${{ steps.check.outputs.changed }}
+      published: ${{ steps.check.outputs.published }}
+      version: ${{ steps.check.outputs.version }}
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          fetch-depth: 2
+
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
+      - name: Check if version changed
+        id: check
+        run: |
+          NEW_VERSION=$(python3 -c 'import pathlib, tomllib; print(tomllib.loads(pathlib.Path("Cargo.toml").read_text())["package"]["version"])')
+          echo "version=$NEW_VERSION" >> "$GITHUB_OUTPUT"
+
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            # Manual dispatch publishes the current version regardless of the
+            # previous commit; the crates.io check below still prevents
+            # double-publishing an already-released version.
+            echo "manual dispatch: publishing v$NEW_VERSION"
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          else
+            OLD_VERSION=$(git show HEAD~1:Cargo.toml | python3 -c 'import sys, tomllib; print(tomllib.loads(sys.stdin.read())["package"]["version"])')
+            echo "old=$OLD_VERSION new=$NEW_VERSION"
+
+            if [ "$NEW_VERSION" = "$OLD_VERSION" ]; then
+              echo "changed=false" >> "$GITHUB_OUTPUT"
+              echo "published=false" >> "$GITHUB_OUTPUT"
+              exit 0
+            fi
+
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          fi
+
+          HTTP_STATUS=$(curl --silent --show-error --output /tmp/crate-version.json --write-out "%{http_code}" \
+            -H "User-Agent: firecrawl/pdf-inspector publish workflow (https://github.com/firecrawl/pdf-inspector)" \
+            "https://crates.io/api/v1/crates/pdf-inspector/$NEW_VERSION")
+
+          case "$HTTP_STATUS" in
+            200)
+              echo "published=true" >> "$GITHUB_OUTPUT"
+              echo "pdf-inspector v$NEW_VERSION is already published"
+              ;;
+            404)
+              echo "published=false" >> "$GITHUB_OUTPUT"
+              ;;
+            *)
+              cat /tmp/crate-version.json
+              echo "Unexpected crates.io response: $HTTP_STATUS" >&2
+              exit 1
+              ;;
+          esac
+
+  publish:
+    name: Publish to crates.io
+    needs: check-version
+    if: needs.check-version.outputs.changed == 'true' && needs.check-version.outputs.published == 'false'
+    runs-on: ubuntu-latest
+    environment: crates-io
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
+
+      - name: Verify package
+        run: cargo publish --dry-run
+
+      - name: Authenticate with crates.io
+        id: auth
+        uses: rust-lang/crates-io-auth-action@c6f97d42243bad5fab37ca0427f495c86d5b1a18 # v1.0.5
+
+      - name: Publish crate
+        run: cargo publish
+        env:
+          CARGO_REGISTRY_TOKEN: ${{ steps.auth.outputs.token }}

.github/workflows/publish-pypi.yml

@@ -0,0 +1,169 @@
+name: Publish Python package
+
+on:
+  push:
+    branches: [main]
+    paths: ['pyproject.toml']
+  # Manual fallback: re-publish the current version without a version bump
+  # (e.g. first run after PyPI trusted publishing is configured).
+  workflow_dispatch:
+
+permissions:
+  contents: read
+
+jobs:
+  check-version:
+    name: Check version change
+    # Guard manual dispatches too: PyPI trusted publishing matches
+    # repo+workflow+environment but NOT branch, so without this a
+    # workflow_dispatch from any branch could publish unmerged code.
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    outputs:
+      changed: ${{ steps.check.outputs.changed }}
+      published: ${{ steps.check.outputs.published }}
+      version: ${{ steps.check.outputs.version }}
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          fetch-depth: 2
+
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
+      - name: Check if version changed
+        id: check
+        run: |
+          NEW_VERSION=$(python3 -c 'import pathlib, tomllib; print(tomllib.loads(pathlib.Path("pyproject.toml").read_text())["project"]["version"])')
+          echo "version=$NEW_VERSION" >> "$GITHUB_OUTPUT"
+
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            # Manual dispatch always rebuilds and publishes. Combined with
+            # skip-existing on the publish step, this repairs partial releases
+            # (PyPI's version endpoint returns 200 even when only some of the
+            # expected wheels were uploaded).
+            echo "manual dispatch: publishing v$NEW_VERSION (skip-existing handles uploaded files)"
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+            echo "published=false" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          # .get(): the parent commit may predate the static version field
+          # (pyproject.toml used dynamic = ["version"]) — treat that as a change
+          # so the very first merge of this workflow publishes.
+          OLD_VERSION=$(git show HEAD~1:pyproject.toml | python3 -c 'import sys, tomllib; print(tomllib.loads(sys.stdin.read())["project"].get("version", ""))')
+          echo "old=$OLD_VERSION new=$NEW_VERSION"
+
+          if [ "$NEW_VERSION" = "$OLD_VERSION" ]; then
+            echo "changed=false" >> "$GITHUB_OUTPUT"
+            echo "published=false" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
+          echo "changed=true" >> "$GITHUB_OUTPUT"
+
+          HTTP_STATUS=$(curl --silent --show-error --output /tmp/pypi-version.json --write-out "%{http_code}" \
+            "https://pypi.org/pypi/pdf-inspector/$NEW_VERSION/json")
+
+          case "$HTTP_STATUS" in
+            200)
+              echo "published=true" >> "$GITHUB_OUTPUT"
+              echo "pdf-inspector v$NEW_VERSION is already published to PyPI"
+              ;;
+            404)
+              echo "published=false" >> "$GITHUB_OUTPUT"
+              ;;
+            *)
+              cat /tmp/pypi-version.json
+              echo "Unexpected PyPI response: $HTTP_STATUS" >&2
+              exit 1
+              ;;
+          esac
+
+  build:
+    needs: check-version
+    if: needs.check-version.outputs.changed == 'true' && needs.check-version.outputs.published == 'false'
+    name: Build ${{ matrix.target }}
+    runs-on: ${{ matrix.os }}
+    strategy:
+      matrix:
+        include:
+          - os: ubuntu-latest
+            target: x86_64-unknown-linux-gnu
+          - os: ubuntu-latest
+            target: aarch64-unknown-linux-gnu
+          # macos-13 was retired by GitHub; macos-15-intel is the remaining
+          # Intel runner label (available through 2027).
+          - os: macos-15-intel
+            target: x86_64-apple-darwin
+          - os: macos-14
+            target: aarch64-apple-darwin
+          - os: windows-latest
+            target: x86_64-pc-windows-msvc
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - uses: actions/setup-python@5fda3b95a4ea91299a34e894583c3862153e4b97 # v7.0.0
+        with:
+          python-version: '3.12'
+
+      - name: Build wheel
+        uses: PyO3/maturin-action@e83996d129638aa358a18fbd1dfb82f0b0fb5d3b # v1.51.0
+        with:
+          target: ${{ matrix.target }}
+          args: --release --out dist
+          manylinux: auto
+
+      - name: Upload wheel
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: wheels-${{ matrix.target }}
+          path: dist/*.whl
+          if-no-files-found: error
+
+  sdist:
+    needs: check-version
+    if: needs.check-version.outputs.changed == 'true' && needs.check-version.outputs.published == 'false'
+    name: Build sdist
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Build sdist
+        uses: PyO3/maturin-action@e83996d129638aa358a18fbd1dfb82f0b0fb5d3b # v1.51.0
+        with:
+          command: sdist
+          args: --out dist
+
+      - name: Upload sdist
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
+        with:
+          name: sdist
+          path: dist/*.tar.gz
+          if-no-files-found: error
+
+  publish:
+    name: Publish to PyPI
+    needs: [check-version, build, sdist]
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - name: Download all artifacts
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          path: dist
+          merge-multiple: true
+
+      - name: List artifacts
+        run: ls -la dist/
+
+      - name: Publish to PyPI
+        uses: pypa/gh-action-pypi-publish@dc37677b2e1c63e2034f94d8a5b11f265b73ba33 # release/v1
+        with:
+          packages-dir: dist
+          # Tolerate already-uploaded files so a manual re-run can complete
+          # a release that previously failed partway through.
+          skip-existing: true

.github/workflows/publish-wasm.yml

@@ -0,0 +1,115 @@
+name: Publish WebAssembly package
+
+on:
+  push:
+    branches: [main]
+    paths: ['wasm/Cargo.toml']
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  id-token: write
+
+env:
+  CARGO_TERM_COLOR: always
+
+jobs:
+  check-version:
+    name: Check version change
+    if: github.ref == 'refs/heads/main'
+    runs-on: ubuntu-latest
+    outputs:
+      changed: ${{ steps.check.outputs.changed }}
+      package_exists: ${{ steps.check.outputs.package_exists }}
+      published: ${{ steps.check.outputs.published }}
+      version: ${{ steps.check.outputs.version }}
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+        with:
+          fetch-depth: 2
+
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
+      - name: Check package version
+        id: check
+        run: |
+          NEW_VERSION=$(python3 -c 'import pathlib, tomllib; print(tomllib.loads(pathlib.Path("wasm/Cargo.toml").read_text())["package"]["version"])')
+          echo "version=$NEW_VERSION" >> "$GITHUB_OUTPUT"
+
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          elif git cat-file -e HEAD~1:wasm/Cargo.toml 2>/dev/null; then
+            OLD_VERSION=$(git show HEAD~1:wasm/Cargo.toml | python3 -c 'import sys, tomllib; print(tomllib.loads(sys.stdin.read())["package"]["version"])')
+            if [ "$NEW_VERSION" = "$OLD_VERSION" ]; then
+              echo "changed=false" >> "$GITHUB_OUTPUT"
+              echo "published=false" >> "$GITHUB_OUTPUT"
+              exit 0
+            fi
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+          fi
+
+          if ! npm view "@firecrawl/pdf-inspector-wasm" name >/dev/null 2>&1; then
+            echo "package_exists=false" >> "$GITHUB_OUTPUT"
+            echo "published=false" >> "$GITHUB_OUTPUT"
+            echo "The initial package must be published once before trusted publishing can be configured."
+            exit 0
+          fi
+
+          echo "package_exists=true" >> "$GITHUB_OUTPUT"
+          if npm view "@firecrawl/pdf-inspector-wasm@$NEW_VERSION" version >/dev/null 2>&1; then
+            echo "published=true" >> "$GITHUB_OUTPUT"
+          else
+            echo "published=false" >> "$GITHUB_OUTPUT"
+          fi
+
+  publish:
+    name: Build and publish
+    needs: check-version
+    if: needs.check-version.outputs.changed == 'true' && needs.check-version.outputs.package_exists == 'true' && needs.check-version.outputs.published == 'false'
+    runs-on: ubuntu-latest
+    permissions:
+      contents: read
+      id-token: write
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Install Rust
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
+          targets: wasm32-unknown-unknown
+
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
+        with:
+          node-version: '24'
+          registry-url: 'https://registry.npmjs.org'
+
+      - name: Install wasm-pack
+        run: cargo install wasm-pack --version 0.15.0 --locked
+
+      - name: Build browser package
+        run: wasm-pack build wasm --target web --scope firecrawl --out-dir pkg --release
+
+      - name: Prepare package metadata
+        run: |
+          node -e '
+            const fs = require("fs")
+            const path = "wasm/pkg/package.json"
+            const pkg = JSON.parse(fs.readFileSync(path, "utf8"))
+            pkg.name = "@firecrawl/pdf-inspector-wasm"
+            pkg.description = "Browser WebAssembly bindings for the pdf-inspector Rust PDF parser"
+            pkg.keywords = ["pdf", "pdf-parser", "webassembly", "wasm", "markdown", "rust", "firecrawl"]
+            pkg.repository = { type: "git", url: "https://github.com/firecrawl/pdf-inspector" }
+            pkg.homepage = "https://github.com/firecrawl/pdf-inspector/tree/main/wasm"
+            pkg.publishConfig = { access: "public" }
+            fs.writeFileSync(path, JSON.stringify(pkg, null, 2) + "\n")
+          '
+
+      - name: Inspect package contents
+        run: npm pack --dry-run ./wasm/pkg
+
+      - name: Publish package
+        run: npm publish ./wasm/pkg --provenance --access public

.github/workflows/publish.yml

@@ -4,6 +4,9 @@ on:
   push:
     branches: [main]
     paths: ['napi/package.json']
+  # Manual fallback: retry a publish that failed partway (per-package
+  # already-published checks make re-runs idempotent).
+  workflow_dispatch:
 
 permissions:
   contents: read
@@ -12,24 +15,41 @@ permissions:
 jobs:
   check-version:
     name: Check version change
+    # Guard manual dispatches: npm trusted publishing matches
+    # repo+workflow+environment but NOT branch, so without this a
+    # workflow_dispatch from any branch could publish unmerged code.
+    if: github.ref == 'refs/heads/main'
     runs-on: ubuntu-latest
     outputs:
       changed: ${{ steps.check.outputs.changed }}
       version: ${{ steps.check.outputs.version }}
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
         with:
           fetch-depth: 2
 
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
       - name: Check if version changed
         id: check
         run: |
           NEW_VERSION=$(node -p "require('./napi/package.json').version")
+          echo "version=$NEW_VERSION" >> "$GITHUB_OUTPUT"
+
+          if [ "${{ github.event_name }}" = "workflow_dispatch" ]; then
+            # Manual dispatch rebuilds and publishes the current version; the
+            # per-package already-published checks in the publish job skip
+            # anything that made it out in a previous partial run.
+            echo "manual dispatch: publishing v$NEW_VERSION"
+            echo "changed=true" >> "$GITHUB_OUTPUT"
+            exit 0
+          fi
+
           OLD_VERSION=$(git show HEAD~1:napi/package.json | node -p "JSON.parse(require('fs').readFileSync('/dev/stdin','utf8')).version")
           echo "old=$OLD_VERSION new=$NEW_VERSION"
           if [ "$NEW_VERSION" != "$OLD_VERSION" ]; then
             echo "changed=true" >> "$GITHUB_OUTPUT"
-            echo "version=$NEW_VERSION" >> "$GITHUB_OUTPUT"
           else
             echo "changed=false" >> "$GITHUB_OUTPUT"
           fi
@@ -44,31 +64,61 @@ jobs:
         include:
           - os: ubuntu-latest
             target: x86_64-unknown-linux-gnu
+          # napi-cross builds gnu targets against an old glibc sysroot for
+          # broad distro compatibility; musl targets cross-compile with
+          # zig via cargo-zigbuild (napi's -x flag).
+          - os: ubuntu-latest
+            target: aarch64-unknown-linux-gnu
+            build-flags: --use-napi-cross
+          - os: ubuntu-latest
+            target: x86_64-unknown-linux-musl
+            build-flags: -x
+          - os: ubuntu-latest
+            target: aarch64-unknown-linux-musl
+            build-flags: -x
           - os: macos-14
             target: aarch64-apple-darwin
           - os: windows-latest
             target: x86_64-pc-windows-msvc
     steps:
-      - uses: actions/checkout@v4
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
       - name: Install Rust
-        uses: dtolnay/rust-toolchain@stable
+        uses: dtolnay/rust-toolchain@4cda84d5c5c54efe2404f9d843567869ab1699d4 # stable
+        with:
+          toolchain: stable
+          targets: ${{ matrix.target }}
 
-      - uses: oven-sh/setup-bun@v2
+      - uses: oven-sh/setup-bun@0c5077e51419868618aeaa5fe8019c62421857d6 # v2.2.0
         with:
           bun-version: latest
 
+      - name: Install zig
+        if: contains(matrix.target, 'musl')
+        uses: mlugg/setup-zig@d1434d08867e3ee9daa34448df10607b98908d29 # v2.2.1
+        with:
+          version: 0.14.1
+
+      - name: Install cargo-zigbuild
+        if: contains(matrix.target, 'musl')
+        uses: taiki-e/install-action@67729d5c413db75907f0ad1e39bb04b9c868ff60 # v2.85.7
+        env:
+          GITHUB_TOKEN: ${{ github.token }}
+        with:
+          tool: cargo-zigbuild
+
       - name: Cache cargo
-        uses: actions/cache@v4
+        uses: actions/cache@55cc8345863c7cc4c66a329aec7e433d2d1c52a9 # v6.1.0
         with:
           path: |
             ~/.cargo/registry/index/
             ~/.cargo/registry/cache/
             ~/.cargo/git/db/
+            ~/.napi-rs
             napi/target/
-          key: ${{ runner.os }}-cargo-napi-${{ hashFiles('**/Cargo.lock') }}
+          key: ${{ matrix.target }}-cargo-napi-${{ hashFiles('**/Cargo.lock') }}
           restore-keys: |
-            ${{ runner.os }}-cargo-napi-
+            ${{ matrix.target }}-cargo-napi-
 
       - name: Install dependencies
         working-directory: napi
@@ -76,10 +126,10 @@ jobs:
 
       - name: Build native addon
         working-directory: napi
-        run: bunx napi build --platform --release
+        run: bunx napi build --platform --release --target ${{ matrix.target }} ${{ matrix.build-flags }}
 
       - name: Upload native binary
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: bindings-${{ matrix.target }}
           path: napi/*.node
@@ -87,7 +137,7 @@ jobs:
 
       - name: Upload generated JS bindings
         if: matrix.target == 'x86_64-unknown-linux-gnu'
-        uses: actions/upload-artifact@v4
+        uses: actions/upload-artifact@043fb46d1a93c77aae656e7c1c64a875d1fc6a0a # v7.0.1
         with:
           name: js-bindings
           path: |
@@ -95,34 +145,149 @@ jobs:
             napi/index.d.ts
           if-no-files-found: error
 
+  smoke-test:
+    name: Smoke test ${{ matrix.target }}
+    needs: [check-version, build]
+    runs-on: ${{ matrix.os }}
+    strategy:
+      fail-fast: false
+      matrix:
+        include:
+          - os: ubuntu-latest
+            target: x86_64-unknown-linux-gnu
+          - os: ubuntu-latest
+            target: x86_64-unknown-linux-musl
+          - os: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-gnu
+          - os: ubuntu-24.04-arm
+            target: aarch64-unknown-linux-musl
+          - os: macos-14
+            target: aarch64-apple-darwin
+          - os: windows-latest
+            target: x86_64-pc-windows-msvc
+    steps:
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
+
+      - name: Download native binary
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: bindings-${{ matrix.target }}
+          path: napi
+
+      - name: Download generated JS bindings
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
+        with:
+          name: js-bindings
+          path: napi
+
+      # musl binaries must load under a real musl libc, so run inside Alpine.
+      - name: Run smoke test (Alpine)
+        if: contains(matrix.target, 'musl')
+        run: docker run --rm -v "$PWD:/repo" -w /repo/napi node:24-alpine node test.mjs
+
+      - name: Run smoke test
+        if: ${{ !contains(matrix.target, 'musl') }}
+        working-directory: napi
+        run: node test.mjs
+
   publish:
     name: Publish to npm
-    needs: [check-version, build]
+    needs: [check-version, build, smoke-test]
     runs-on: ubuntu-latest
     permissions:
       contents: read
       id-token: write
     steps:
-      - uses: actions/checkout@v6
+      - uses: actions/checkout@3d3c42e5aac5ba805825da76410c181273ba90b1 # v7.0.1
 
-      - uses: actions/setup-node@v6
+      - uses: actions/setup-node@820762786026740c76f36085b0efc47a31fe5020 # v7.0.0
         with:
           node-version: '24'
           registry-url: 'https://registry.npmjs.org'
 
       - name: Download all artifacts
-        uses: actions/download-artifact@v4
+        uses: actions/download-artifact@3e5f45b2cfb9172054b4087a40e8e0b5a5461e7c # v8.0.1
         with:
           path: napi/artifacts
 
-      - name: Collect binaries and publish
+      - name: Publish platform packages
         working-directory: napi
         run: |
-          cp artifacts/bindings-*/*.node .
+          VERSION="${{ needs.check-version.outputs.version }}"
+
+          for node_file in artifacts/bindings-*/pdf-inspector.*.node; do
+            base=$(basename "$node_file")
+            suffix=${base#pdf-inspector.}
+            suffix=${suffix%.node}
+            pkg="@firecrawl/pdf-inspector-$suffix"
+
+            if npm view "$pkg@$VERSION" version >/dev/null 2>&1; then
+              echo "$pkg@$VERSION already published — skipping"
+              continue
+            fi
+
+            dir="npm-dist/$suffix"
+            mkdir -p "$dir"
+            cp "$node_file" "$dir/"
+            node -e '
+              const [suffix, version] = process.argv.slice(1)
+              const meta = {
+                "linux-x64-gnu":    { os: ["linux"],  cpu: ["x64"],   libc: ["glibc"] },
+                "linux-x64-musl":   { os: ["linux"],  cpu: ["x64"],   libc: ["musl"] },
+                "linux-arm64-gnu":  { os: ["linux"],  cpu: ["arm64"], libc: ["glibc"] },
+                "linux-arm64-musl": { os: ["linux"],  cpu: ["arm64"], libc: ["musl"] },
+                "darwin-arm64":     { os: ["darwin"], cpu: ["arm64"] },
+                "win32-x64-msvc":   { os: ["win32"],  cpu: ["x64"] },
+              }[suffix]
+              if (!meta) {
+                console.error(`unknown platform suffix: ${suffix} — add it to the meta map`)
+                process.exit(1)
+              }
+              const pkg = {
+                name: `@firecrawl/pdf-inspector-${suffix}`,
+                version,
+                description: `Prebuilt ${suffix} binary for @firecrawl/pdf-inspector`,
+                main: `pdf-inspector.${suffix}.node`,
+                files: [`pdf-inspector.${suffix}.node`],
+                license: "MIT",
+                engines: { node: ">= 10" },
+                repository: { type: "git", url: "https://github.com/firecrawl/pdf-inspector" },
+                publishConfig: { access: "public" },
+                ...meta,
+              }
+              require("fs").writeFileSync(`npm-dist/${suffix}/package.json`, JSON.stringify(pkg, null, 2) + "\n")
+            ' "$suffix" "$VERSION"
+
+            echo "=== $pkg@$VERSION ==="
+            ls -la "$dir"
+            (cd "$dir" && npm publish --provenance --access public)
+          done
+
+      - name: Publish main package
+        working-directory: napi
+        run: |
+          VERSION="${{ needs.check-version.outputs.version }}"
+
+          if npm view "@firecrawl/pdf-inspector@$VERSION" version >/dev/null 2>&1; then
+            echo "@firecrawl/pdf-inspector@$VERSION already published — skipping"
+            exit 0
+          fi
+
           cp artifacts/js-bindings/index.js .
           cp artifacts/js-bindings/index.d.ts .
 
-          echo "=== Package contents ==="
-          ls -la *.node index.js index.d.ts
+          # Stamp optionalDependencies to this exact version so the platform
+          # pins can never drift from the main package version.
+          node -e '
+            const fs = require("fs")
+            const pkg = JSON.parse(fs.readFileSync("package.json", "utf8"))
+            for (const dep of Object.keys(pkg.optionalDependencies ?? {})) {
+              pkg.optionalDependencies[dep] = pkg.version
+            }
+            fs.writeFileSync("package.json", JSON.stringify(pkg, null, 2) + "\n")
+          '
+
+          echo "=== Main package contents ==="
+          npm pack --dry-run
 
           npm publish --provenance --access public

.gitignore

@@ -1,10 +1,13 @@
 # Rust build artifacts
 /target/
+/wasm/target/
+/wasm/pkg/
 debug/
 *.pdb
 
 # Cargo lock (optional for libraries)
 Cargo.lock
+!/wasm/Cargo.lock
 
 # IDE
 .idea/
@@ -28,15 +31,14 @@ Thumbs.db
 napi/index.js
 napi/index.d.ts
 
-# Local samples and scripts
+# Local samples
 samples/
-scripts/
 
 # Test output
 test_output/
+.firecrawl/
 
 # Python
 __pycache__/
 *.pyc
 .pytest_cache/
-

AGENTS.md

@@ -61,7 +61,8 @@ src/
 
 - **Unit tests**: inline `#[cfg(test)] mod tests` in each module with synthetic data.
 - **Integration tests**: `tests/integration_tests.rs` with fixture PDFs in `tests/fixtures/`.
-- **Regression suite**: sibling repo `pdf-evals` with 179+ snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing.
+- **Regression suite**: sibling repo `pdf-evals` with ~200 snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing. While iterating, prefer a subset run (`bench.py test -q` for the quick set, or `-s <name>` for a named test set) and save the full `bench.py test` for the final pre-commit check.
+- **Semantic quality**: run `bench.py score` in `pdf-evals` for the semantic verdict (TEDS + MHS + reading order + char/word + list preservation, composited). Character-level diff alone misclassifies structural improvements (e.g., column-detection rewrites) as regressions — `score` is the tie-breaker. See `pdf-evals/CLAUDE.md` "Semantic scoring".
 
 ## Debugging
 

CLAUDE.md

@@ -61,7 +61,7 @@ src/
 
 - **Unit tests**: inline `#[cfg(test)] mod tests` in each module with synthetic data.
 - **Integration tests**: `tests/integration_tests.rs` with fixture PDFs in `tests/fixtures/`.
-- **Regression suite**: sibling repo `pdf-evals` with 187+ snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing.
+- **Regression suite**: sibling repo `pdf-evals` with ~200 snapshot PDFs. Run `cargo build --release` then `bench.py test` in that repo before committing. While iterating, prefer a subset run (`bench.py test -q` for the quick set, or `-s <name>` for a named test set) and save the full `bench.py test` for the final pre-commit check.
 - **Semantic quality**: run `bench.py score` in `pdf-evals` for the semantic verdict (TEDS + MHS + reading order + char/word + list preservation, composited). Character-level diff alone misclassifies structural improvements (e.g., column-detection rewrites) as regressions — `score` is the tie-breaker. See `pdf-evals/CLAUDE.md` "Semantic scoring".
 
 ## Debugging

Cargo.toml

@@ -1,12 +1,25 @@
 [package]
 name = "pdf-inspector"
-version = "0.1.0"
+version = "1.14.0"
 edition = "2021"
 autobins = false
 authors = ["Firecrawl Team"]
 description = "Fast PDF inspection, classification, and text extraction with smart scanned vs text-based detection"
 license = "MIT"
 repository = "https://github.com/firecrawl/pdf-inspector"
+readme = "docs/rust-api.md"
+# Explicit allowlist: crates.io caps uploads at 10 MiB and tests/fixtures
+# alone exceeds that. external/bcmaps ships in the crate — tounicode.rs
+# loads it at runtime relative to CARGO_MANIFEST_DIR.
+include = [
+    "/src/**",
+    "/external/bcmaps/**",
+    "/docs/rust-api.md",
+    "/LICENSE",
+    # maturin derives the sdist file list from this allowlist; the stub must
+    # ship so wheels built from the sdist keep their type hints.
+    "/pdf_inspector.pyi",
+]
 
 [lib]
 name = "pdf_inspector"
@@ -14,20 +27,13 @@ crate-type = ["lib", "cdylib"]
 
 [dependencies]
 # Python bindings
-pyo3 = { version = "0.25", features = ["extension-module"], optional = true }
-
-# PDF parsing
-lopdf = { git = "https://github.com/J-F-Liu/lopdf", rev = "7a05512d831415b1f2b1ce522391d6beab8a1284", features = ["rayon"] }
+pyo3 = { version = "0.25", features = ["extension-module", "abi3-py38"], optional = true }
 
 # Error handling
 thiserror = "2.0"
 
-# Parallel processing
-rayon = "1.10"
-
 # Logging
 log = "0.4"
-env_logger = "0.11"
 
 # Text processing
 regex = "1.10"
@@ -37,6 +43,19 @@ unicode-normalization = "0.1"
 # TrueType font parsing (for Identity-H CID font cmap extraction)
 ttf-parser = "0.25"
 
+# Native builds keep lopdf's parallel parser and CLI logging. Browser WASM is
+# deliberately single-threaded so it works without cross-origin isolation.
+[target.'cfg(not(target_arch = "wasm32"))'.dependencies]
+lopdf = { version = "0.42.0", features = ["rayon"] }
+rayon = "1.10"
+env_logger = "0.11"
+
+# Browser builds use JavaScript randomness for encrypted PDFs and embed the
+# bundled CMaps because there is no filesystem at runtime.
+[target.'cfg(target_arch = "wasm32")'.dependencies]
+lopdf = { version = "0.42.0", default-features = false, features = ["wasm_js"] }
+include_dir = "0.7"
+
 [dev-dependencies]
 tempfile = "3.3"
 

LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Firecrawl
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

README.md

@@ -1,6 +1,11 @@
 # pdf-inspector
 
-Fast Rust library for PDF classification and text extraction. Detects whether a PDF is text-based or scanned, extracts text with position awareness, and converts to clean Markdown — all without OCR. Includes bindings for [Python](docs/python.md) and [Node.js](napi/README.md).
+[![Crates.io](https://img.shields.io/crates/v/pdf-inspector.svg)](https://crates.io/crates/pdf-inspector)
+[![npm](https://img.shields.io/npm/v/@firecrawl/pdf-inspector.svg)](https://www.npmjs.com/package/@firecrawl/pdf-inspector)
+[![PyPI](https://img.shields.io/pypi/v/pdf-inspector.svg)](https://pypi.org/project/pdf-inspector/)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
+
+Fast Rust library for PDF classification and text extraction. Detects whether a PDF is text-based or scanned, extracts text with position awareness, and converts to clean Markdown — all without OCR. Includes bindings for [Python](docs/python.md), [Node.js](napi/README.md), and [browser WebAssembly](wasm/README.md).
 
 Built by [Firecrawl](https://firecrawl.dev) to handle text-based PDFs locally in under 200ms, skipping expensive OCR services for the ~54% of PDFs that don't need them.
 
@@ -14,24 +19,28 @@ Built by [Firecrawl](https://firecrawl.dev) to handle text-based PDFs locally in
 - **Multi-column layout** — Automatic detection of newspaper-style columns, sequential reading order, and RTL text support.
 - **Encoding issue detection** — Automatically flags broken font encodings so callers can fall back to OCR.
 - **Single document load** — The document is parsed once and shared between detection and extraction, avoiding redundant I/O.
+- **Browser WebAssembly** — Run the same Rust parser locally in browsers and Web Workers, with embedded CMaps and no server round trip.
 - **Lightweight** — Pure Rust, no ML models, no external services. Single dependency on `lopdf` for PDF parsing.
 
 ## Benchmark
 
-Evaluated on the [opendataloader-bench](https://github.com/opendataloader-project/opendataloader-bench) corpus (200 PDFs). Only direct text extraction engines are shown — no OCR, no ML models. Scores are 0-1, higher is better.
+Evaluated on the [opendataloader-bench](https://github.com/opendataloader-project/opendataloader-bench) corpus (200 PDFs). Only local engines without model-based PDF parsing are shown; OCR was disabled. Scores are 0-1, higher is better.
 
 | Engine | Overall | Reading Order (NID) | Tables (TEDS) | Headings (MHS) | Speed (200 docs) |
 |---|---|---|---|---|---|
-| pdf-inspector | 0.78 | 0.87 | 0.59 | 0.57 | 4s |
-| opendataloader | 0.84 | 0.91 | 0.49 | 0.74 | 11s |
-| pymupdf4llm | 0.73 | 0.89 | 0.40 | 0.41 | 18s |
-| markitdown | 0.58 | 0.88 | 0.00 | 0.00 | 8s |
+| pdf-inspector | **0.875** | **0.915** | **0.814** | 0.788 | **0.470s** |
+| liteparse | 0.873 | 0.913 | 0.693 | **0.811** | 0.750s |
+| opendataloader | 0.831 | 0.902 | 0.489 | 0.739 | 2.569s |
+| pymupdf4llm | 0.735 | 0.886 | 0.401 | 0.424 | 17.117s |
+| markitdown | 0.589 | 0.844 | 0.273 | 0.000 | 16.165s |
 
-For context, engines that use OCR/ML (docling, marker, mineru) score 0.83-0.88 overall but take 2-180 minutes on the same corpus.
+Results were refreshed on July 31, 2026, on an Apple M4 Pro. Engine versions were pdf-inspector 0.2.6, LiteParse 2.10.1, OpenDataLoader 2.2.1, PyMuPDF4LLM 0.2.0, and MarkItDown 0.1.5. Speed is the median of five alternating or rotating complete corpus runs after an excluded warm-up run, with each parser processing documents sequentially in a single process.
 
-**Where we do well:** Speed (fastest of all engines), reading order, table detection vs other direct-text tools.
+The complete parser configuration, per-document predictions, evaluator output, and generated charts are available in the [reproducible results branch](https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results).
 
-**Where we lag:** Heading detection trails opendataloader — many PDFs use bold text at body font size for headings, or headings that are only slightly larger than body text. Table detection trails OCR-based engines that can see visual table structure.
+**Best fit:** Native-text PDFs where speed, reading order, and table structure matter. In this comparison, pdf-inspector delivered the higher overall, reading-order, and table scores, along with the fastest complete run. That makes it a strong local default for reports, research papers, financial documents, invoices, and legal PDFs that need clean, structured Markdown without adding OCR latency or infrastructure.
+
+Use the [paired benchmark harness](docs/benchmarking.md) to compare two local builds against the exact same corpus and evaluator revision.
 
 ## Quick start
 
@@ -69,11 +78,39 @@ console.log(result.markdown);  // Markdown string or null
 
 > Full API reference: [napi/README.md](napi/README.md)
 
+### Browser WebAssembly
+
+```bash
+npm install @firecrawl/pdf-inspector-wasm
+```
+
+```javascript
+import init, { processPdf } from '@firecrawl/pdf-inspector-wasm';
+
+await init();
+const response = await fetch('/document.pdf');
+const pdf = new Uint8Array(await response.arrayBuffer());
+const result = processPdf(pdf);
+
+console.log(result.pdfType);
+console.log(result.markdown);
+```
+
+> Full API reference: [wasm/README.md](wasm/README.md)
+
 ### Rust
 
+Install from [crates.io](https://crates.io/crates/pdf-inspector):
+
+```bash
+cargo add pdf-inspector
+```
+
+Or add it manually:
+
 ```toml
 [dependencies]
-pdf-inspector = { git = "https://github.com/firecrawl/pdf-inspector" }
+pdf-inspector = "1"
 ```
 
 ```rust
@@ -91,29 +128,40 @@ if let Some(markdown) = &result.markdown {
 ### CLI
 
 ```bash
+# Install the CLI tools
+cargo install pdf-inspector
+
 # Convert PDF to Markdown
-cargo run --bin pdf2md -- document.pdf
+pdf2md document.pdf
 
 # JSON output (for piping)
-cargo run --bin pdf2md -- document.pdf --json
+pdf2md document.pdf --json
+
+# Positioned TextItem JSON, including is_underline metadata
+pdf2md document.pdf --items-json
 
 # Raw markdown only (no headers)
-cargo run --bin pdf2md -- document.pdf --raw
+pdf2md document.pdf --raw
+
+# Token-efficient output (collapses long dot leaders and similar source padding)
+pdf2md document.pdf --compact
 
 # Insert page break markers (<!-- Page N -->)
-cargo run --bin pdf2md -- document.pdf --pages
+pdf2md document.pdf --pages
 
 # Process only specific pages
-cargo run --bin pdf2md -- document.pdf --select-pages 1,3,5-10
+pdf2md document.pdf --select-pages 1,3,5-10
 
 # Detection only (no extraction)
-cargo run --bin detect-pdf -- document.pdf
-cargo run --bin detect-pdf -- document.pdf --json
+detect-pdf document.pdf
+detect-pdf document.pdf --json
 
 # Detection + layout analysis (tables, columns)
-cargo run --bin detect-pdf -- document.pdf --analyze --json
+detect-pdf document.pdf --analyze --json
 ```
 
+From a source checkout, use `cargo run --bin pdf2md -- document.pdf` or `cargo run --bin detect-pdf -- document.pdf` instead.
+
 ## Architecture
 
 ```
@@ -161,6 +209,7 @@ src/
   markdown/             — Markdown conversion and structure detection
   bin/                  — CLI tools (pdf2md, detect_pdf)
 napi/                   — Node.js/Bun bindings (napi-rs)
+wasm/                   — Browser bindings (wasm-bindgen)
 ```
 
 ## How classification works
@@ -189,7 +238,7 @@ The converter handles:
 |---|---|
 | Headings (H1-H4) | Font size tiers relative to body text, with 0.5pt clustering |
 | Bold/italic | Font name patterns (Bold, Italic, Oblique) |
-| Bullet lists | `*`, `-`, `*`, `○`, `●`, `◦` prefixes |
+| Bullet lists | `•`, `-`, `*`, `○`, `●`, `◦` prefixes |
 | Numbered lists | `1.`, `1)`, `(1)` patterns |
 | Letter lists | `a.`, `a)`, `(a)` patterns |
 | Code blocks | Monospace fonts (Courier, Consolas, Monaco, Menlo, Fira Code, JetBrains Mono) and keyword detection |
@@ -223,4 +272,4 @@ See [docs/debugging.md](docs/debugging.md) for `RUST_LOG` environment variable u
 
 ## License
 
-MIT
+[MIT](LICENSE)

SECURITY.md

@@ -5,14 +5,15 @@
 If you believe you've found a security vulnerability in pdf-inspector, please
 report it privately so we can fix it before public disclosure.
 
-**Preferred:** Email **help@firecrawl.dev** with:
+**Preferred:** Submit through Firecrawl's Bugcrowd vulnerability disclosure
+program at <https://bugcrowd.com/engagements/firecrawl-vdp-ess>. Please include:
 
 - A description of the issue and its impact
 - Steps to reproduce (a minimal PDF or input that triggers the bug is ideal)
 - The version or commit hash of pdf-inspector you tested against
 
-**Alternative:** Use GitHub's private vulnerability reporting under the
-[Security tab](https://github.com/firecrawl/pdf-inspector/security/advisories/new).
+**Alternative:** If you'd rather not use Bugcrowd, email
+**help@firecrawl.dev** with the same details.
 
 We'll acknowledge your report in a timely manner and keep you updated on
 remediation progress. Please do not open a public GitHub issue for security

docs/benchmarking.md

@@ -0,0 +1,62 @@
+# Benchmarking against OpenDataLoader
+
+The paired harness runs two `pdf2md` binaries through the same local
+OpenDataLoader corpus, evaluates both outputs, and reports aggregate and
+per-document deltas. This avoids comparing results produced from different
+corpus revisions or evaluator versions.
+
+Build a candidate and provide a released or worktree build as the baseline:
+
+```bash
+cargo build --release
+python3 scripts/bench_opendataloader.py \
+  --bench-dir ../opendataloader-bench \
+  --baseline ../pdf-inspector-main/target/release/pdf2md \
+  --candidate target/release/pdf2md \
+  --max-document-regression 0.02 \
+  --json-output /tmp/pdf-inspector-benchmark.json
+```
+
+Pass `--reference-evaluation path/to/evaluation.json` to report the candidate
+delta against another evaluation, and add `--require-reference-lead` to make a
+negative reference delta fail the run. By default, the candidate must not
+regress the baseline overall score or introduce missing predictions. Use
+`--min-overall-delta` to require a specific aggregate gain.
+
+The OpenDataLoader repository is external and keeps its normal
+`prediction/pdf-inspector` output. Paired evaluation copies each run into a
+temporary directory before evaluating it, so the baseline and candidate cannot
+overwrite one another.
+
+## Published comparison protocol
+
+The public benchmark table was refreshed on July 31, 2026, on an Apple M4 Pro
+using pdf-inspector 0.2.6, LiteParse 2.10.1, OpenDataLoader 2.2.1,
+PyMuPDF4LLM 0.2.0, and MarkItDown 0.1.5. Every engine processed the same 200
+PDFs sequentially in a single process with OCR disabled. Reported speed is the
+median of five alternating or rotating complete corpus runs after an excluded
+warm-up run; quality scores come from the benchmark evaluator over all 200
+outputs. Raw timings, predictions, evaluations, and charts are available in the
+[results branch](https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results).
+
+## Optional backend evidence probe
+
+The evidence probe compares positioned `pdf2md` items with MuPDF structured
+text on the same pages. It is intended to find deterministic extraction or
+layout evidence that could justify a future native implementation; it does not
+merge MuPDF output into Markdown, invoke OCR, or add a runtime dependency.
+
+Install MuPDF's `mutool`, build `pdf2md`, then run:
+
+```bash
+python3 scripts/probe_backend_evidence.py document.pdf \
+  --pdf2md target/release/pdf2md \
+  --json-output /tmp/backend-evidence.json
+```
+
+The report flags pages when MuPDF exposes a material net token gain, repeated
+alignment anchors absent from local evidence, or additional image blocks. The
+JSON includes bounded token samples and page-level counts so promising cases
+can be inspected without treating backend disagreement as automatically
+correct. Thresholds are configurable with `--min-token-gain`,
+`--min-alternate-only-ratio`, and `--min-anchor-gain`.

docs/publishing.md

@@ -0,0 +1,57 @@
+# Publishing
+
+Every pdf-inspector distribution uses one shared semantic version:
+
+- Rust crate: `pdf-inspector`
+- Python package: `pdf-inspector`
+- Node package: `@firecrawl/pdf-inspector` and its platform packages
+- Browser package: `@firecrawl/pdf-inspector-wasm`
+- Internal NAPI and WASM Rust crates
+
+`Cargo.toml` is the canonical version source. Update every manifest and lockfile
+with:
+
+```bash
+python3 scripts/version.py <version>
+```
+
+Verify that nothing has diverged with:
+
+```bash
+python3 scripts/version.py --check
+```
+
+CI and every publishing workflow run this check before building or publishing.
+
+## Release steps
+
+1. Choose the next shared semantic version and run `scripts/version.py`.
+2. Review the manifest and lockfile changes in the version-bump pull request.
+3. Merge the pull request to `main`.
+4. The crates.io, PyPI, Node, and WASM workflows independently build and
+   publish that version from the same commit.
+5. After all registries succeed, create one `v<version>` GitHub release that
+   links to each package and describes changes since the previous shared tag.
+
+The independent workflows are intentionally idempotent. A manual dispatch from
+`main` can repair a partial release, and already-published artifacts are skipped.
+
+## Trusted publishers
+
+The repositories use GitHub Actions OIDC instead of long-lived registry tokens.
+Configure each registry's trusted publisher for `firecrawl/pdf-inspector` and
+its corresponding workflow:
+
+- crates.io: `publish-crate.yml`, environment `crates-io`
+- PyPI: `publish-pypi.yml`, environment `pypi`
+- npm Node package: `publish.yml`
+- npm WASM package: `publish-wasm.yml`
+
+The WASM package must exist before npm trusted publishing can be configured. If
+it ever needs to be bootstrapped again, build and inspect it before publishing:
+
+```bash
+wasm-pack build wasm --target web --scope firecrawl --out-dir pkg --release
+npm pack --dry-run ./wasm/pkg
+npm publish ./wasm/pkg --access public
+```

docs/python.md

@@ -1,9 +1,39 @@
-# Python API
+# pdf-inspector
 
-Python bindings via [PyO3](https://pyo3.rs). Requires Rust toolchain for building from source.
+Fast PDF classification and text extraction. Detects whether a PDF is text-based or scanned, extracts text with position awareness, and converts to clean Markdown — all without OCR. Python bindings via [PyO3](https://pyo3.rs) for the [pdf-inspector](https://github.com/firecrawl/pdf-inspector) Rust library.
+
+Built by [Firecrawl](https://firecrawl.dev) to handle text-based PDFs locally in under 200ms, skipping expensive OCR services for the ~54% of PDFs that don't need them.
+
+## Features
+
+- **Smart classification** — `text_based` / `scanned` / `image_based` / `mixed` in ~10–50ms, with a confidence score and per-page OCR routing.
+- **Markdown conversion** — headings, lists, code blocks, bold/italic, URL linking, and dual-mode table detection (PDF drawing ops + text-alignment heuristics).
+- **Layout-aware extraction** — multi-column reading order, position and font info per text item, RTL support.
+- **Robust text decoding** — CID/Type0 fonts via ToUnicode CMaps, plus automatic flagging of broken encodings so callers can fall back to OCR.
+- **Lightweight** — native Rust core, no ML models, no external services; ships type stubs.
+
+## Benchmark
+
+[opendataloader-bench](https://github.com/opendataloader-project/opendataloader-bench) corpus (200 PDFs), local engines without model-based PDF parsing; OCR disabled. Scores 0–1, higher is better:
+
+| Engine | Overall | Reading order | Tables (TEDS) | Headings | Speed |
+|---|---|---|---|---|---|
+| **pdf-inspector** | **0.875** | **0.915** | **0.814** | 0.788 | **0.470s** |
+| liteparse | 0.873 | 0.913 | 0.693 | **0.811** | 0.750s |
+| opendataloader | 0.831 | 0.902 | 0.489 | 0.739 | 2.569s |
+| pymupdf4llm | 0.735 | 0.886 | 0.401 | 0.424 | 17.117s |
+| markitdown | 0.589 | 0.844 | 0.273 | 0.000 | 16.165s |
+
+Refreshed July 31, 2026, on Apple M4 Pro; speed is the median of five complete corpus runs after an excluded warm-up. Full methodology and versions are in the [repo README](https://github.com/firecrawl/pdf-inspector#benchmark), with raw timings and artifacts in the [results branch](https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results).
 
 ## Install
 
+```bash
+pip install pdf-inspector
+```
+
+Prebuilt wheels cover CPython ≥3.8 on Linux (x86_64, aarch64), macOS (Intel, Apple Silicon), and Windows (x64). Other platforms build from source, which requires a Rust toolchain. For local development in a repo checkout:
+
 ```bash
 pip install maturin
 maturin develop --release
@@ -50,6 +80,17 @@ for page in result.pages:
 
 # Restrict to specific 0-indexed pages (preserves caller order)
 result = pdf_inspector.extract_pages_markdown("document.pdf", pages=[0, 2])
+
+# Structure-tree elements from tagged PDFs (empty list when untagged).
+# Pages are 1-indexed to match TextItem.page, so (page, mcid) joins directly
+# against extract_text_with_positions — e.g. to recover real heading levels:
+elements = pdf_inspector.extract_structure_elements("tagged.pdf")
+roles = {(e.page, e.mcid): e.role for e in elements}
+headings = [
+    item.text
+    for item in pdf_inspector.extract_text_with_positions("tagged.pdf")
+    if item.mcid is not None and roles.get((item.page, item.mcid), "").startswith("H")
+]
 ```
 
 ## API reference
@@ -70,19 +111,73 @@ result = pdf_inspector.extract_pages_markdown("document.pdf", pages=[0, 2])
 | `extract_text_in_regions_bytes(data, page_regions)` | Region extraction from bytes |
 | `extract_pages_markdown(path, pages=None)` | Per-page Markdown + layout metadata (all pages by default) |
 | `extract_pages_markdown_bytes(data, pages=None)` | Per-page Markdown from bytes |
+| `extract_structure_elements(path, pages=None)` | Structure-tree elements from tagged PDFs (page, mcid, role) |
+| `extract_structure_elements_bytes(data, pages=None)` | Structure-tree elements from bytes |
 
 ## Types
 
-**`PdfResult` fields:** `pdf_type`, `markdown`, `page_count`, `processing_time_ms`, `pages_needing_ocr`, `title`, `confidence`, `is_complex_layout`, `pages_with_tables`, `pages_with_columns`, `has_encoding_issues`
+Type stubs (`pdf_inspector.pyi`) ship with the package. Result types at a glance:
 
-**`PdfClassification` fields:** `pdf_type`, `page_count`, `pages_needing_ocr` (0-indexed), `confidence`
+```python
+class PdfResult:                     # process_pdf / detect_pdf
+    pdf_type: str                    # "text_based" | "scanned" | "image_based" | "mixed"
+    markdown: str | None             # extracted Markdown (None for detect_pdf)
+    page_count: int
+    processing_time_ms: int
+    pages_needing_ocr: list[int]     # 1-indexed
+    ocr_reasons_by_page: list[PageOcrReasons]
+    title: str | None
+    confidence: float                # 0.0 - 1.0
+    is_complex_layout: bool
+    pages_with_tables: list[int]
+    pages_with_columns: list[int]
+    has_encoding_issues: bool        # broken font encodings — consider OCR fallback
 
-**`TextItem` fields:** `text`, `x`, `y`, `width`, `height`, `font`, `font_size`, `page`, `is_bold`, `is_italic`, `item_type`
+class PageOcrReasons:                # per-page OCR diagnostics
+    page: int                        # 1-indexed
+    reasons: list[str]               # machine-readable reason identifiers
 
-**`RegionText` fields:** `text`, `needs_ocr`
+class PdfClassification:             # classify_pdf
+    pdf_type: str
+    page_count: int
+    pages_needing_ocr: list[int]     # 0-indexed
+    confidence: float
 
-**`PageRegionTexts` fields:** `page` (0-indexed), `regions` (list of RegionText)
+class TextItem:                      # extract_text_with_positions
+    text: str
+    x: float
+    y: float
+    width: float
+    height: float
+    font: str
+    font_size: float
+    page: int
+    is_bold: bool
+    is_italic: bool
+    is_underline: bool
+    is_strikeout: bool
+    item_type: str
+    mcid: int | None                 # marked-content ID for tagged PDFs (None otherwise)
 
-**`PageMarkdown` fields:** `page` (0-indexed), `markdown`, `needs_ocr`
+class StructureElement:              # extract_structure_elements
+    page: int                        # 1-indexed (matches TextItem.page)
+    mcid: int
+    role: str                        # "H1".."H6", "P", "Table", ... (resolved via /RoleMap)
 
-**`PagesExtractionResult` fields:** `pages` (list of PageMarkdown), `pages_with_tables` (1-indexed), `pages_with_columns` (1-indexed), `pages_needing_ocr` (1-indexed), `is_complex`
+class RegionText:                    # extract_text_in_regions
+    text: str
+    needs_ocr: bool
+    ocr_reason: str | None           # machine-readable OCR reason
+
+class PageRegionTexts:               # extract_text_in_regions
+    page: int                        # 0-indexed
+    regions: list[RegionText]
+
+class PagesExtractionResult:         # extract_pages_markdown
+    pages: list[PageMarkdown]        # PageMarkdown: page (0-indexed), markdown, needs_ocr, ocr_reason
+    pages_with_tables: list[int]     # 1-indexed
+    pages_with_columns: list[int]    # 1-indexed
+    pages_needing_ocr: list[int]     # 1-indexed
+    ocr_reasons_by_page: list[PageOcrReasons]
+    is_complex: bool                 # any page has tables or multi-column layout
+```

docs/rust-api.md

@@ -1,12 +1,50 @@
-# Rust API
+# pdf-inspector
 
-Add to your `Cargo.toml`:
+Fast PDF classification and text extraction. Detects whether a PDF is text-based or scanned, extracts text with position awareness, and converts to clean Markdown — all without OCR. Pure Rust, no ML models, no external services; the only PDF dependency is [lopdf](https://crates.io/crates/lopdf). Also available for [Python](https://pypi.org/project/pdf-inspector/) and [Node.js](https://www.npmjs.com/package/@firecrawl/pdf-inspector).
+
+Built by [Firecrawl](https://firecrawl.dev) to handle text-based PDFs locally in under 200ms, skipping expensive OCR services for the ~54% of PDFs that don't need them.
+
+## Features
+
+- **Smart classification** — TextBased / Scanned / ImageBased / Mixed in ~10–50ms, with a confidence score and per-page OCR routing.
+- **Markdown conversion** — headings, lists, code blocks, bold/italic, URL linking, and dual-mode table detection (PDF drawing ops + text-alignment heuristics).
+- **Layout-aware extraction** — multi-column reading order, position and font info per text item, RTL support.
+- **Robust text decoding** — CID/Type0 fonts via ToUnicode CMaps, plus automatic flagging of broken encodings so callers can fall back to OCR.
+- **Lightweight** — pure Rust, no ML models, no external services; single PDF dependency ([lopdf](https://crates.io/crates/lopdf)).
+
+## Benchmark
+
+[opendataloader-bench](https://github.com/opendataloader-project/opendataloader-bench) corpus (200 PDFs), local engines without model-based PDF parsing; OCR disabled. Scores 0–1, higher is better:
+
+| Engine | Overall | Reading order | Tables (TEDS) | Headings | Speed |
+|---|---|---|---|---|---|
+| **pdf-inspector** | **0.875** | **0.915** | **0.814** | 0.788 | **0.470s** |
+| liteparse | 0.873 | 0.913 | 0.693 | **0.811** | 0.750s |
+| opendataloader | 0.831 | 0.902 | 0.489 | 0.739 | 2.569s |
+| pymupdf4llm | 0.735 | 0.886 | 0.401 | 0.424 | 17.117s |
+| markitdown | 0.589 | 0.844 | 0.273 | 0.000 | 16.165s |
+
+Refreshed July 31, 2026, on Apple M4 Pro; speed is the median of five complete corpus runs after an excluded warm-up. Full methodology and versions are in the [repo README](https://github.com/firecrawl/pdf-inspector#benchmark), with raw timings and artifacts in the [results branch](https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results).
+
+## Install
+
+```bash
+cargo add pdf-inspector
+```
+
+For the latest unreleased changes, use the git dependency instead:
 
 ```toml
 [dependencies]
 pdf-inspector = { git = "https://github.com/firecrawl/pdf-inspector" }
 ```
 
+The crate also ships CLI binaries — `pdf2md` (PDF → Markdown, with `--json`, `--pages`, `--select-pages`, and the opt-in token-saving `--compact` profile) and `detect-pdf` (classification, with `--analyze --json`):
+
+```bash
+cargo install pdf-inspector
+```
+
 ## Usage
 
 Detect and extract in one call:
@@ -100,6 +138,34 @@ for page in &result.pages {
 println!("Complex layout? {}", result.is_complex);
 ```
 
+Extract structure-tree elements from tagged PDFs, and join them against
+`extract_text_with_positions` to attach semantic roles (heading levels,
+paragraphs, table cells) to extracted text:
+
+```rust
+use pdf_inspector::{extract_structure_elements, extract_text_with_positions};
+use std::collections::HashMap;
+
+// One entry per marked-content reference, sorted by (page, mcid); empty for
+// untagged PDFs. Pages are 1-indexed to match `TextItem::page`, so the
+// (page, mcid) pair is a direct join key.
+let elements = extract_structure_elements("tagged.pdf", None)?;
+let roles: HashMap<(u32, i64), &str> = elements
+    .iter()
+    .map(|e| ((e.page, e.mcid), e.role.as_str()))
+    .collect();
+
+for item in extract_text_with_positions("tagged.pdf")? {
+    if let Some(mcid) = item.mcid {
+        if let Some(role) = roles.get(&(item.page, mcid)) {
+            if role.starts_with('H') {
+                println!("{}: {}", role, item.text);
+            }
+        }
+    }
+}
+```
+
 ## Processing modes
 
 | Mode | What it does | Returns |
@@ -125,6 +191,8 @@ println!("Complex layout? {}", result.is_complex);
 | `to_markdown_from_items_with_rects(items, options, rects)` | Markdown with rectangle-based table detection |
 | `extract_pages_markdown(path, pages)` | Per-page Markdown + layout metadata (file) |
 | `extract_pages_markdown_mem(bytes, pages)` | Per-page Markdown from bytes |
+| `extract_structure_elements(path, pages)` | Structure-tree elements from tagged PDFs (page, mcid, role) |
+| `extract_structure_elements_mem(bytes, pages)` | Structure-tree elements from bytes |
 
 Low-level detection functions are also available via the `detector` module (`detect_pdf_type`, `detect_pdf_type_with_config`, etc.) for callers who need `PdfTypeResult` instead of `PdfProcessResult`.
 
@@ -140,7 +208,8 @@ Low-level detection functions are also available via the `detector` module (`det
 | `DetectionConfig` | Configuration for detection: scan strategy, thresholds |
 | `ScanStrategy` | `EarlyExit`, `Full`, `Sample(n)`, `Pages(vec)` |
 | `LayoutComplexity` | Layout analysis: is_complex, pages_with_tables, pages_with_columns |
-| `TextItem` | Text with position, font info, and page number |
+| `TextItem` | Text with position, font info, page number, and optional structure-tree `mcid` |
+| `StructureElement` | Tagged-PDF structure reference: page (1-indexed), mcid, role (`"H1"`..`"H6"`, `"P"`, …) |
 | `MarkdownOptions` | Configuration for Markdown formatting (page numbers, etc.) |
 | `PageMarkdown` | Per-page result: page (0-indexed), markdown, needs_ocr |
 | `PagesExtractionResult` | Per-page output + 1-indexed pages_with_tables / pages_with_columns / pages_needing_ocr, is_complex |

napi/Cargo.lock

@@ -499,11 +499,13 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "0de51e6874e94e7bf76d726fc5d13ba782deca734ff60d5bb2fb2607c7406555"
 dependencies = [
  "cfg-if",
+ "js-sys",
  "libc",
  "r-efi",
  "rand_core",
  "wasip2",
  "wasip3",
+ "wasm-bindgen",
 ]
 
 [[package]]
@@ -557,6 +559,25 @@ version = "2.3.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "3d3067d79b975e8844ca9eb072e16b31c3c1c36928edf9c6789548c524d0d954"
 
+[[package]]
+name = "include_dir"
+version = "0.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "923d117408f1e49d914f1a379a309cffe4f18c05cf4e3d12e613a15fc81bd0dd"
+dependencies = [
+ "include_dir_macros",
+]
+
+[[package]]
+name = "include_dir_macros"
+version = "0.7.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7cab85a7ed0bd5f0e76d93846e0147172bed2e2d3f859bcc33a8d9699cad1a75"
+dependencies = [
+ "proc-macro2",
+ "quote",
+]
+
 [[package]]
 name = "indexmap"
 version = "2.13.0"
@@ -672,8 +693,9 @@ checksum = "5e5032e24019045c762d3c0f28f5b6b8bbf38563a65908389bf7978758920897"
 
 [[package]]
 name = "lopdf"
-version = "0.40.0"
-source = "git+https://github.com/J-F-Liu/lopdf?rev=7a05512d831415b1f2b1ce522391d6beab8a1284#7a05512d831415b1f2b1ce522391d6beab8a1284"
+version = "0.42.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "25aab26d99567469098e64a02f42679f8965c6401263eefa31d8f2dcc37a221c"
 dependencies = [
  "aes",
  "bitflags",
@@ -829,9 +851,10 @@ checksum = "384b8ab6d37215f3c5301a95a4accb5d64aa607f1fcb26a11b5303878451b4fe"
 
 [[package]]
 name = "pdf-inspector"
-version = "0.1.0"
+version = "1.14.0"
 dependencies = [
  "env_logger",
+ "include_dir",
  "log",
  "lopdf",
  "once_cell",
@@ -844,7 +867,7 @@ dependencies = [
 
 [[package]]
 name = "pdf-inspector-napi"
-version = "0.2.0"
+version = "1.14.0"
 dependencies = [
  "napi",
  "napi-build",

napi/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "pdf-inspector-napi"
-version = "0.2.0"
+version = "1.14.0"
 edition = "2021"
 
 [lib]

napi/README.md

@@ -4,6 +4,28 @@ Fast PDF classification and region-based text extraction for Node.js/Bun. Native
 
 Built by [Firecrawl](https://firecrawl.dev) for hybrid OCR pipelines — extract text from PDF structure where possible, fall back to OCR only when needed.
 
+## Features
+
+- **Smart classification** — text-based / scanned / image-based / mixed in ~10–50ms, with a confidence score and per-page OCR routing.
+- **Region-based extraction** — pull text from bounding boxes with per-region quality checks (`needsOcr`).
+- **Layout-aware** — multi-column reading order, position and font info per text item, RTL support.
+- **Robust text decoding** — CID/Type0 fonts via ToUnicode CMaps, plus automatic flagging of broken encodings so callers can fall back to OCR.
+- **Lightweight** — native Rust core via napi-rs, no ML models, no external services; ~5–6 MB platform binary, TypeScript definitions included.
+
+## Benchmark
+
+[opendataloader-bench](https://github.com/opendataloader-project/opendataloader-bench) corpus (200 PDFs), local engines without model-based PDF parsing; OCR disabled. Scores 0–1, higher is better:
+
+| Engine | Overall | Reading order | Tables (TEDS) | Headings | Speed |
+|---|---|---|---|---|---|
+| **pdf-inspector** | **0.875** | **0.915** | **0.814** | 0.788 | **0.470s** |
+| liteparse | 0.873 | 0.913 | 0.693 | **0.811** | 0.750s |
+| opendataloader | 0.831 | 0.902 | 0.489 | 0.739 | 2.569s |
+| pymupdf4llm | 0.735 | 0.886 | 0.401 | 0.424 | 17.117s |
+| markitdown | 0.589 | 0.844 | 0.273 | 0.000 | 16.165s |
+
+Refreshed July 31, 2026, on Apple M4 Pro; speed is the median of five complete corpus runs after an excluded warm-up. Full methodology and versions are in the [repo README](https://github.com/firecrawl/pdf-inspector#benchmark), with raw timings and artifacts in the [results branch](https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results).
+
 ## Install
 
 ```bash
@@ -12,7 +34,7 @@ npm install @firecrawl/pdf-inspector
 bun add @firecrawl/pdf-inspector
 ```
 
-Prebuilt binaries included for **linux-x64** and **macOS ARM64**. No Rust toolchain needed.
+Prebuilt binaries for **Linux x64/ARM64** (glibc and musl/Alpine), **macOS ARM64**, and **Windows x64** — npm installs only the one matching your platform. No Rust toolchain needed.
 
 ## API
 
@@ -37,7 +59,7 @@ console.log(result.confidence)     // 0.875
 
 Extract text within bounding-box regions from a PDF. Designed for hybrid OCR pipelines where a layout model detects regions in rendered page images, and this function extracts text from the PDF structure for text-based pages — skipping GPU OCR.
 
-Each region result includes a `needsOcr` flag that signals unreliable extraction (empty text, GID-encoded fonts, garbage text, encoding issues).
+Each region result includes a `needsOcr` flag that signals unreliable extraction (empty text, GID-encoded fonts, garbage text, encoding issues). When the cause is a suspected garbled text layer, `ocrReason` is set to `"suspected_garbled_text"`.
 
 ```typescript
 import { extractTextInRegions } from '@firecrawl/pdf-inspector'
@@ -61,6 +83,22 @@ for (const region of result[0].regions) {
 }
 ```
 
+### Async variants
+
+`processPdf`, `classifyPdf`, and `extractPagesMarkdown` are synchronous and parse on the calling thread — in Node, that's the event loop. For a one-off call in a script that's fine, but in a server a large document can hold the loop for tens to hundreds of milliseconds.
+
+`processPdfAsync`, `classifyPdfAsync`, and `extractPagesMarkdownAsync` take the same arguments and produce the same results, but run the parse on the libuv thread pool and return a promise, keeping the event loop free. The input buffer is copied before the call returns, so it's safe to reuse or mutate immediately:
+
+```typescript
+import { classifyPdfAsync, extractPagesMarkdownAsync } from '@firecrawl/pdf-inspector'
+
+const classification = await classifyPdfAsync(pdf)
+if (classification.pdfType === 'TextBased') {
+  const { pages } = await extractPagesMarkdownAsync(pdf)
+  // ...
+}
+```
+
 ## Types
 
 ```typescript
@@ -84,15 +122,22 @@ interface PageRegionTexts {
 interface RegionText {
   text: string
   needsOcr: boolean         // true when text is unreliable
+  ocrReason?: string        // "suspected_garbled_text" when known
 }
 ```
 
 ## Platforms
 
-| Platform | Architecture | Supported |
-|----------|-------------|-----------|
-| Linux    | x64         | Yes       |
-| macOS    | ARM64       | Yes       |
+Prebuilt binaries ship as platform-specific packages installed automatically via `optionalDependencies`:
+
+| Platform | Architecture | Package |
+|----------|-------------|---------|
+| Linux    | x64 (glibc)         | `@firecrawl/pdf-inspector-linux-x64-gnu` |
+| Linux    | x64 (musl/Alpine)   | `@firecrawl/pdf-inspector-linux-x64-musl` |
+| Linux    | ARM64 (glibc)       | `@firecrawl/pdf-inspector-linux-arm64-gnu` |
+| Linux    | ARM64 (musl/Alpine) | `@firecrawl/pdf-inspector-linux-arm64-musl` |
+| macOS    | ARM64               | `@firecrawl/pdf-inspector-darwin-arm64` |
+| Windows  | x64                 | `@firecrawl/pdf-inspector-win32-x64-msvc` |
 
 ## License
 

napi/bun.lock

@@ -7,6 +7,14 @@
       "devDependencies": {
         "@napi-rs/cli": "^3.4.1",
       },
+      "optionalDependencies": {
+        "@firecrawl/pdf-inspector-darwin-arm64": "1.14.0",
+        "@firecrawl/pdf-inspector-linux-arm64-gnu": "1.14.0",
+        "@firecrawl/pdf-inspector-linux-arm64-musl": "1.14.0",
+        "@firecrawl/pdf-inspector-linux-x64-gnu": "1.14.0",
+        "@firecrawl/pdf-inspector-linux-x64-musl": "1.14.0",
+        "@firecrawl/pdf-inspector-win32-x64-msvc": "1.14.0",
+      },
     },
   },
   "packages": {

napi/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@firecrawl/pdf-inspector",
-  "version": "1.9.2",
+  "version": "1.14.0",
   "description": "Fast PDF classification and text extraction. Detect text-based vs scanned PDFs, extract text by region with quality checks. Native Rust performance via napi-rs.",
   "main": "index.js",
   "types": "index.d.ts",
@@ -22,7 +22,6 @@
   "files": [
     "index.js",
     "index.d.ts",
-    "*.node",
     "bin/",
     "README.md"
   ],
@@ -38,12 +37,12 @@
     "binaryName": "pdf-inspector",
     "targets": [
       "x86_64-unknown-linux-gnu",
+      "x86_64-unknown-linux-musl",
+      "aarch64-unknown-linux-gnu",
+      "aarch64-unknown-linux-musl",
       "aarch64-apple-darwin",
       "x86_64-pc-windows-msvc"
-    ],
-    "package": {
-      "name": "@firecrawl/pdf-inspector-js"
-    }
+    ]
   },
   "scripts": {
     "build": "napi build --platform --release",
@@ -51,5 +50,13 @@
   },
   "devDependencies": {
     "@napi-rs/cli": "^3.4.1"
+  },
+  "optionalDependencies": {
+    "@firecrawl/pdf-inspector-linux-x64-gnu": "1.14.0",
+    "@firecrawl/pdf-inspector-linux-x64-musl": "1.14.0",
+    "@firecrawl/pdf-inspector-linux-arm64-gnu": "1.14.0",
+    "@firecrawl/pdf-inspector-linux-arm64-musl": "1.14.0",
+    "@firecrawl/pdf-inspector-darwin-arm64": "1.14.0",
+    "@firecrawl/pdf-inspector-win32-x64-msvc": "1.14.0"
   }
 }

napi/src/lib.rs

@@ -40,6 +40,8 @@ pub struct PdfResult {
     pub processing_time_ms: u32,
     /// 1-indexed page numbers that need OCR.
     pub pages_needing_ocr: Vec<u32>,
+    /// Machine-readable OCR reasons by 1-indexed page.
+    pub ocr_reasons_by_page: Vec<PageOcrReasons>,
     pub title: Option<String>,
     pub confidence: f64,
     pub is_complex_layout: bool,
@@ -48,6 +50,13 @@ pub struct PdfResult {
     pub has_encoding_issues: bool,
 }
 
+/// OCR reasons for a single 1-indexed page.
+#[napi(object)]
+pub struct PageOcrReasons {
+    pub page: u32,
+    pub reasons: Vec<String>,
+}
+
 /// Lightweight PDF classification result.
 #[napi(object)]
 pub struct PdfClassification {
@@ -71,9 +80,20 @@ pub struct TextItem {
     pub page: u32,
     pub is_bold: bool,
     pub is_italic: bool,
+    /// Underline detected geometrically (drawn rule/thin rect under the
+    /// baseline) — PDFs carry no underline font flag.
+    pub is_underline: bool,
+    /// Strikeout detected geometrically (rule crossing the glyphs at mid
+    /// x-height).
+    pub is_strikeout: bool,
     pub item_type: ItemType,
     /// URL for link items, `None` for other types.
     pub link_url: Option<String>,
+    /// Marked Content ID from the content stream's BDC/BMC operator, `None`
+    /// when the text is not part of marked content. Join with the
+    /// `page`/`mcid` pairs from [`extractStructureElements`] to attach
+    /// structure-tree roles (headings, paragraphs, …) in tagged PDFs.
+    pub mcid: Option<i64>,
 }
 
 /// A page's regions for text extraction: (page_index_0based, bboxes).
@@ -90,6 +110,8 @@ pub struct RegionText {
     pub text: String,
     /// `true` when the text should not be trusted (empty, GID fonts, garbage, encoding issues).
     pub needs_ocr: bool,
+    /// Machine-readable OCR reason when the cause is known.
+    pub ocr_reason: Option<String>,
 }
 
 /// Extracted text for one page's regions.
@@ -126,6 +148,7 @@ fn to_napi_result(r: pdf_inspector::PdfProcessResult) -> PdfResult {
         page_count: r.page_count,
         processing_time_ms: r.processing_time_ms as u32,
         pages_needing_ocr: r.pages_needing_ocr,
+        ocr_reasons_by_page: to_napi_page_ocr_reasons(r.ocr_reasons_by_page),
         title: r.title,
         confidence: r.confidence as f64,
         is_complex_layout: r.layout.is_complex,
@@ -135,6 +158,16 @@ fn to_napi_result(r: pdf_inspector::PdfProcessResult) -> PdfResult {
     }
 }
 
+fn to_napi_page_ocr_reasons(reasons: Vec<pdf_inspector::PageOcrReasons>) -> Vec<PageOcrReasons> {
+    reasons
+        .into_iter()
+        .map(|reason| PageOcrReasons {
+            page: reason.page,
+            reasons: reason.reasons,
+        })
+        .collect()
+}
+
 fn convert_item_type(t: &pdf_inspector::types::ItemType) -> (ItemType, Option<String>) {
     match t {
         pdf_inspector::types::ItemType::Text => (ItemType::Text, None),
@@ -172,6 +205,31 @@ where
     }
 }
 
+// ---------------------------------------------------------------------------
+// Shared implementations (single body behind sync and async entry points)
+// ---------------------------------------------------------------------------
+
+fn process_pdf_impl(bytes: &[u8], pages: Option<Vec<u32>>) -> Result<PdfResult> {
+    let mut opts = pdf_inspector::PdfOptions::new();
+    if let Some(p) = pages {
+        opts = opts.pages(p);
+    }
+    let result = pdf_inspector::process_pdf_mem_with_options(bytes, opts)
+        .map_err(|e| to_napi_err(e, "process_pdf"))?;
+    Ok(to_napi_result(result))
+}
+
+fn classify_pdf_impl(bytes: &[u8]) -> Result<PdfClassification> {
+    let result =
+        pdf_inspector::classify_pdf_mem(bytes).map_err(|e| to_napi_err(e, "classify_pdf"))?;
+    Ok(PdfClassification {
+        pdf_type: convert_pdf_type(result.pdf_type),
+        page_count: result.page_count,
+        pages_needing_ocr: result.pages_needing_ocr,
+        confidence: result.confidence as f64,
+    })
+}
+
 // ---------------------------------------------------------------------------
 // Public NAPI API
 // ---------------------------------------------------------------------------
@@ -180,15 +238,7 @@ where
 #[napi]
 pub fn process_pdf(buffer: Buffer, pages: Option<Vec<u32>>) -> Result<PdfResult> {
     let bytes: Vec<u8> = buffer.to_vec();
-    catch_panic("process_pdf", move || {
-        let mut opts = pdf_inspector::PdfOptions::new();
-        if let Some(p) = pages {
-            opts = opts.pages(p);
-        }
-        let result = pdf_inspector::process_pdf_mem_with_options(&bytes, opts)
-            .map_err(|e| to_napi_err(e, "process_pdf"))?;
-        Ok(to_napi_result(result))
-    })
+    catch_panic("process_pdf", move || process_pdf_impl(&bytes, pages))
 }
 
 /// Fast detection only — no text extraction or markdown.
@@ -208,16 +258,7 @@ pub fn detect_pdf(buffer: Buffer) -> Result<PdfResult> {
 #[napi]
 pub fn classify_pdf(buffer: Buffer) -> Result<PdfClassification> {
     let bytes: Vec<u8> = buffer.to_vec();
-    catch_panic("classify_pdf", move || {
-        let result =
-            pdf_inspector::classify_pdf_mem(&bytes).map_err(|e| to_napi_err(e, "classify_pdf"))?;
-        Ok(PdfClassification {
-            pdf_type: convert_pdf_type(result.pdf_type),
-            page_count: result.page_count,
-            pages_needing_ocr: result.pages_needing_ocr,
-            confidence: result.confidence as f64,
-        })
-    })
+    catch_panic("classify_pdf", move || classify_pdf_impl(&bytes))
 }
 
 /// Extract plain text from a PDF Buffer.
@@ -266,14 +307,65 @@ pub fn extract_text_with_positions(
                     page: item.page,
                     is_bold: item.is_bold,
                     is_italic: item.is_italic,
+                    is_underline: item.is_underline,
+                    is_strikeout: item.is_strikeout,
                     item_type,
                     link_url,
+                    mcid: item.mcid,
                 }
             })
             .collect())
     })
 }
 
+/// One structure-tree element reference from a tagged PDF.
+#[napi(object)]
+pub struct StructureElementJs {
+    /// 1-indexed page number (matches `TextItem.page`).
+    pub page: u32,
+    /// Marked Content ID from the page's content stream (matches
+    /// `TextItem.mcid`).
+    pub mcid: i64,
+    /// Standard structure type name ("H1".."H6", "P", "Table", "TD", …).
+    /// Custom tags are resolved through the document's role map; tags with
+    /// no standard mapping are returned verbatim.
+    pub role: String,
+}
+
+/// Extract structure-tree element references from a tagged PDF.
+///
+/// Parses the document's structure tree (when present) and returns one
+/// entry per marked-content reference, resolved to its 1-indexed page,
+/// MCID, and structure type name. Returns an empty array when the PDF is
+/// not tagged.
+///
+/// Join `(page, mcid)` against the `page`/`mcid` fields from
+/// [`extractTextWithPositions`] to attach heading levels (H1..H6) and other
+/// semantic roles to extracted text.
+///
+/// Pass 1-indexed page numbers (matching `TextItem.page`) to restrict
+/// output; omit `pages` for the whole document. Entries are sorted by
+/// `(page, mcid)`.
+#[napi]
+pub fn extract_structure_elements(
+    buffer: Buffer,
+    pages: Option<Vec<u32>>,
+) -> Result<Vec<StructureElementJs>> {
+    let bytes: Vec<u8> = buffer.to_vec();
+    catch_panic("extract_structure_elements", move || {
+        let elements = pdf_inspector::extract_structure_elements_mem(&bytes, pages.as_deref())
+            .map_err(|e| to_napi_err(e, "extract_structure_elements"))?;
+        Ok(elements
+            .into_iter()
+            .map(|e| StructureElementJs {
+                page: e.page,
+                mcid: e.mcid,
+                role: e.role,
+            })
+            .collect())
+    })
+}
+
 /// Extract text within bounding-box regions from a PDF.
 ///
 /// For hybrid OCR: layout model detects regions in rendered images,
@@ -563,6 +655,8 @@ pub struct PageMarkdownResult {
     pub markdown: String,
     /// `true` when text on this page is unreliable.
     pub needs_ocr: bool,
+    /// Machine-readable OCR reason when the cause is known.
+    pub ocr_reason: Option<String>,
 }
 
 /// Combined per-page markdown extraction and layout classification result.
@@ -576,6 +670,8 @@ pub struct PagesExtractionResult {
     pub pages_with_columns: Vec<u32>,
     /// 1-indexed pages that need OCR (scanned/image-based).
     pub pages_needing_ocr: Vec<u32>,
+    /// Machine-readable OCR reasons by 1-indexed page.
+    pub ocr_reasons_by_page: Vec<PageOcrReasons>,
     /// True if any page has tables or columns.
     pub is_complex: bool,
 }
@@ -597,23 +693,32 @@ pub fn extract_pages_markdown(
 ) -> Result<PagesExtractionResult> {
     let bytes: Vec<u8> = buffer.to_vec();
     catch_panic("extract_pages_markdown", move || {
-        let result = pdf_inspector::extract_pages_markdown_mem(&bytes, pages.as_deref())
-            .map_err(|e| to_napi_err(e, "extract_pages_markdown"))?;
-        Ok(PagesExtractionResult {
-            pages: result
-                .pages
-                .into_iter()
-                .map(|r| PageMarkdownResult {
-                    page: r.page,
-                    markdown: r.markdown,
-                    needs_ocr: r.needs_ocr,
-                })
-                .collect(),
-            pages_with_tables: result.pages_with_tables,
-            pages_with_columns: result.pages_with_columns,
-            pages_needing_ocr: result.pages_needing_ocr,
-            is_complex: result.is_complex,
-        })
+        extract_pages_markdown_impl(&bytes, pages.as_deref())
+    })
+}
+
+fn extract_pages_markdown_impl(
+    bytes: &[u8],
+    pages: Option<&[u32]>,
+) -> Result<PagesExtractionResult> {
+    let result = pdf_inspector::extract_pages_markdown_mem(bytes, pages)
+        .map_err(|e| to_napi_err(e, "extract_pages_markdown"))?;
+    Ok(PagesExtractionResult {
+        pages: result
+            .pages
+            .into_iter()
+            .map(|r| PageMarkdownResult {
+                page: r.page,
+                markdown: r.markdown,
+                needs_ocr: r.needs_ocr,
+                ocr_reason: r.ocr_reason,
+            })
+            .collect(),
+        pages_with_tables: result.pages_with_tables,
+        pages_with_columns: result.pages_with_columns,
+        pages_needing_ocr: result.pages_needing_ocr,
+        ocr_reasons_by_page: to_napi_page_ocr_reasons(result.ocr_reasons_by_page),
+        is_complex: result.is_complex,
     })
 }
 
@@ -648,8 +753,137 @@ fn to_page_region_texts(results: Vec<pdf_inspector::PageRegionResult>) -> Vec<Pa
                 .map(|r| RegionText {
                     text: r.text,
                     needs_ocr: r.needs_ocr,
+                    ocr_reason: r.ocr_reason,
                 })
                 .collect(),
         })
         .collect()
 }
+
+// ---------------------------------------------------------------------------
+// Async variants (libuv thread pool via AsyncTask)
+//
+// The synchronous exports above parse on the calling thread, which in Node is
+// the event loop. These `*Async` variants run the same shared implementations
+// on the libuv thread pool and hand JavaScript a promise, so servers under
+// concurrent load keep answering requests while a document parses. The sync
+// exports keep their names, signatures, and behaviour.
+//
+// Each factory copies the input Buffer to an owned `Vec<u8>` on the calling
+// (JS) thread — deliberately. JS execution is single-threaded, so no JS code
+// can mutate the buffer while the synchronous part of the call copies it.
+// Holding the napi `Buffer` and reading it from the worker instead would be
+// zero-copy, but a caller mutating the buffer before the promise settles
+// would then race the worker's reads — undefined behavior, not a recoverable
+// error (a known napi-rs soundness hazard with cross-thread Buffer access).
+// The copy is a one-time memcpy, negligible next to the parse it unblocks.
+// ---------------------------------------------------------------------------
+
+pub struct ProcessPdfTask {
+    bytes: Vec<u8>,
+    pages: Option<Vec<u32>>,
+}
+
+impl Task for ProcessPdfTask {
+    type Output = PdfResult;
+    type JsValue = PdfResult;
+
+    fn compute(&mut self) -> Result<Self::Output> {
+        let bytes = std::mem::take(&mut self.bytes);
+        let pages = self.pages.take();
+        // AssertUnwindSafe: `bytes`/`pages` are moved into the closure and
+        // dropped on unwind — no shared state can be observed broken.
+        catch_panic(
+            "process_pdf",
+            panic::AssertUnwindSafe(move || process_pdf_impl(&bytes, pages)),
+        )
+    }
+
+    fn resolve(&mut self, _env: Env, output: Self::Output) -> Result<Self::JsValue> {
+        Ok(output)
+    }
+}
+
+/// Async variant of [`processPdf`]: same result, but the parse runs on the
+/// libuv thread pool instead of the event loop and the call returns a
+/// promise. The buffer is copied before the call returns, so it may be
+/// reused or mutated immediately.
+// ts_return_type is required: napi-rs emits `Promise<unknown>` for
+// `AsyncTask<T>` returns without it.
+#[napi(ts_return_type = "Promise<PdfResult>")]
+pub fn process_pdf_async(buffer: Buffer, pages: Option<Vec<u32>>) -> AsyncTask<ProcessPdfTask> {
+    AsyncTask::new(ProcessPdfTask {
+        bytes: buffer.to_vec(),
+        pages,
+    })
+}
+
+pub struct ClassifyPdfTask {
+    bytes: Vec<u8>,
+}
+
+impl Task for ClassifyPdfTask {
+    type Output = PdfClassification;
+    type JsValue = PdfClassification;
+
+    fn compute(&mut self) -> Result<Self::Output> {
+        let bytes = std::mem::take(&mut self.bytes);
+        catch_panic(
+            "classify_pdf",
+            panic::AssertUnwindSafe(move || classify_pdf_impl(&bytes)),
+        )
+    }
+
+    fn resolve(&mut self, _env: Env, output: Self::Output) -> Result<Self::JsValue> {
+        Ok(output)
+    }
+}
+
+/// Async variant of [`classifyPdf`]: same result, but the classification runs
+/// on the libuv thread pool instead of the event loop and the call returns a
+/// promise. The buffer is copied before the call returns, so it may be
+/// reused or mutated immediately.
+#[napi(ts_return_type = "Promise<PdfClassification>")]
+pub fn classify_pdf_async(buffer: Buffer) -> AsyncTask<ClassifyPdfTask> {
+    AsyncTask::new(ClassifyPdfTask {
+        bytes: buffer.to_vec(),
+    })
+}
+
+pub struct ExtractPagesMarkdownTask {
+    bytes: Vec<u8>,
+    pages: Option<Vec<u32>>,
+}
+
+impl Task for ExtractPagesMarkdownTask {
+    type Output = PagesExtractionResult;
+    type JsValue = PagesExtractionResult;
+
+    fn compute(&mut self) -> Result<Self::Output> {
+        let bytes = std::mem::take(&mut self.bytes);
+        let pages = self.pages.take();
+        catch_panic(
+            "extract_pages_markdown",
+            panic::AssertUnwindSafe(move || extract_pages_markdown_impl(&bytes, pages.as_deref())),
+        )
+    }
+
+    fn resolve(&mut self, _env: Env, output: Self::Output) -> Result<Self::JsValue> {
+        Ok(output)
+    }
+}
+
+/// Async variant of [`extractPagesMarkdown`]: same result, but the extraction
+/// runs on the libuv thread pool instead of the event loop and the call
+/// returns a promise. The buffer is copied before the call returns, so it
+/// may be reused or mutated immediately.
+#[napi(ts_return_type = "Promise<PagesExtractionResult>")]
+pub fn extract_pages_markdown_async(
+    buffer: Buffer,
+    pages: Option<Vec<u32>>,
+) -> AsyncTask<ExtractPagesMarkdownTask> {
+    AsyncTask::new(ExtractPagesMarkdownTask {
+        bytes: buffer.to_vec(),
+        pages,
+    })
+}

napi/test.mjs

@@ -2,16 +2,21 @@ import { readFileSync } from 'fs';
 import { strict as assert } from 'assert';
 import {
   processPdf,
+  processPdfAsync,
   detectPdf,
   classifyPdf,
+  classifyPdfAsync,
   extractText,
   extractTextWithPositions,
+  extractStructureElements,
   extractTextInRegions,
   detectVectorGridInRegion,
   extractPagesMarkdown,
+  extractPagesMarkdownAsync,
 } from './index.js';
 
 const fixture = readFileSync('../tests/fixtures/thermo-freon12.pdf');
+const taggedFixture = readFileSync('../tests/fixtures/firecrawl_docs_tagged.pdf');
 
 // --- processPdf ---
 console.log('Testing processPdf...');
@@ -79,6 +84,46 @@ assert.ok(page1Items.length > 0);
 assert.ok(page1Items.every(i => i.page === 1));
 console.log('  extractTextWithPositions with pages: OK');
 
+// mcid: undefined on untagged PDFs, numeric on tagged marked content
+assert.ok(items.every(i => i.mcid === undefined || typeof i.mcid === 'number'));
+const taggedItems = extractTextWithPositions(taggedFixture);
+assert.ok(
+  taggedItems.some(i => typeof i.mcid === 'number'),
+  'tagged PDF text items should carry Marked Content IDs',
+);
+console.log('  extractTextWithPositions mcid: OK');
+
+// --- extractStructureElements ---
+console.log('Testing extractStructureElements...');
+const structureElements = extractStructureElements(taggedFixture);
+assert.ok(structureElements.length > 0);
+assert.ok(structureElements.every(e => typeof e.page === 'number'));
+assert.ok(structureElements.every(e => typeof e.mcid === 'number'));
+assert.ok(structureElements.every(e => typeof e.role === 'string' && e.role.length > 0));
+assert.ok(
+  structureElements.some(e => e.role === 'H1'),
+  'tagged fixture should surface H1 heading roles',
+);
+
+// (page, mcid) joins against extractTextWithPositions to recover heading text
+const h1Refs = new Set(
+  structureElements.filter(e => e.role === 'H1').map(e => `${e.page}:${e.mcid}`),
+);
+const h1Text = taggedItems
+  .filter(i => typeof i.mcid === 'number' && h1Refs.has(`${i.page}:${i.mcid}`))
+  .map(i => i.text)
+  .join('');
+assert.ok(h1Text.trim().length > 0, 'H1 join should recover heading text');
+
+// pages filter is 1-indexed, matching TextItem.page
+const page1Elements = extractStructureElements(taggedFixture, [1]);
+assert.ok(page1Elements.length > 0);
+assert.ok(page1Elements.every(e => e.page === 1));
+
+// untagged PDFs yield an empty array
+assert.deepEqual(extractStructureElements(fixture), []);
+console.log('  extractStructureElements: OK');
+
 // --- extractTextInRegions ---
 console.log('Testing extractTextInRegions...');
 const regionResults = extractTextInRegions(fixture, [
@@ -124,10 +169,75 @@ assert.equal(picked.pages[0].page, 2);
 assert.equal(picked.pages[1].page, 0);
 console.log('  extractPagesMarkdown with pages: OK');
 
+// --- Async variants ---
+console.log('Testing async variants...');
+
+// processPdfAsync returns a promise and matches the sync result
+const asyncResultPromise = processPdfAsync(fixture);
+assert.ok(asyncResultPromise instanceof Promise);
+const asyncResult = await asyncResultPromise;
+assert.equal(asyncResult.pdfType, result.pdfType);
+assert.equal(asyncResult.pageCount, result.pageCount);
+assert.equal(asyncResult.markdown, result.markdown);
+console.log('  processPdfAsync: OK');
+
+// processPdfAsync with pages
+const asyncResult2 = await processPdfAsync(fixture, [1]);
+assert.equal(asyncResult2.markdown, result2.markdown);
+console.log('  processPdfAsync with pages: OK');
+
+// classifyPdfAsync matches the sync result
+const asyncClassified = await classifyPdfAsync(fixture);
+assert.equal(asyncClassified.pdfType, classified.pdfType);
+assert.equal(asyncClassified.pageCount, classified.pageCount);
+assert.equal(asyncClassified.confidence, classified.confidence);
+assert.deepEqual(asyncClassified.pagesNeedingOcr, classified.pagesNeedingOcr);
+console.log('  classifyPdfAsync: OK');
+
+// extractPagesMarkdownAsync matches the sync result
+const asyncAllPages = await extractPagesMarkdownAsync(fixture);
+assert.equal(asyncAllPages.pages.length, allPages.pages.length);
+assert.deepEqual(
+  asyncAllPages.pages.map(p => p.markdown),
+  allPages.pages.map(p => p.markdown),
+);
+assert.equal(asyncAllPages.isComplex, allPages.isComplex);
+console.log('  extractPagesMarkdownAsync: OK');
+
+// selected pages preserve caller order
+const asyncPicked = await extractPagesMarkdownAsync(fixture, [2, 0]);
+assert.equal(asyncPicked.pages.length, 2);
+assert.equal(asyncPicked.pages[0].page, 2);
+assert.equal(asyncPicked.pages[1].page, 0);
+console.log('  extractPagesMarkdownAsync with pages: OK');
+
+// input buffer is copied at call time: mutating it immediately after the
+// call must not affect the in-flight parse
+const scratch = Buffer.from(fixture);
+const inFlight = processPdfAsync(scratch);
+scratch.fill(0);
+const fromMutated = await inFlight;
+assert.equal(fromMutated.markdown, result.markdown);
+console.log('  processPdfAsync input copied at call time: OK');
+
+// concurrent async calls all settle
+const [c1, c2, c3] = await Promise.all([
+  processPdfAsync(fixture),
+  classifyPdfAsync(fixture),
+  extractPagesMarkdownAsync(fixture),
+]);
+assert.equal(c1.pdfType, 'TextBased');
+assert.equal(c2.pdfType, 'TextBased');
+assert.equal(c3.pages.length, 3);
+console.log('  concurrent async calls: OK');
+
 // --- Error handling ---
 console.log('Testing error handling...');
 assert.throws(() => processPdf(Buffer.from('not a pdf')), /process_pdf/);
 assert.throws(() => classifyPdf(Buffer.from('')), /classify_pdf/);
+await assert.rejects(processPdfAsync(Buffer.from('not a pdf')), /process_pdf/);
+await assert.rejects(classifyPdfAsync(Buffer.from('')), /classify_pdf/);
+await assert.rejects(extractPagesMarkdownAsync(Buffer.from('')), /extract_pages_markdown/);
 console.log('  error handling: OK');
 
 console.log('\nAll NAPI tests passed!');

pdf_inspector.pyi

@@ -10,6 +10,9 @@ class PdfResult:
     page_count: int
     processing_time_ms: int
     pages_needing_ocr: list[int]
+    """1-indexed page numbers that need OCR."""
+    ocr_reasons_by_page: list["PageOcrReasons"]
+    """Machine-readable OCR reasons by 1-indexed page."""
     title: Optional[str]
     confidence: float
     is_complex_layout: bool
@@ -17,6 +20,13 @@ class PdfResult:
     pages_with_columns: list[int]
     has_encoding_issues: bool
 
+class PageOcrReasons:
+    """OCR reasons for a single 1-indexed page."""
+    page: int
+    """1-indexed page number."""
+    reasons: list[str]
+    """Machine-readable OCR reason identifiers."""
+
 class PdfClassification:
     """Lightweight PDF classification result."""
     pdf_type: str
@@ -38,13 +48,31 @@ class TextItem:
     page: int
     is_bold: bool
     is_italic: bool
+    is_underline: bool
+    is_strikeout: bool
     item_type: str
+    mcid: Optional[int]
+    """Marked Content ID from the content stream's BDC/BMC operator, None when
+    the text is not part of marked content. Join with the (page, mcid) pairs
+    from extract_structure_elements to attach structure-tree roles in tagged
+    PDFs."""
+
+class StructureElement:
+    """One structure-tree element reference from a tagged PDF."""
+    page: int
+    """1-indexed page number (matches TextItem.page)."""
+    mcid: int
+    """Marked Content ID from the page's content stream (matches TextItem.mcid)."""
+    role: str
+    """Standard structure type name ("H1".."H6", "P", "Table", "TD", ...)."""
 
 class RegionText:
     """Extracted text for a single region."""
     text: str
     needs_ocr: bool
     """True when the text should not be trusted."""
+    ocr_reason: Optional[str]
+    """Machine-readable OCR reason when the cause is known."""
 
 class PageRegionTexts:
     """Extracted text for one page's regions."""
@@ -60,6 +88,8 @@ class PageMarkdown:
     """Formatted markdown for this page (empty string when needs_ocr is True)."""
     needs_ocr: bool
     """True when text on this page is unreliable and OCR should be used instead."""
+    ocr_reason: Optional[str]
+    """Machine-readable OCR reason when the cause is known."""
 
 class PagesExtractionResult:
     """Per-page markdown output with document-wide layout classification."""
@@ -71,6 +101,8 @@ class PagesExtractionResult:
     """1-indexed pages where multi-column layout was detected."""
     pages_needing_ocr: list[int]
     """1-indexed pages that need OCR."""
+    ocr_reasons_by_page: list[PageOcrReasons]
+    """Machine-readable OCR reasons by 1-indexed page."""
     is_complex: bool
     """True if any page has tables or multi-column layout."""
 
@@ -114,6 +146,27 @@ def extract_text_with_positions_bytes(data: bytes, pages: Optional[list[int]] =
     """Extract text with position information from bytes."""
     ...
 
+def extract_structure_elements(path: str, pages: Optional[list[int]] = None) -> list[StructureElement]:
+    """Extract structure-tree element references from a tagged PDF file.
+
+    Returns one entry per marked-content reference, resolved to its 1-indexed
+    page, MCID, and structure type name ("H1".."H6", "P", "Table", ...), sorted
+    by (page, mcid). Returns an empty list when the PDF is not tagged.
+
+    Args:
+        path: Path to the PDF file.
+        pages: Optional list of 1-indexed pages (matching ``TextItem.page``).
+            When ``None`` (default), the whole document is returned.
+    """
+    ...
+
+def extract_structure_elements_bytes(data: bytes, pages: Optional[list[int]] = None) -> list[StructureElement]:
+    """Extract structure-tree element references from tagged PDF bytes.
+
+    See :func:`extract_structure_elements` for details.
+    """
+    ...
+
 def extract_text_in_regions(
     path: str,
     page_regions: list[tuple[int, list[list[float]]]],

pyproject.toml

@@ -4,8 +4,11 @@ build-backend = "maturin"
 
 [project]
 name = "pdf-inspector"
-version = "0.1.0"
+# Keep package versions in sync with `python3 scripts/version.py <version>`.
+# CI publishes automatically when the synchronized change lands on main.
+version = "1.14.0"
 description = "Fast PDF inspection, classification, and text extraction with smart scanned vs text-based detection"
+readme = "docs/python.md"
 license = { text = "MIT" }
 requires-python = ">=3.8"
 classifiers = [
@@ -17,5 +20,10 @@ classifiers = [
     "Topic :: Text Processing",
 ]
 
+[project.urls]
+Homepage = "https://github.com/firecrawl/pdf-inspector"
+Repository = "https://github.com/firecrawl/pdf-inspector"
+Documentation = "https://github.com/firecrawl/pdf-inspector/blob/main/docs/python.md"
+
 [tool.maturin]
 features = ["python"]

scripts/bench_opendataloader.py

@@ -0,0 +1,351 @@
+#!/usr/bin/env python3
+"""Run a paired pdf-inspector OpenDataLoader benchmark and report deltas."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import math
+import os
+import shutil
+import subprocess
+import sys
+import tempfile
+from pathlib import Path
+from typing import Any
+
+
+SCORE_KEYS = (
+    "overall_mean",
+    "nid_mean",
+    "nid_s_mean",
+    "teds_mean",
+    "teds_s_mean",
+    "mhs_mean",
+    "mhs_s_mean",
+)
+
+
+def _non_negative_int(value: str) -> int:
+    parsed = int(value)
+    if parsed < 0:
+        raise argparse.ArgumentTypeError("must be non-negative")
+    return parsed
+
+
+def _non_negative_float(value: str) -> float:
+    parsed = float(value)
+    if not math.isfinite(parsed) or parsed < 0.0:
+        raise argparse.ArgumentTypeError("must be finite and non-negative")
+    return parsed
+
+
+def _finite_float(value: str) -> float:
+    parsed = float(value)
+    if not math.isfinite(parsed):
+        raise argparse.ArgumentTypeError("must be finite")
+    return parsed
+
+
+def _scores(evaluation: dict[str, Any]) -> dict[str, float]:
+    score = evaluation.get("metrics", {}).get("score", {})
+    return {key: float(score[key]) for key in SCORE_KEYS if score.get(key) is not None}
+
+
+def _documents(evaluation: dict[str, Any]) -> dict[str, float]:
+    documents: dict[str, float] = {}
+    for document in evaluation.get("documents", []):
+        overall = document.get("scores", {}).get("overall")
+        if overall is not None:
+            documents[str(document["document_id"])] = float(overall)
+    return documents
+
+
+def compare_evaluations(
+    baseline: dict[str, Any],
+    candidate: dict[str, Any],
+    reference: dict[str, Any] | None = None,
+    *,
+    top: int = 10,
+) -> dict[str, Any]:
+    """Build aggregate and per-document deltas from evaluator JSON payloads."""
+    baseline_scores = _scores(baseline)
+    candidate_scores = _scores(candidate)
+    metric_deltas = {
+        key: candidate_scores[key] - baseline_scores[key]
+        for key in SCORE_KEYS
+        if key in baseline_scores and key in candidate_scores
+    }
+
+    baseline_documents = _documents(baseline)
+    candidate_documents = _documents(candidate)
+    shared = sorted(baseline_documents.keys() & candidate_documents.keys())
+    document_deltas = [
+        {
+            "document_id": document_id,
+            "baseline": baseline_documents[document_id],
+            "candidate": candidate_documents[document_id],
+            "delta": candidate_documents[document_id] - baseline_documents[document_id],
+        }
+        for document_id in shared
+    ]
+    epsilon = 1e-12
+    improvements = sorted(document_deltas, key=lambda item: item["delta"], reverse=True)
+    regressions = sorted(document_deltas, key=lambda item: item["delta"])
+
+    result: dict[str, Any] = {
+        "baseline": baseline_scores,
+        "candidate": candidate_scores,
+        "deltas": metric_deltas,
+        "missing_predictions": {
+            "baseline": int(baseline.get("metrics", {}).get("missing_predictions", 0)),
+            "candidate": int(candidate.get("metrics", {}).get("missing_predictions", 0)),
+        },
+        "documents": {
+            "shared": len(shared),
+            "improved": sum(item["delta"] > epsilon for item in document_deltas),
+            "regressed": sum(item["delta"] < -epsilon for item in document_deltas),
+            "unchanged": sum(abs(item["delta"]) <= epsilon for item in document_deltas),
+            "largest_improvements": [
+                item for item in improvements if item["delta"] > epsilon
+            ][:top],
+            "largest_regressions": [
+                item for item in regressions if item["delta"] < -epsilon
+            ][:top],
+            "worst_regression": next(
+                (item for item in regressions if item["delta"] < -epsilon), None
+            ),
+        },
+    }
+    if reference is not None:
+        reference_scores = _scores(reference)
+        result["reference"] = reference_scores
+        result["candidate_vs_reference"] = {
+            key: candidate_scores[key] - reference_scores[key]
+            for key in SCORE_KEYS
+            if key in candidate_scores and key in reference_scores
+        }
+    return result
+
+
+def evaluate_gates(
+    comparison: dict[str, Any],
+    *,
+    min_overall_delta: float,
+    max_document_regression: float | None,
+    max_missing: int,
+    require_reference_lead: bool,
+) -> list[str]:
+    """Return human-readable gate failures; an empty list means pass."""
+    failures: list[str] = []
+    overall_delta = comparison["deltas"].get("overall_mean")
+    if overall_delta is None or overall_delta < min_overall_delta:
+        failures.append(
+            f"overall delta {overall_delta!r} is below {min_overall_delta:+.6f}"
+        )
+    candidate_missing = comparison["missing_predictions"]["candidate"]
+    if candidate_missing > max_missing:
+        failures.append(
+            f"candidate has {candidate_missing} missing predictions (maximum {max_missing})"
+        )
+    if max_document_regression is not None:
+        regression = comparison["documents"].get("worst_regression")
+        if regression is not None and regression["delta"] < -max_document_regression:
+            failures.append(
+                "largest document regression "
+                f"{regression['document_id']}={regression['delta']:+.6f} "
+                f"exceeds {-max_document_regression:+.6f}"
+            )
+    if require_reference_lead:
+        reference_delta = comparison.get("candidate_vs_reference", {}).get("overall_mean")
+        if reference_delta is None:
+            failures.append("reference overall score is unavailable")
+        elif reference_delta < 0.0:
+            failures.append(
+                f"candidate trails reference overall by {reference_delta!r}"
+            )
+    return failures
+
+
+def _run(command: list[str], *, cwd: Path, env: dict[str, str] | None = None) -> None:
+    print("+", " ".join(command), flush=True)
+    subprocess.run(command, cwd=cwd, env=env, check=True)
+
+
+def _run_engine(
+    *,
+    bench_dir: Path,
+    python: Path,
+    binary: Path,
+    label: str,
+    scratch_root: Path,
+) -> dict[str, Any]:
+    env = os.environ.copy()
+    env["PDF_INSPECTOR_BINARY"] = str(binary)
+    source = bench_dir / "prediction" / "pdf-inspector"
+    if source.exists():
+        if source.is_dir():
+            shutil.rmtree(source)
+        else:
+            source.unlink()
+    _run(
+        [
+            str(python),
+            "src/pdf_parser.py",
+            "--engine",
+            "pdf-inspector",
+            "--log-level",
+            "WARNING",
+        ],
+        cwd=bench_dir,
+        env=env,
+    )
+
+    if not source.is_dir():
+        raise RuntimeError(f"parser did not produce predictions: {source}")
+    destination = scratch_root / label
+    shutil.copytree(source, destination)
+    _run(
+        [
+            str(python),
+            "src/evaluator.py",
+            "--prediction-root",
+            str(scratch_root),
+            "--engine",
+            label,
+            "--log-level",
+            "WARNING",
+        ],
+        cwd=bench_dir,
+    )
+    with (destination / "evaluation.json").open(encoding="utf-8") as handle:
+        return json.load(handle)
+
+
+def _print_report(comparison: dict[str, Any]) -> None:
+    print("\nMetric                 baseline   candidate       delta")
+    print("--------------------  ----------  ----------  ----------")
+    for key in SCORE_KEYS:
+        if key not in comparison["deltas"]:
+            continue
+        print(
+            f"{key:<20}  {comparison['baseline'][key]:>10.6f}  "
+            f"{comparison['candidate'][key]:>10.6f}  "
+            f"{comparison['deltas'][key]:>+10.6f}"
+        )
+    if "reference" in comparison:
+        delta = comparison["candidate_vs_reference"].get("overall_mean")
+        reference = comparison["reference"].get("overall_mean")
+        reference_display = f"{reference:.6f}" if reference is not None else "n/a"
+        delta_display = f"{delta:+.6f}" if delta is not None else "n/a"
+        print(f"\nReference overall: {reference_display}; candidate delta: {delta_display}")
+
+    documents = comparison["documents"]
+    print(
+        "\nDocuments: "
+        f"{documents['improved']} improved, {documents['regressed']} regressed, "
+        f"{documents['unchanged']} unchanged ({documents['shared']} shared)"
+    )
+    for heading, key in (
+        ("Largest improvements", "largest_improvements"),
+        ("Largest regressions", "largest_regressions"),
+    ):
+        print(f"\n{heading}:")
+        rows = documents[key]
+        if not rows:
+            print("  none")
+        for row in rows:
+            print(
+                f"  {row['document_id']}: {row['delta']:+.6f} "
+                f"({row['baseline']:.6f} -> {row['candidate']:.6f})"
+            )
+
+
+def _arguments(argv: list[str] | None = None) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("--bench-dir", type=Path, required=True)
+    parser.add_argument("--baseline", type=Path, required=True)
+    parser.add_argument("--candidate", type=Path, required=True)
+    parser.add_argument("--python", type=Path)
+    parser.add_argument("--reference-evaluation", type=Path)
+    parser.add_argument("--json-output", type=Path)
+    parser.add_argument("--top", type=_non_negative_int, default=10)
+    parser.add_argument("--min-overall-delta", type=_finite_float, default=0.0)
+    parser.add_argument("--max-document-regression", type=_non_negative_float)
+    parser.add_argument("--max-missing", type=_non_negative_int, default=0)
+    parser.add_argument("--require-reference-lead", action="store_true")
+    return parser.parse_args(argv)
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = _arguments(argv)
+    bench_dir = args.bench_dir.resolve()
+    baseline = args.baseline.resolve()
+    candidate = args.candidate.resolve()
+    # Keep the virtualenv launcher path intact. Resolving its symlink would
+    # invoke the underlying system interpreter without the benchmark's site
+    # packages.
+    python = (args.python or bench_dir / ".venv" / "bin" / "python").absolute()
+    for path, description in (
+        (bench_dir / "src" / "pdf_parser.py", "OpenDataLoader parser"),
+        (bench_dir / "src" / "evaluator.py", "OpenDataLoader evaluator"),
+        (baseline, "baseline binary"),
+        (candidate, "candidate binary"),
+        (python, "Python interpreter"),
+    ):
+        if not path.exists():
+            raise SystemExit(f"{description} not found: {path}")
+
+    with tempfile.TemporaryDirectory(prefix="pdf-inspector-opendataloader-") as temporary:
+        scratch_root = Path(temporary)
+        baseline_evaluation = _run_engine(
+            bench_dir=bench_dir,
+            python=python,
+            binary=baseline,
+            label="baseline",
+            scratch_root=scratch_root,
+        )
+        candidate_evaluation = _run_engine(
+            bench_dir=bench_dir,
+            python=python,
+            binary=candidate,
+            label="candidate",
+            scratch_root=scratch_root,
+        )
+
+        reference = None
+        if args.reference_evaluation is not None:
+            with args.reference_evaluation.resolve().open(encoding="utf-8") as handle:
+                reference = json.load(handle)
+
+        comparison = compare_evaluations(
+            baseline_evaluation,
+            candidate_evaluation,
+            reference,
+            top=args.top,
+        )
+
+    _print_report(comparison)
+    if args.json_output is not None:
+        args.json_output.resolve().write_text(
+            json.dumps(comparison, indent=2) + "\n", encoding="utf-8"
+        )
+
+    failures = evaluate_gates(
+        comparison,
+        min_overall_delta=args.min_overall_delta,
+        max_document_regression=args.max_document_regression,
+        max_missing=args.max_missing,
+        require_reference_lead=args.require_reference_lead,
+    )
+    if failures:
+        print("\nBenchmark gate failed:", file=sys.stderr)
+        for failure in failures:
+            print(f"  - {failure}", file=sys.stderr)
+        return 1
+    print("\nBenchmark gate passed.")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/probe_backend_evidence.py

@@ -0,0 +1,351 @@
+#!/usr/bin/env python3
+"""Compare pdf-inspector evidence with optional MuPDF structured text.
+
+This is an experiment and diagnostic tool, not an extraction fallback. It runs
+MuPDF's deterministic ``stext.json`` backend without OCR and highlights pages
+where that backend exposes materially different text or layout evidence.
+"""
+
+from __future__ import annotations
+
+import argparse
+from collections import Counter
+import json
+from pathlib import Path
+import re
+import shutil
+import subprocess
+import sys
+from typing import Any, Iterable
+
+
+TOKEN_PATTERN = re.compile(r"[^\W_]+(?:[\u2019'][^\W_]+)*", re.UNICODE)
+
+
+def _tokens(texts: Iterable[str]) -> Counter[str]:
+    tokens: Counter[str] = Counter()
+    for text in texts:
+        for token in TOKEN_PATTERN.findall(text.casefold()):
+            # Lone letters are frequently bullets, chart labels, or fragmented
+            # glyphs. Digits remain useful even when they are one character.
+            if len(token) > 1 or token.isdigit():
+                tokens[token] += 1
+    return tokens
+
+
+def _repeated_x_anchors(xs: Iterable[float], *, tolerance: float = 4.0) -> int:
+    buckets = Counter(round(float(x) / tolerance) for x in xs)
+    return sum(count >= 3 for count in buckets.values())
+
+
+def local_pages(payload: dict[str, Any]) -> dict[int, dict[str, Any]]:
+    """Summarize positioned ``pdf2md --items-json`` evidence by page."""
+    pages: dict[int, dict[str, Any]] = {}
+    for item in payload.get("items", []):
+        page_number = int(item["page"])
+        page = pages.setdefault(
+            page_number,
+            {"texts": [], "xs": [], "text_items": 0, "image_items": 0},
+        )
+        if item.get("item_type") == "image":
+            page["image_items"] += 1
+            continue
+        text = str(item.get("text", ""))
+        if text.strip():
+            page["texts"].append(text)
+            page["xs"].append(float(item.get("x", 0.0)))
+            page["text_items"] += 1
+    return pages
+
+
+def alternate_pages(
+    payload: dict[str, Any] | list[dict[str, Any]],
+) -> dict[int, dict[str, Any]]:
+    """Summarize MuPDF ``stext.json`` evidence by page."""
+    pages: dict[int, dict[str, Any]] = {}
+    raw_pages = payload if isinstance(payload, list) else payload.get("pages", [])
+    for index, raw_page in enumerate(raw_pages, start=1):
+        page_number = int(raw_page.get("number", index))
+        page = {
+            "texts": [],
+            "xs": [],
+            "text_blocks": 0,
+            "text_lines": 0,
+            "image_blocks": 0,
+        }
+        for block in raw_page.get("blocks", []):
+            if block.get("type") == "image":
+                page["image_blocks"] += 1
+                continue
+            if block.get("type") != "text":
+                continue
+            page["text_blocks"] += 1
+            for line in block.get("lines", []):
+                text = str(line.get("text", ""))
+                if text.strip():
+                    page["texts"].append(text)
+                    bbox = line.get("bbox", {})
+                    page["xs"].append(float(bbox.get("x", line.get("x", 0.0))))
+                    page["text_lines"] += 1
+        pages[page_number] = page
+    return pages
+
+
+def compare_page(
+    local: dict[str, Any],
+    alternate: dict[str, Any],
+    *,
+    min_token_gain: int,
+    min_alternate_only_ratio: float,
+    min_anchor_gain: int,
+) -> dict[str, Any]:
+    """Compare semantic and coarse layout evidence for one page."""
+    local_tokens = _tokens(local.get("texts", []))
+    alternate_tokens = _tokens(alternate.get("texts", []))
+    shared = local_tokens & alternate_tokens
+    alternate_only = alternate_tokens - local_tokens
+    local_only = local_tokens - alternate_tokens
+    local_total = sum(local_tokens.values())
+    alternate_total = sum(alternate_tokens.values())
+    shared_total = sum(shared.values())
+    alternate_only_total = sum(alternate_only.values())
+    local_only_total = sum(local_only.values())
+    net_token_gain = alternate_total - local_total
+    alternate_only_ratio = alternate_only_total / max(alternate_total, 1)
+
+    local_anchors = _repeated_x_anchors(local.get("xs", []))
+    alternate_anchors = _repeated_x_anchors(alternate.get("xs", []))
+    anchor_gain = alternate_anchors - local_anchors
+    image_gain = int(alternate.get("image_blocks", 0)) - int(
+        local.get("image_items", 0)
+    )
+
+    reasons: list[str] = []
+    if local_total == 0 and alternate_total >= max(5, min_token_gain // 2):
+        reasons.append("local_text_empty")
+    elif (
+        net_token_gain >= min_token_gain
+        and alternate_only_ratio >= min_alternate_only_ratio
+    ):
+        reasons.append("alternate_has_more_text")
+    if anchor_gain >= min_anchor_gain:
+        reasons.append("alternate_has_more_alignment_anchors")
+    if image_gain > 0:
+        reasons.append("alternate_has_more_image_blocks")
+
+    if reasons:
+        classification = "investigate_alternate_evidence"
+    elif local_total - alternate_total >= min_token_gain:
+        classification = "local_has_more_text"
+    elif alternate_only_total + local_only_total:
+        classification = "different_segmentation_or_decoding"
+    else:
+        classification = "equivalent_text_evidence"
+
+    return {
+        "classification": classification,
+        "reasons": reasons,
+        "tokens": {
+            "local": local_total,
+            "alternate": alternate_total,
+            "shared": shared_total,
+            "net_alternate_gain": net_token_gain,
+            "alternate_only": alternate_only_total,
+            "local_only": local_only_total,
+            "alternate_only_ratio": alternate_only_ratio,
+            "alternate_only_sample": sorted(alternate_only)[:12],
+            "local_only_sample": sorted(local_only)[:12],
+        },
+        "layout": {
+            "local_text_items": int(local.get("text_items", 0)),
+            "local_image_items": int(local.get("image_items", 0)),
+            "local_repeated_x_anchors": local_anchors,
+            "alternate_text_blocks": int(alternate.get("text_blocks", 0)),
+            "alternate_text_lines": int(alternate.get("text_lines", 0)),
+            "alternate_image_blocks": int(alternate.get("image_blocks", 0)),
+            "alternate_repeated_x_anchors": alternate_anchors,
+        },
+    }
+
+
+def compare_documents(
+    local_payload: dict[str, Any],
+    alternate_payload: dict[str, Any] | list[dict[str, Any]],
+    *,
+    min_token_gain: int = 20,
+    min_alternate_only_ratio: float = 0.15,
+    min_anchor_gain: int = 2,
+) -> dict[str, Any]:
+    """Return a page-level evidence report for already extracted payloads."""
+    local = local_pages(local_payload)
+    alternate = alternate_pages(alternate_payload)
+    page_numbers = sorted(local.keys() | alternate.keys())
+    pages = []
+    for page_number in page_numbers:
+        result = compare_page(
+            local.get(page_number, {}),
+            alternate.get(page_number, {}),
+            min_token_gain=min_token_gain,
+            min_alternate_only_ratio=min_alternate_only_ratio,
+            min_anchor_gain=min_anchor_gain,
+        )
+        result["page"] = page_number
+        pages.append(result)
+
+    flagged = [
+        page
+        for page in pages
+        if page["classification"] == "investigate_alternate_evidence"
+    ]
+    return {
+        "summary": {
+            "pages": len(pages),
+            "flagged_pages": len(flagged),
+            "flagged_page_numbers": [page["page"] for page in flagged],
+            "local_tokens": sum(page["tokens"]["local"] for page in pages),
+            "alternate_tokens": sum(page["tokens"]["alternate"] for page in pages),
+            "alternate_only_tokens": sum(
+                page["tokens"]["alternate_only"] for page in pages
+            ),
+        },
+        "pages": pages,
+    }
+
+
+def _json_command(command: list[str]) -> Any:
+    try:
+        completed = subprocess.run(
+            command,
+            check=True,
+            stdout=subprocess.PIPE,
+            stderr=subprocess.PIPE,
+            text=True,
+        )
+    except subprocess.CalledProcessError as error:
+        detail = error.stderr.strip() or error.stdout.strip() or "no diagnostic output"
+        raise RuntimeError(f"command failed: {' '.join(command)}\n{detail}") from error
+    try:
+        return json.loads(completed.stdout)
+    except json.JSONDecodeError as error:
+        raise RuntimeError(
+            f"command did not return JSON: {' '.join(command)}: {error}"
+        ) from error
+
+
+def probe_pdf(
+    pdf: Path,
+    *,
+    pdf2md: Path,
+    mutool: Path,
+    min_token_gain: int,
+    min_alternate_only_ratio: float,
+    min_anchor_gain: int,
+) -> dict[str, Any]:
+    local_payload = _json_command([str(pdf2md), str(pdf), "--items-json"])
+    # `stext.json` is MuPDF's native structured text output. The OCR formats
+    # are intentionally not used so this remains a deterministic no-model
+    # comparison.
+    alternate_payload = _json_command(
+        [str(mutool), "draw", "-q", "-F", "stext.json", "-o", "-", str(pdf)]
+    )
+    report = compare_documents(
+        local_payload,
+        alternate_payload,
+        min_token_gain=min_token_gain,
+        min_alternate_only_ratio=min_alternate_only_ratio,
+        min_anchor_gain=min_anchor_gain,
+    )
+    report["pdf"] = str(pdf)
+    return report
+
+
+def _arguments(argv: list[str] | None = None) -> argparse.Namespace:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("pdf", type=Path, nargs="+")
+    parser.add_argument("--pdf2md", type=Path, default=Path("target/release/pdf2md"))
+    parser.add_argument("--mutool", type=Path)
+    parser.add_argument("--json-output", type=Path)
+    parser.add_argument("--min-token-gain", type=int, default=20)
+    parser.add_argument("--min-alternate-only-ratio", type=float, default=0.15)
+    parser.add_argument("--min-anchor-gain", type=int, default=2)
+    return parser.parse_args(argv)
+
+
+def _print_report(result: dict[str, Any]) -> None:
+    summary = result["summary"]
+    print(f"\n{result['pdf']}")
+    print(
+        f"  {summary['flagged_pages']}/{summary['pages']} pages flagged; "
+        f"tokens local={summary['local_tokens']} alternate={summary['alternate_tokens']} "
+        f"alternate-only={summary['alternate_only_tokens']}"
+    )
+    for page in result["pages"]:
+        if page["classification"] != "investigate_alternate_evidence":
+            continue
+        reasons = ", ".join(page["reasons"])
+        tokens = page["tokens"]
+        print(
+            f"  page {page['page']}: {reasons}; "
+            f"net tokens={tokens['net_alternate_gain']:+d}, "
+            f"alternate-only={tokens['alternate_only']}"
+        )
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = _arguments(argv)
+    pdf2md = args.pdf2md.absolute()
+    mutool = args.mutool or (Path(found) if (found := shutil.which("mutool")) else None)
+    if not pdf2md.is_file():
+        print(f"error: pdf2md binary not found: {pdf2md}", file=sys.stderr)
+        return 2
+    if mutool is None or not mutool.is_file():
+        print("error: mutool not found; install MuPDF or pass --mutool", file=sys.stderr)
+        return 2
+    if (
+        args.min_token_gain < 0
+        or args.min_anchor_gain < 0
+        or not 0.0 <= args.min_alternate_only_ratio <= 1.0
+    ):
+        print("error: thresholds must be non-negative and ratio must be in [0, 1]", file=sys.stderr)
+        return 2
+
+    results = []
+    for pdf in args.pdf:
+        path = pdf.absolute()
+        if not path.is_file():
+            print(f"error: PDF not found: {path}", file=sys.stderr)
+            return 2
+        try:
+            result = probe_pdf(
+                path,
+                pdf2md=pdf2md,
+                mutool=mutool,
+                min_token_gain=args.min_token_gain,
+                min_alternate_only_ratio=args.min_alternate_only_ratio,
+                min_anchor_gain=args.min_anchor_gain,
+            )
+        except RuntimeError as error:
+            print(f"error: {error}", file=sys.stderr)
+            return 1
+        results.append(result)
+        _print_report(result)
+
+    payload = {
+        "schema_version": 1,
+        "experiment": "optional_mupdf_stext_evidence",
+        "ocr": False,
+        "thresholds": {
+            "min_token_gain": args.min_token_gain,
+            "min_alternate_only_ratio": args.min_alternate_only_ratio,
+            "min_anchor_gain": args.min_anchor_gain,
+        },
+        "documents": results,
+    }
+    if args.json_output:
+        args.json_output.parent.mkdir(parents=True, exist_ok=True)
+        args.json_output.write_text(json.dumps(payload, indent=2) + "\n", encoding="utf-8")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

scripts/tests/test_bench_opendataloader.py

@@ -0,0 +1,203 @@
+import io
+import json
+import sys
+import tempfile
+import unittest
+from contextlib import redirect_stderr, redirect_stdout
+from pathlib import Path
+from unittest.mock import patch
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from bench_opendataloader import (
+    _arguments,
+    _print_report,
+    _run_engine,
+    compare_evaluations,
+    evaluate_gates,
+)
+
+
+def evaluation(overall, documents, *, missing=0):
+    return {
+        "metrics": {
+            "score": {
+                "overall_mean": overall,
+                "nid_mean": overall + 0.01,
+            },
+            "missing_predictions": missing,
+        },
+        "documents": [
+            {
+                "document_id": document_id,
+                "scores": {"overall": score},
+            }
+            for document_id, score in documents.items()
+        ],
+    }
+
+
+class ComparisonTests(unittest.TestCase):
+    def test_reports_metric_and_document_deltas(self):
+        baseline = evaluation(0.80, {"a": 0.8, "b": 0.6, "c": 0.7})
+        candidate = evaluation(0.82, {"a": 0.9, "b": 0.5, "c": 0.7})
+
+        result = compare_evaluations(baseline, candidate, top=1)
+
+        self.assertAlmostEqual(result["deltas"]["overall_mean"], 0.02)
+        self.assertEqual(result["documents"]["improved"], 1)
+        self.assertEqual(result["documents"]["regressed"], 1)
+        self.assertEqual(result["documents"]["unchanged"], 1)
+        self.assertEqual(
+            result["documents"]["largest_improvements"][0]["document_id"], "a"
+        )
+        self.assertEqual(
+            result["documents"]["largest_regressions"][0]["document_id"], "b"
+        )
+
+    def test_reference_delta_is_reported(self):
+        baseline = evaluation(0.80, {})
+        candidate = evaluation(0.82, {})
+        reference = evaluation(0.81, {})
+
+        result = compare_evaluations(baseline, candidate, reference)
+
+        self.assertAlmostEqual(
+            result["candidate_vs_reference"]["overall_mean"], 0.01
+        )
+
+    def test_gates_cover_aggregate_document_missing_and_reference(self):
+        comparison = compare_evaluations(
+            evaluation(0.80, {"a": 0.8}),
+            evaluation(0.79, {"a": 0.7}, missing=1),
+            evaluation(0.81, {}),
+        )
+
+        failures = evaluate_gates(
+            comparison,
+            min_overall_delta=0.0,
+            max_document_regression=0.05,
+            max_missing=0,
+            require_reference_lead=True,
+        )
+
+        self.assertEqual(len(failures), 4)
+
+    def test_regression_gate_is_independent_of_report_limit(self):
+        comparison = compare_evaluations(
+            evaluation(0.80, {"a": 0.8}),
+            evaluation(0.80, {"a": 0.7}),
+            top=0,
+        )
+
+        failures = evaluate_gates(
+            comparison,
+            min_overall_delta=0.0,
+            max_document_regression=0.05,
+            max_missing=0,
+            require_reference_lead=False,
+        )
+
+        self.assertEqual(len(failures), 1)
+        self.assertIn("largest document regression", failures[0])
+
+    def test_report_handles_reference_without_overall_score(self):
+        result = compare_evaluations(
+            evaluation(0.80, {}),
+            evaluation(0.82, {}),
+            {"metrics": {"score": {"nid_mean": 0.81}}},
+        )
+
+        output = io.StringIO()
+        with redirect_stdout(output):
+            _print_report(result)
+
+        self.assertIn("Reference overall: n/a; candidate delta: n/a", output.getvalue())
+
+    def test_reference_gate_reports_missing_score_as_unavailable(self):
+        comparison = compare_evaluations(
+            evaluation(0.80, {}),
+            evaluation(0.82, {}),
+        )
+
+        failures = evaluate_gates(
+            comparison,
+            min_overall_delta=0.0,
+            max_document_regression=None,
+            max_missing=0,
+            require_reference_lead=True,
+        )
+
+        self.assertEqual(failures, ["reference overall score is unavailable"])
+
+    def test_arguments_reject_negative_counts_and_allow_zero_top(self):
+        required = [
+            "--bench-dir",
+            ".",
+            "--baseline",
+            "baseline",
+            "--candidate",
+            "candidate",
+        ]
+        self.assertEqual(_arguments(required + ["--top", "0"]).top, 0)
+        for option in ("--top", "--max-document-regression", "--max-missing"):
+            with self.subTest(option=option), redirect_stderr(io.StringIO()):
+                with self.assertRaises(SystemExit):
+                    _arguments(required + [option, "-1"])
+
+    def test_arguments_reject_nonfinite_float_thresholds(self):
+        required = [
+            "--bench-dir",
+            ".",
+            "--baseline",
+            "baseline",
+            "--candidate",
+            "candidate",
+        ]
+        for option in ("--min-overall-delta", "--max-document-regression"):
+            for value in ("nan", "inf", "-inf"):
+                with self.subTest(option=option, value=value), redirect_stderr(
+                    io.StringIO()
+                ):
+                    with self.assertRaises(SystemExit):
+                        _arguments(required + [option, value])
+
+    def test_run_engine_clears_stale_predictions_before_parser(self):
+        with tempfile.TemporaryDirectory() as temporary:
+            root = Path(temporary)
+            bench_dir = root / "bench"
+            source = bench_dir / "prediction" / "pdf-inspector"
+            source.mkdir(parents=True)
+            (source / "stale.md").write_text("stale", encoding="utf-8")
+            scratch = root / "scratch"
+            scratch.mkdir()
+
+            def fake_run(command, *, cwd, env=None):
+                if any(part.endswith("pdf_parser.py") for part in command):
+                    self.assertFalse(source.exists())
+                    (source / "markdown").mkdir(parents=True)
+                    (source / "markdown" / "new.md").write_text(
+                        "new", encoding="utf-8"
+                    )
+                else:
+                    destination = scratch / "candidate"
+                    (destination / "evaluation.json").write_text(
+                        json.dumps(evaluation(0.82, {})), encoding="utf-8"
+                    )
+
+            with patch("bench_opendataloader._run", side_effect=fake_run):
+                result = _run_engine(
+                    bench_dir=bench_dir,
+                    python=Path("python"),
+                    binary=Path("pdf2md"),
+                    label="candidate",
+                    scratch_root=scratch,
+                )
+
+            self.assertEqual(result["metrics"]["score"]["overall_mean"], 0.82)
+            self.assertFalse((source / "stale.md").exists())
+            self.assertFalse((scratch / "candidate" / "stale.md").exists())
+
+
+if __name__ == "__main__":
+    unittest.main()

scripts/tests/test_probe_backend_evidence.py

@@ -0,0 +1,103 @@
+import sys
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from probe_backend_evidence import compare_documents
+
+
+def local_payload(items):
+    return {"items": items}
+
+
+def item(page, text, x=10, item_type="text"):
+    return {"page": page, "text": text, "x": x, "item_type": item_type}
+
+
+def alternate_payload(pages):
+    return {"pages": pages}
+
+
+def page(lines, *, images=0):
+    blocks = [
+        {
+            "type": "text",
+            "lines": [
+                {"text": text, "bbox": {"x": x, "y": index * 10, "w": 80, "h": 8}}
+                for index, (text, x) in enumerate(lines)
+            ],
+        }
+    ]
+    blocks.extend({"type": "image"} for _ in range(images))
+    return {"blocks": blocks}
+
+
+class EvidenceComparisonTests(unittest.TestCase):
+    def test_accepts_real_top_level_page_array(self):
+        local = local_payload([item(1, "alpha beta")])
+        alternate = [page([("alpha beta gamma", 10)])]
+
+        result = compare_documents(local, alternate)["pages"][0]
+
+        self.assertEqual(result["tokens"]["alternate"], 3)
+        self.assertEqual(result["tokens"]["net_alternate_gain"], 1)
+
+    def test_flags_material_alternate_text_gain(self):
+        local = local_payload([item(1, "alpha beta")])
+        alternate = alternate_payload(
+            [page([("alpha beta gamma delta epsilon zeta", 10)])]
+        )
+
+        report = compare_documents(
+            local,
+            alternate,
+            min_token_gain=3,
+            min_alternate_only_ratio=0.2,
+        )
+
+        result = report["pages"][0]
+        self.assertEqual(result["classification"], "investigate_alternate_evidence")
+        self.assertIn("alternate_has_more_text", result["reasons"])
+        self.assertEqual(result["tokens"]["net_alternate_gain"], 4)
+
+    def test_repeated_alignment_and_image_evidence_are_reported(self):
+        local = local_payload([item(1, "one two", 10)])
+        alternate = alternate_payload(
+            [
+                page(
+                    [
+                        ("one two", 10),
+                        ("row three", 100),
+                        ("row four", 100),
+                        ("row five", 100),
+                    ],
+                    images=1,
+                )
+            ]
+        )
+
+        result = compare_documents(
+            local,
+            alternate,
+            min_token_gain=99,
+            min_anchor_gain=1,
+        )["pages"][0]
+
+        self.assertIn("alternate_has_more_alignment_anchors", result["reasons"])
+        self.assertIn("alternate_has_more_image_blocks", result["reasons"])
+        self.assertEqual(result["layout"]["alternate_repeated_x_anchors"], 1)
+
+    def test_token_segmentation_difference_does_not_imply_more_evidence(self):
+        local = local_payload([item(1, "Revenue 2025")])
+        alternate = alternate_payload([page([("Revenue 2024", 10)])])
+
+        result = compare_documents(local, alternate, min_token_gain=2)["pages"][0]
+
+        self.assertEqual(result["classification"], "different_segmentation_or_decoding")
+        self.assertEqual(result["reasons"], [])
+        self.assertEqual(result["tokens"]["alternate_only_sample"], ["2024"])
+
+
+if __name__ == "__main__":
+    unittest.main()

scripts/tests/test_version.py

@@ -0,0 +1,111 @@
+import json
+import sys
+import tempfile
+import unittest
+from pathlib import Path
+
+sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
+
+from version import PLATFORM_PACKAGES, check_versions, set_versions
+
+
+class VersionTests(unittest.TestCase):
+    def setUp(self):
+        self.temporary = tempfile.TemporaryDirectory()
+        self.root = Path(self.temporary.name)
+        (self.root / "napi").mkdir()
+        (self.root / "site").mkdir()
+        (self.root / "wasm").mkdir()
+
+        self._write_manifest("Cargo.toml", "package", "0.1.0")
+        self._write_manifest("pyproject.toml", "project", "0.1.0")
+        self._write_manifest("napi/Cargo.toml", "package", "0.1.0")
+        self._write_manifest("wasm/Cargo.toml", "package", "0.1.0")
+
+        package = {
+            "name": "@firecrawl/pdf-inspector",
+            "version": "0.1.0",
+            "optionalDependencies": {
+                dependency: "0.1.0" for dependency in PLATFORM_PACKAGES
+            },
+        }
+        (self.root / "napi/package.json").write_text(
+            json.dumps(package), encoding="utf-8"
+        )
+        (self.root / "napi/bun.lock").write_text(
+            "\n".join(
+                f'        "{dependency}": "0.1.0",'
+                for dependency in PLATFORM_PACKAGES
+            )
+            + "\n",
+            encoding="utf-8",
+        )
+        (self.root / "site/index.html").write_text(
+            'https://cdn.jsdelivr.net/npm/@firecrawl/pdf-inspector-wasm@0.1.0/'
+            'pdf_inspector_wasm.js\n',
+            encoding="utf-8",
+        )
+        self._write_lock(
+            "napi/Cargo.lock", ("pdf-inspector", "pdf-inspector-napi")
+        )
+        self._write_lock(
+            "wasm/Cargo.lock", ("pdf-inspector", "pdf-inspector-wasm")
+        )
+
+    def tearDown(self):
+        self.temporary.cleanup()
+
+    def _write_manifest(self, relative, section, version):
+        (self.root / relative).write_text(
+            f'[{section}]\nname = "fixture"\nversion = "{version}"\n',
+            encoding="utf-8",
+        )
+
+    def _write_lock(self, relative, packages):
+        content = "\n".join(
+            f'[[package]]\nname = "{package}"\nversion = "0.1.0"\n'
+            for package in packages
+        )
+        (self.root / relative).write_text(content, encoding="utf-8")
+
+    def test_updates_every_version_location(self):
+        set_versions("1.14.0", self.root)
+
+        self.assertEqual(check_versions(self.root), "1.14.0")
+
+    def test_reports_a_divergent_package(self):
+        self._write_manifest("wasm/Cargo.toml", "package", "0.2.0")
+
+        with self.assertRaisesRegex(ValueError, "WASM package: 0.2.0"):
+            check_versions(self.root)
+
+    def test_rejects_an_invalid_version(self):
+        with self.assertRaisesRegex(ValueError, "Invalid semantic version"):
+            set_versions("next", self.root)
+
+    def test_rejects_numeric_prerelease_with_leading_zero(self):
+        before = (self.root / "Cargo.toml").read_text(encoding="utf-8")
+
+        with self.assertRaisesRegex(ValueError, "Invalid semantic version"):
+            set_versions("1.2.3-01", self.root)
+
+        self.assertEqual(
+            (self.root / "Cargo.toml").read_text(encoding="utf-8"), before
+        )
+
+    def test_preflight_failure_does_not_partially_update(self):
+        before = (self.root / "Cargo.toml").read_text(encoding="utf-8")
+        (self.root / "site/index.html").write_text(
+            "missing module URL\n", encoding="utf-8"
+        )
+
+        with self.assertRaisesRegex(ValueError, "Missing pinned WASM package URL"):
+            set_versions("1.14.0", self.root)
+
+        self.assertEqual(
+            (self.root / "Cargo.toml").read_text(encoding="utf-8"), before
+        )
+
+
+if __name__ == "__main__":
+    unittest.main()

scripts/version.py

@@ -0,0 +1,262 @@
+#!/usr/bin/env python3
+"""Keep every pdf-inspector package on one release version."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import re
+from pathlib import Path
+
+
+ROOT = Path(__file__).resolve().parents[1]
+PRERELEASE_IDENTIFIER = (
+    r"(?:0|[1-9]\d*|[0-9A-Za-z-]*[A-Za-z-][0-9A-Za-z-]*)"
+)
+SEMVER = re.compile(
+    r"^(0|[1-9]\d*)\.(0|[1-9]\d*)\.(0|[1-9]\d*)"
+    rf"(?:-{PRERELEASE_IDENTIFIER}(?:\.{PRERELEASE_IDENTIFIER})*)?"
+    r"(?:\+[0-9A-Za-z-]+(?:\.[0-9A-Za-z-]+)*)?$"
+)
+VERSION_LINE = re.compile(r'^(\s*version\s*=\s*")[^"]+(".*)$')
+SECTION_LINE = re.compile(r"^\s*\[([^]]+)]\s*$")
+PLATFORM_PACKAGES = (
+    "@firecrawl/pdf-inspector-linux-x64-gnu",
+    "@firecrawl/pdf-inspector-linux-x64-musl",
+    "@firecrawl/pdf-inspector-linux-arm64-gnu",
+    "@firecrawl/pdf-inspector-linux-arm64-musl",
+    "@firecrawl/pdf-inspector-darwin-arm64",
+    "@firecrawl/pdf-inspector-win32-x64-msvc",
+)
+TOML_VERSIONS = (
+    ("Rust crate", Path("Cargo.toml"), "package"),
+    ("Python package", Path("pyproject.toml"), "project"),
+    ("NAPI crate", Path("napi/Cargo.toml"), "package"),
+    ("WASM package", Path("wasm/Cargo.toml"), "package"),
+)
+LOCK_VERSIONS = (
+    ("NAPI lock: core", Path("napi/Cargo.lock"), "pdf-inspector"),
+    ("NAPI lock: binding", Path("napi/Cargo.lock"), "pdf-inspector-napi"),
+    ("WASM lock: core", Path("wasm/Cargo.lock"), "pdf-inspector"),
+    ("WASM lock: binding", Path("wasm/Cargo.lock"), "pdf-inspector-wasm"),
+)
+SITE_WASM_VERSION = re.compile(
+    r"(@firecrawl/pdf-inspector-wasm@)([^/\"]+)(/pdf_inspector_wasm\.js)"
+)
+
+
+def _read_section_version(path: Path, section: str) -> str:
+    active = False
+    for line in path.read_text(encoding="utf-8").splitlines():
+        section_match = SECTION_LINE.match(line)
+        if section_match:
+            active = section_match.group(1) == section
+        elif active:
+            version_match = VERSION_LINE.match(line)
+            if version_match:
+                return line.split('"', 2)[1]
+    raise ValueError(f"No version found in [{section}] of {path}")
+
+
+def _write_section_version(path: Path, section: str, version: str) -> None:
+    lines = path.read_text(encoding="utf-8").splitlines(keepends=True)
+    active = False
+    for index, line in enumerate(lines):
+        section_match = SECTION_LINE.match(line)
+        if section_match:
+            active = section_match.group(1) == section
+        elif active:
+            version_match = VERSION_LINE.match(line)
+            if version_match:
+                newline = "\n" if line.endswith("\n") else ""
+                replacement = (
+                    f"{version_match.group(1)}{version}"
+                    f"{version_match.group(2).rstrip()}"
+                )
+                lines[index] = (
+                    f"{replacement}{newline}"
+                )
+                path.write_text("".join(lines), encoding="utf-8")
+                return
+    raise ValueError(f"No version found in [{section}] of {path}")
+
+
+def _package_block(lines: list[str], package: str) -> tuple[int, int]:
+    for start, line in enumerate(lines):
+        if line.strip() != "[[package]]":
+            continue
+        end = next(
+            (
+                index
+                for index in range(start + 1, len(lines))
+                if lines[index].strip() == "[[package]]"
+            ),
+            len(lines),
+        )
+        if any(line.strip() == f'name = "{package}"' for line in lines[start:end]):
+            return start, end
+    raise ValueError(f"No lockfile entry found for {package}")
+
+
+def _read_lock_version(path: Path, package: str) -> str:
+    lines = path.read_text(encoding="utf-8").splitlines()
+    start, end = _package_block(lines, package)
+    for line in lines[start:end]:
+        version_match = VERSION_LINE.match(line)
+        if version_match:
+            return line.split('"', 2)[1]
+    raise ValueError(f"No version found for {package} in {path}")
+
+
+def _write_lock_version(path: Path, package: str, version: str) -> None:
+    lines = path.read_text(encoding="utf-8").splitlines(keepends=True)
+    start, end = _package_block(lines, package)
+    for index in range(start, end):
+        version_match = VERSION_LINE.match(lines[index])
+        if version_match:
+            newline = "\n" if lines[index].endswith("\n") else ""
+            lines[index] = (
+                f'{version_match.group(1)}{version}{version_match.group(2).rstrip()}'
+                f"{newline}"
+            )
+            path.write_text("".join(lines), encoding="utf-8")
+            return
+    raise ValueError(f"No version found for {package} in {path}")
+
+
+def _node_versions(root: Path) -> dict[str, str]:
+    package = json.loads((root / "napi/package.json").read_text(encoding="utf-8"))
+    versions = {"Node package": package["version"]}
+    optional = package.get("optionalDependencies", {})
+    for dependency in PLATFORM_PACKAGES:
+        if dependency not in optional:
+            raise ValueError(f"Missing Node optional dependency: {dependency}")
+        versions[f"Node optional dependency: {dependency}"] = optional[dependency]
+    return versions
+
+
+def _bun_versions(root: Path) -> dict[str, str]:
+    text = (root / "napi/bun.lock").read_text(encoding="utf-8")
+    versions = {}
+    for dependency in PLATFORM_PACKAGES:
+        match = re.search(
+            rf'"{re.escape(dependency)}": "([^"]+)"[,]', text
+        )
+        if not match:
+            raise ValueError(f"Missing Bun lock dependency: {dependency}")
+        versions[f"Bun lock: {dependency}"] = match.group(1)
+    return versions
+
+
+def _site_wasm_version(root: Path) -> str:
+    text = (root / "site/index.html").read_text(encoding="utf-8")
+    match = SITE_WASM_VERSION.search(text)
+    if not match:
+        raise ValueError("Missing pinned WASM package URL in site/index.html")
+    return match.group(2)
+
+
+def package_versions(root: Path = ROOT) -> dict[str, str]:
+    versions = {
+        label: _read_section_version(root / relative, section)
+        for label, relative, section in TOML_VERSIONS
+    }
+    versions.update(_node_versions(root))
+    versions.update(_bun_versions(root))
+    versions["Website WASM module"] = _site_wasm_version(root)
+    versions.update(
+        {
+            label: _read_lock_version(root / relative, package)
+            for label, relative, package in LOCK_VERSIONS
+        }
+    )
+    return versions
+
+
+def check_versions(root: Path = ROOT) -> str:
+    versions = package_versions(root)
+    expected = versions["Rust crate"]
+    if not SEMVER.fullmatch(expected):
+        raise ValueError(f"Rust crate has an invalid semantic version: {expected}")
+    mismatches = {
+        label: version for label, version in versions.items() if version != expected
+    }
+    if mismatches:
+        details = "\n".join(
+            f"  - {label}: {version}" for label, version in mismatches.items()
+        )
+        raise ValueError(f"Expected every package to use {expected}:\n{details}")
+    return expected
+
+
+def set_versions(version: str, root: Path = ROOT) -> None:
+    if not SEMVER.fullmatch(version):
+        raise ValueError(f"Invalid semantic version: {version}")
+
+    # Validate every expected location before writing the first file. This
+    # prevents a stale manifest or generated file from leaving a partial bump.
+    package_versions(root)
+
+    for _, relative, section in TOML_VERSIONS:
+        _write_section_version(root / relative, section, version)
+
+    package_path = root / "napi/package.json"
+    package = json.loads(package_path.read_text(encoding="utf-8"))
+    package["version"] = version
+    optional = package.get("optionalDependencies", {})
+    for dependency in PLATFORM_PACKAGES:
+        if dependency not in optional:
+            raise ValueError(f"Missing Node optional dependency: {dependency}")
+        optional[dependency] = version
+    package_path.write_text(json.dumps(package, indent=2) + "\n", encoding="utf-8")
+
+    bun_path = root / "napi/bun.lock"
+    bun_text = bun_path.read_text(encoding="utf-8")
+    for dependency in PLATFORM_PACKAGES:
+        pattern = rf'("{re.escape(dependency)}": ")[^"]+("[,])'
+        bun_text, count = re.subn(
+            pattern, rf"\g<1>{version}\g<2>", bun_text, count=1
+        )
+        if count != 1:
+            raise ValueError(f"Missing Bun lock dependency: {dependency}")
+    bun_path.write_text(bun_text, encoding="utf-8")
+
+    site_path = root / "site/index.html"
+    site_text = site_path.read_text(encoding="utf-8")
+    site_text, count = SITE_WASM_VERSION.subn(
+        rf"\g<1>{version}\g<3>", site_text, count=1
+    )
+    if count != 1:
+        raise ValueError("Missing pinned WASM package URL in site/index.html")
+    site_path.write_text(site_text, encoding="utf-8")
+
+    for _, relative, package_name in LOCK_VERSIONS:
+        _write_lock_version(root / relative, package_name, version)
+
+    check_versions(root)
+
+
+def main() -> int:
+    parser = argparse.ArgumentParser(description=__doc__)
+    parser.add_argument("version", nargs="?", help="new shared semantic version")
+    parser.add_argument(
+        "--check", action="store_true", help="fail if package versions have diverged"
+    )
+    arguments = parser.parse_args()
+    if arguments.check == bool(arguments.version):
+        parser.error("provide either a version or --check")
+
+    try:
+        if arguments.check:
+            version = check_versions()
+            print(f"All packages use {version}")
+        else:
+            set_versions(arguments.version)
+            print(f"Updated all packages to {arguments.version}")
+    except ValueError as error:
+        parser.exit(1, f"{error}\n")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

site/assets/firecrawl-mark.svg

@@ -0,0 +1,3 @@
+<svg width="200" height="284" viewBox="0 0 200 284" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path d="M166.862 90.7716C155.812 94.0514 147.483 101.471 141.383 109.53C140.073 111.26 137.343 109.96 137.863 107.841C149.543 59.8136 134.113 19.896 86.0157 0.247269C83.5758 -0.752669 81.0359 1.43719 81.6759 3.99704C103.555 91.8416 11.5294 84.432 23.1588 184.016C23.3588 185.726 21.4389 186.896 20.039 185.896C15.6792 182.766 10.8095 176.236 7.46963 171.647C6.48968 170.297 4.36978 170.677 3.9198 172.287C1.25994 181.906 0 190.965 0 199.965C0 234.963 17.9891 265.771 45.2177 283.63C46.7777 284.65 48.7776 283.19 48.2476 281.4C46.8477 276.7 46.0577 271.74 45.9977 266.611C45.9977 263.461 46.1977 260.241 46.6877 257.241C47.8276 249.702 50.4475 242.522 54.8473 235.983C69.9365 213.334 100.185 191.455 95.3552 161.747C95.0453 159.867 97.2651 158.627 98.6651 159.917C119.974 179.386 124.194 205.575 120.694 229.063C120.394 231.103 122.954 232.193 124.244 230.593C127.504 226.513 131.483 222.933 135.813 220.244C136.893 219.574 138.333 220.084 138.743 221.284C141.153 228.293 144.733 234.873 148.113 241.452C152.152 249.362 154.302 258.391 153.962 267.951C153.792 272.6 153.022 277.1 151.732 281.38C151.182 283.19 153.162 284.7 154.752 283.66C182.001 265.801 200 234.993 200 199.975C200 187.806 197.87 175.876 193.84 164.697C185.391 141.248 163.952 123.64 169.372 93.0815C169.632 91.6216 168.282 90.3517 166.862 90.7716Z" fill="#FA5D19" style="fill:#FA5D19;fill:color(display-p3 0.9816 0.3634 0.0984);fill-opacity:1;"/>
+</svg>

site/assets/firecrawl-wordmark.svg

@@ -0,0 +1,12 @@
+<svg width="172" height="40" viewBox="0 0 172 40" fill="none" xmlns="http://www.w3.org/2000/svg">
+<path d="M23.3606 12.8281C21.8137 13.2873 20.6476 14.3261 19.7936 15.4544C19.6102 15.6966 19.228 15.5146 19.3008 15.2178C20.936 8.49401 18.7759 2.90556 12.0422 0.154735C11.7006 0.0147436 11.345 0.321324 11.4346 0.679702C14.4977 12.9779 1.61412 11.9406 3.24224 25.8823C3.27024 26.1217 3.00145 26.2855 2.80546 26.1455C2.19509 25.7073 1.51332 24.7932 1.04575 24.1506C0.908555 23.9616 0.611769 24.0148 0.548773 24.2402C0.176391 25.5869 0 26.8553 0 28.1152C0 33.0149 2.51847 37.328 6.33048 39.8283C6.54887 39.9711 6.82886 39.7667 6.75466 39.5161C6.55867 38.8581 6.44808 38.1638 6.43968 37.4456C6.43968 37.0046 6.46768 36.5539 6.53627 36.1339C6.69587 35.0784 7.06265 34.0732 7.67862 33.1577C9.79111 29.9869 14.0259 26.9239 13.3497 22.7647C13.3063 22.5015 13.6171 22.328 13.8131 22.5085C16.7964 25.2342 17.3871 28.9005 16.8972 32.1889C16.8552 32.4745 17.2135 32.6271 17.3941 32.4031C17.8505 31.832 18.4077 31.3308 19.0138 30.9542C19.165 30.8604 19.3666 30.9318 19.424 31.0998C19.7614 32.0811 20.2626 33.0023 20.7358 33.9234C21.3013 35.0308 21.6023 36.2949 21.5547 37.6332C21.5309 38.2842 21.4231 38.9141 21.2425 39.5133C21.1655 39.7667 21.4427 39.9781 21.6653 39.8325C25.4801 37.3322 28 33.0191 28 28.1166C28 26.4129 27.7018 24.7428 27.1376 23.1777C25.9547 19.8949 22.9533 17.4297 23.712 13.1515C23.7484 12.9471 23.5594 12.7693 23.3606 12.8281Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M41 34.0521V10.9618H55.7586V14.3264H44.7969V21.0226H53.8436V24.2882H44.7969V34.0521H41Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M59.9569 14.7882C58.7352 14.7882 57.7777 13.8976 57.7777 12.6441C57.7777 11.3906 58.7352 10.5 59.9569 10.5C61.1785 10.5 62.136 11.3906 62.136 12.6441C62.136 13.8976 61.1785 14.7882 59.9569 14.7882ZM58.1409 34.0521V17.1632H61.7068V34.0521H58.1409Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M73.5885 17.1632H74.3809V20.4948H72.796C69.6264 20.4948 68.6029 22.9687 68.6029 25.5747V34.0521H65.0371V17.1632H68.2067L68.6029 19.7031C69.4613 18.2847 70.815 17.1632 73.5885 17.1632Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M83.632 34.25C78.3163 34.25 74.9816 30.8194 74.9816 25.6406C74.9816 20.4288 78.3163 16.9653 83.3019 16.9653C88.1884 16.9653 91.457 20.066 91.5561 25.0139C91.5561 25.4427 91.5231 25.9045 91.457 26.3663H78.7125V26.5972C78.8116 29.467 80.6275 31.3472 83.4339 31.3472C85.613 31.3472 87.1979 30.2587 87.6931 28.3785H91.2589C90.6646 31.7101 87.8252 34.25 83.632 34.25ZM78.8446 23.7604H87.8582C87.561 21.2535 85.8112 19.8351 83.3349 19.8351C81.0567 19.8351 79.1087 21.3524 78.8446 23.7604Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M102.033 34.25C96.9151 34.25 93.6465 30.9184 93.6465 25.6406C93.6465 20.4288 97.0142 16.9653 102.132 16.9653C106.49 16.9653 109.197 19.3733 109.891 23.1997H106.16C105.698 21.2205 104.278 20 102.066 20C99.1933 20 97.3113 22.309 97.3113 25.6406C97.3113 28.9392 99.1933 31.2153 102.066 31.2153C104.245 31.2153 105.698 29.9618 106.127 28.0156H109.891C109.23 31.842 106.358 34.25 102.033 34.25Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M121.006 17.1632H121.799V20.4948H120.214C117.044 20.4948 116.021 22.9687 116.021 25.5747V34.0521H112.455V17.1632H115.625L116.021 19.7031C116.879 18.2847 118.233 17.1632 121.006 17.1632Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M130.614 16.9653C135.104 16.9653 137.679 19.1094 137.679 23.1007V34.0521H134.576L134.279 31.6441C133.123 33.1615 131.505 34.25 128.831 34.25C125.133 34.25 122.657 32.4358 122.657 29.3021C122.657 25.8385 125.166 23.8924 129.92 23.8924H134.147V22.8698C134.147 20.9896 132.793 19.8351 130.449 19.8351C128.336 19.8351 126.916 20.8247 126.652 22.309H123.152C123.515 19.0104 126.355 16.9653 130.614 16.9653ZM129.425 31.4792C132.397 31.4792 134.114 29.7309 134.147 27.125V26.5312H129.722C127.51 26.5312 126.289 27.3559 126.289 29.0712C126.289 30.4896 127.477 31.4792 129.425 31.4792Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M144.653 34.0521L139.139 17.1632H142.903L146.766 30.0937L150.629 17.1632H153.897L157.595 30.0937L161.59 17.1632H165.222L159.609 34.0521H155.779L152.214 22.5729L148.516 34.0521H144.653Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+<path d="M166.934 34.0521V10.9618H170.5V34.0521H166.934Z" fill="#262626" style="fill:#262626;fill:color(display-p3 0.1500 0.1500 0.1500);fill-opacity:1;"/>
+</svg>

src/bin/detect_pdf.rs

@@ -11,6 +11,37 @@ use std::process;
 use std::time::Instant;
 
 /// Escape a string for embedding in a JSON string value.
+fn format_detector_ocr_reasons(reasons: &std::collections::BTreeMap<u32, Vec<String>>) -> String {
+    reasons
+        .iter()
+        .map(|(page, page_reasons)| {
+            let reasons_json = page_reasons
+                .iter()
+                .map(|reason| format!(r#""{}""#, json_escape(reason)))
+                .collect::<Vec<_>>()
+                .join(",");
+            format!(r#"{{"page":{},"reasons":[{}]}}"#, page, reasons_json)
+        })
+        .collect::<Vec<_>>()
+        .join(",")
+}
+
+fn format_ocr_reasons_by_page(reasons: &[pdf_inspector::PageOcrReasons]) -> String {
+    reasons
+        .iter()
+        .map(|entry| {
+            let reasons_json = entry
+                .reasons
+                .iter()
+                .map(|reason| format!(r#""{}""#, json_escape(reason)))
+                .collect::<Vec<_>>()
+                .join(",");
+            format!(r#"{{"page":{},"reasons":[{}]}}"#, entry.page, reasons_json)
+        })
+        .collect::<Vec<_>>()
+        .join(",")
+}
+
 fn json_escape(s: &str) -> String {
     let mut out = String::with_capacity(s.len() + 16);
     for ch in s.chars() {
@@ -32,6 +63,7 @@ fn json_escape(s: &str) -> String {
 }
 
 fn main() {
+    #[cfg(not(target_arch = "wasm32"))]
     env_logger::init();
     let args: Vec<String> = env::args().collect();
 
@@ -117,11 +149,13 @@ fn run_analyze(pdf_path: &str, json_output: bool, start: Instant) {
                     .iter()
                     .map(|p| p.to_string())
                     .collect();
+                let ocr_reasons = format_ocr_reasons_by_page(&result.ocr_reasons_by_page);
                 println!(
-                    r#"{{"pdf_type":"{}","page_count":{},"pages_needing_ocr":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"detection_time_ms":{}}}"#,
+                    r#"{{"pdf_type":"{}","page_count":{},"pages_needing_ocr":[{}],"ocr_reasons_by_page":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"detection_time_ms":{}}}"#,
                     pdf_type_str(&result.pdf_type),
                     result.page_count,
                     ocr_pages.join(","),
+                    ocr_reasons,
                     result.layout.is_complex,
                     table_pages.join(","),
                     col_pages.join(","),
@@ -144,6 +178,9 @@ fn run_analyze(pdf_path: &str, json_output: bool, start: Instant) {
                 println!("Page count: {}", result.page_count);
                 if !result.pages_needing_ocr.is_empty() {
                     println!("Pages needing OCR: {:?}", result.pages_needing_ocr);
+                    for entry in &result.ocr_reasons_by_page {
+                        println!("  page {}: {}", entry.page, entry.reasons.join(", "));
+                    }
                 }
                 println!();
                 if result.layout.is_complex {
@@ -184,8 +221,9 @@ fn run_detect_only(pdf_path: &str, json_output: bool, start: Instant) {
                     .iter()
                     .map(|p| p.to_string())
                     .collect();
+                let ocr_reasons = format_detector_ocr_reasons(&result.ocr_reasons_by_page);
                 println!(
-                    r#"{{"pdf_type":"{}","page_count":{},"pages_sampled":{},"pages_with_text":{},"confidence":{:.2},"title":{},"ocr_recommended":{},"pages_needing_ocr":[{}],"detection_time_ms":{}}}"#,
+                    r#"{{"pdf_type":"{}","page_count":{},"pages_sampled":{},"pages_with_text":{},"confidence":{:.2},"title":{},"ocr_recommended":{},"pages_needing_ocr":[{}],"ocr_reasons_by_page":[{}],"detection_time_ms":{}}}"#,
                     pdf_type_str(&result.pdf_type),
                     result.page_count,
                     result.pages_sampled,
@@ -198,6 +236,7 @@ fn run_detect_only(pdf_path: &str, json_output: bool, start: Instant) {
                         .unwrap_or_else(|| "null".to_string()),
                     result.ocr_recommended,
                     ocr_pages.join(","),
+                    ocr_reasons,
                     elapsed.as_millis()
                 );
             } else {
@@ -232,6 +271,9 @@ fn run_detect_only(pdf_path: &str, json_output: bool, start: Instant) {
                             result.pages_needing_ocr, result.page_count
                         );
                     }
+                    for (page, reasons) in &result.ocr_reasons_by_page {
+                        println!("  page {}: {}", page, reasons.join(", "));
+                    }
                 }
                 if let Some(title) = &result.title {
                     println!("Title: {}", title);

src/bin/pdf2md.rs

@@ -1,6 +1,10 @@
 //! CLI tool for PDF to Markdown conversion
 
-use pdf_inspector::{process_pdf_with_options, LayoutComplexity, PdfOptions, PdfType, ProcessMode};
+use pdf_inspector::extractor::ItemType;
+use pdf_inspector::{
+    extract_text_with_positions_pages_with_password, process_pdf_with_options, LayoutComplexity,
+    PdfOptions, PdfType, ProcessMode, TextItem,
+};
 use std::collections::HashSet;
 use std::env;
 use std::fmt::Write;
@@ -31,6 +35,137 @@ fn json_escape(s: &str) -> String {
     out
 }
 
+fn format_ocr_reasons_by_page(reasons: &[pdf_inspector::PageOcrReasons]) -> String {
+    reasons
+        .iter()
+        .map(|entry| {
+            let reasons_json = entry
+                .reasons
+                .iter()
+                .map(|reason| format!(r#""{}""#, json_escape(reason)))
+                .collect::<Vec<_>>()
+                .join(",");
+            format!(r#"{{"page":{},"reasons":[{}]}}"#, entry.page, reasons_json)
+        })
+        .collect::<Vec<_>>()
+        .join(",")
+}
+
+fn item_type_label(item_type: &ItemType) -> &'static str {
+    match item_type {
+        ItemType::Text => "text",
+        ItemType::Image => "image",
+        ItemType::Link(_) => "link",
+        ItemType::FormField => "form_field",
+    }
+}
+
+fn format_items_json(items: &[TextItem]) -> String {
+    let underlined_count = items.iter().filter(|item| item.is_underline).count();
+    let items_json = items
+        .iter()
+        .map(|item| {
+            let mcid = item
+                .mcid
+                .map(|value| value.to_string())
+                .unwrap_or_else(|| "null".to_string());
+            let link_url = match &item.item_type {
+                ItemType::Link(url) => format!(r#","url":"{}""#, json_escape(url)),
+                _ => String::new(),
+            };
+            format!(
+                r#"{{"text":"{}","page":{},"x":{:.2},"y":{:.2},"width":{:.2},"height":{:.2},"font":"{}","font_size":{:.2},"is_bold":{},"is_italic":{},"is_underline":{},"is_strikeout":{},"item_type":"{}","mcid":{}{}}}"#,
+                json_escape(&item.text),
+                item.page,
+                item.x,
+                item.y,
+                item.width,
+                item.height,
+                json_escape(&item.font),
+                item.font_size,
+                item.is_bold,
+                item.is_italic,
+                item.is_underline,
+                item.is_strikeout,
+                item_type_label(&item.item_type),
+                mcid,
+                link_url,
+            )
+        })
+        .collect::<Vec<_>>()
+        .join(",");
+
+    format!(
+        r#"{{"total_items":{},"underlined_count":{},"items":[{}]}}"#,
+        items.len(),
+        underlined_count,
+        items_json
+    )
+}
+
+fn extract_items_json(
+    pdf_path: &str,
+    page_filter: Option<&HashSet<u32>>,
+    password: Option<&str>,
+) -> Result<String, pdf_inspector::PdfError> {
+    extract_text_with_positions_pages_with_password(pdf_path, page_filter, password)
+        .map(|items| format_items_json(&items))
+}
+
+#[cfg(test)]
+mod tests {
+    use super::{extract_items_json, format_items_json};
+    use pdf_inspector::extractor::ItemType;
+    use pdf_inspector::TextItem;
+
+    #[test]
+    fn items_json_includes_position_and_underline_metadata() {
+        let items = vec![TextItem {
+            text: "A \"quoted\" item".to_string(),
+            x: 12.345,
+            y: 67.891,
+            width: 23.456,
+            height: 9.876,
+            font: "F1".to_string(),
+            font_size: 10.0,
+            page: 2,
+            is_bold: false,
+            is_italic: true,
+            is_underline: true,
+            is_strikeout: true,
+            item_type: ItemType::Text,
+            mcid: Some(7),
+        }];
+
+        let json = format_items_json(&items);
+
+        assert!(json.contains(r#""text":"A \"quoted\" item""#));
+        assert!(json.contains(r#""page":2"#));
+        assert!(json.contains(r#""x":12.35"#));
+        assert!(json.contains(r#""is_underline":true"#));
+        assert!(json.contains(r#""item_type":"text""#));
+        assert!(json.contains(r#""mcid":7"#));
+    }
+
+    #[test]
+    fn items_json_uses_supplied_pdf_password() {
+        let path = "tests/fixtures/encrypted-secret123.pdf";
+
+        let without_password = extract_items_json(path, None, None);
+        assert!(
+            without_password.is_err(),
+            "encrypted fixture unexpectedly extracted without a password"
+        );
+
+        let json = extract_items_json(path, None, Some("secret123"))
+            .expect("correct password should decrypt positioned text");
+        assert!(
+            json.contains("Procurement"),
+            "decrypted item JSON should contain fixture text, got {json}"
+        );
+    }
+}
+
 /// Parse a page specification like "1,3,5-10,20" into a HashSet of page numbers.
 fn parse_page_spec(spec: &str) -> Result<HashSet<u32>, String> {
     let mut pages = HashSet::new();
@@ -82,12 +217,14 @@ fn print_layout_info(layout: &LayoutComplexity) {
 }
 
 fn main() {
+    #[cfg(not(target_arch = "wasm32"))]
     env_logger::init();
     let args: Vec<String> = env::args().collect();
 
     if args.len() < 2 {
         eprintln!("Usage: {} <pdf_file> [output_file]", args[0]);
         eprintln!("       {} <pdf_file> --json", args[0]);
+        eprintln!("       {} <pdf_file> --items-json", args[0]);
         eprintln!("       {} <pdf_file> --raw", args[0]);
         eprintln!();
         eprintln!("Converts PDF to Markdown with smart type detection.");
@@ -95,9 +232,14 @@ fn main() {
         eprintln!();
         eprintln!("Options:");
         eprintln!("  --json              Output result as JSON");
+        eprintln!("  --items-json        Output positioned TextItem JSON");
         eprintln!("  --raw               Output only markdown (no headers)");
+        eprintln!(
+            "  --compact           Collapse token-heavy source formatting such as dot leaders"
+        );
         eprintln!("  --pages             Insert page break markers (<!-- Page N -->)");
         eprintln!("  --select-pages N    Only process specified pages (e.g. 1,3,5-10)");
+        eprintln!("  --password PW       Password for an encrypted PDF");
         eprintln!("  --detect-only       Only detect PDF type (no extraction)");
         eprintln!("  --analyze           Detect + extract + layout analysis (no markdown)");
         process::exit(1);
@@ -105,11 +247,23 @@ fn main() {
 
     let pdf_path = &args[1];
     let json_output = args.iter().any(|a| a == "--json");
+    let items_json_output = args.iter().any(|a| a == "--items-json");
     let raw_output = args.iter().any(|a| a == "--raw");
+    let compact_output = args.iter().any(|a| a == "--compact");
     let page_numbers = args.iter().any(|a| a == "--pages");
     let detect_only = args.iter().any(|a| a == "--detect-only");
     let analyze = args.iter().any(|a| a == "--analyze");
 
+    // Parse --password value
+    let password = args.iter().position(|a| a == "--password").map(|i| {
+        args.get(i + 1)
+            .unwrap_or_else(|| {
+                eprintln!("Error: --password requires a value");
+                process::exit(1);
+            })
+            .clone()
+    });
+
     // Parse --select-pages value
     let page_filter = args
         .iter()
@@ -129,6 +283,17 @@ fn main() {
             })
         });
 
+    if items_json_output {
+        match extract_items_json(pdf_path, page_filter.as_ref(), password.as_deref()) {
+            Ok(json) => println!("{}", json),
+            Err(e) => {
+                println!(r#"{{"error":"{}"}}"#, json_escape(&e.to_string()));
+                process::exit(1);
+            }
+        }
+        return;
+    }
+
     let output_file = args
         .get(2)
         .filter(|a| !a.starts_with("--"))
@@ -143,10 +308,14 @@ fn main() {
     };
 
     let mut options = PdfOptions::new().mode(process_mode);
+    if compact_output {
+        options.markdown.profile = pdf_inspector::MarkdownProfile::Compact;
+    }
     options.markdown.include_page_numbers = page_numbers;
     if let Some(pages) = page_filter {
         options.page_filter = Some(pages);
     }
+    options.password = password;
 
     match process_pdf_with_options(pdf_path, options) {
         Ok(result) => {
@@ -177,12 +346,14 @@ fn main() {
                         .iter()
                         .map(|p| p.to_string())
                         .collect();
+                    let ocr_reasons = format_ocr_reasons_by_page(&result.ocr_reasons_by_page);
                     println!(
-                        r#"{{"pdf_type":"{}","page_count":{},"processing_time_ms":{},"pages_needing_ocr":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"has_encoding_issues":{}}}"#,
+                        r#"{{"pdf_type":"{}","page_count":{},"processing_time_ms":{},"pages_needing_ocr":[{}],"ocr_reasons_by_page":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"has_encoding_issues":{}}}"#,
                         pdf_type_str,
                         result.page_count,
                         result.processing_time_ms,
                         ocr_pages.join(","),
+                        ocr_reasons,
                         result.layout.is_complex,
                         table_pages.join(","),
                         col_pages.join(","),
@@ -223,8 +394,9 @@ fn main() {
                     .iter()
                     .map(|p| p.to_string())
                     .collect();
+                let ocr_reasons = format_ocr_reasons_by_page(&result.ocr_reasons_by_page);
                 println!(
-                    r#"{{"pdf_type":"{}","page_count":{},"has_text":{},"processing_time_ms":{},"markdown_length":{},"pages_needing_ocr":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"has_encoding_issues":{},"markdown":"{}"}}"#,
+                    r#"{{"pdf_type":"{}","page_count":{},"has_text":{},"processing_time_ms":{},"markdown_length":{},"pages_needing_ocr":[{}],"ocr_reasons_by_page":[{}],"is_complex":{},"pages_with_tables":[{}],"pages_with_columns":[{}],"has_encoding_issues":{},"markdown":"{}"}}"#,
                     match result.pdf_type {
                         PdfType::TextBased => "text_based",
                         PdfType::Scanned => "scanned",
@@ -236,6 +408,7 @@ fn main() {
                     result.processing_time_ms,
                     result.markdown.as_ref().map(|m| m.len()).unwrap_or(0),
                     ocr_pages.join(","),
+                    ocr_reasons,
                     result.layout.is_complex,
                     table_pages.join(","),
                     col_pages.join(","),

src/detector.rs

@@ -60,6 +60,10 @@ pub struct PdfTypeResult {
     /// 1-indexed page numbers that need OCR (image-only or insufficient text).
     /// Empty for TextBased. All pages for Scanned/ImageBased. Specific pages for Mixed.
     pub pages_needing_ocr: Vec<u32>,
+    /// Per-page explanation for `pages_needing_ocr`: 1-indexed page → reason
+    /// codes (`scanned`, `no_text`, `vector_text`, `suspected_garbled_text`).
+    /// Only contains pages that need OCR.
+    pub ocr_reasons_by_page: std::collections::BTreeMap<u32, Vec<String>>,
 }
 
 /// Configuration for PDF type detection
@@ -382,7 +386,12 @@ pub(crate) fn detect_from_document(
                 let analysis = if let Some(cached) = analysis_cache.get(&page_num) {
                     cached.clone()
                 } else if let Some(&page_id) = pages.get(&page_num) {
-                    analyze_page_content(doc, page_id)
+                    // Cache the fresh analysis so the reason-classification pass
+                    // below sees the real signals (vector_text, etc.) instead of
+                    // defaulting to "scanned".
+                    let a = analyze_page_content(doc, page_id);
+                    analysis_cache.insert(page_num, a.clone());
+                    a
                 } else {
                     continue;
                 };
@@ -429,6 +438,9 @@ pub(crate) fn detect_from_document(
                 let analysis = analyze_page_content(doc, page_id);
                 if analysis.has_identity_h_no_tounicode || analysis.has_only_type3_fonts {
                     pages_needing_ocr.push(page_num);
+                    // Cache so the reason pass reports suspected_garbled_text
+                    // rather than defaulting to "scanned".
+                    analysis_cache.insert(page_num, analysis);
                 }
             }
         }
@@ -436,6 +448,19 @@ pub(crate) fn detect_from_document(
     pages_needing_ocr.sort();
     pages_needing_ocr.dedup();
 
+    // Explain each OCR-flagged page. Pages we analyzed get a signal-derived
+    // reason; pages flagged only by whole-document classification (unsampled
+    // pages of a Scanned/ImageBased doc) default to `scanned`.
+    let mut ocr_reasons_by_page: std::collections::BTreeMap<u32, Vec<String>> =
+        std::collections::BTreeMap::new();
+    for &page_num in &pages_needing_ocr {
+        let reasons = match analysis_cache.get(&page_num) {
+            Some(analysis) => page_ocr_reasons(analysis),
+            None => vec![crate::OCR_REASON_SCANNED],
+        };
+        ocr_reasons_by_page.insert(page_num, reasons.into_iter().map(String::from).collect());
+    }
+
     // Try to get title from metadata
     let title = get_document_title(doc);
 
@@ -448,6 +473,7 @@ pub(crate) fn detect_from_document(
         title,
         ocr_recommended,
         pages_needing_ocr,
+        ocr_reasons_by_page,
     })
 }
 
@@ -487,7 +513,7 @@ fn distribute_pages(n: u32, total: u32) -> Vec<u32> {
 }
 
 /// Page content analysis result
-#[derive(Clone)]
+#[derive(Clone, Default)]
 struct PageAnalysis {
     text_operator_count: u32,
     has_images: bool,
@@ -523,6 +549,31 @@ struct PageAnalysis {
     has_decodable_text_fonts: bool,
 }
 
+/// Explain *why* a page needs OCR, from its content analysis. Priority:
+/// undecodable fonts (`suspected_garbled_text`) and vector-outlined text
+/// (`vector_text`) come first because they persist even when a text layer is
+/// present; otherwise a page with no extractable text is `scanned` when an
+/// image backs it or `no_text` when nothing does.
+fn page_ocr_reasons(a: &PageAnalysis) -> Vec<&'static str> {
+    let mut reasons = Vec::new();
+    if a.has_identity_h_no_tounicode || a.has_only_type3_fonts {
+        reasons.push(crate::OCR_REASON_SUSPECTED_GARBLED_TEXT);
+    }
+    if a.has_vector_text {
+        reasons.push(crate::OCR_REASON_VECTOR_TEXT);
+    }
+    if reasons.is_empty() {
+        let has_extractable_text = a.text_operator_count > 0 && a.unique_text_chars > 0;
+        if !has_extractable_text && !a.has_images && !a.has_template_image {
+            reasons.push(crate::OCR_REASON_NO_TEXT);
+        } else {
+            // Image-backed with no usable text, or too little text to trust.
+            reasons.push(crate::OCR_REASON_SCANNED);
+        }
+    }
+    reasons
+}
+
 /// Extracted font information from a Resource dictionary entry.
 /// Stores the properties needed for decodability/identity-h checks
 /// without holding a reference to the document.
@@ -1608,7 +1659,15 @@ fn hex_val(b: u8) -> Option<u8> {
 /// Standard page: 612x792 points (US Letter) = ~485,000 sq points
 /// At 2x resolution that's ~1.9M pixels, so we use 250K pixels as threshold
 /// (accounting for varying DPI and page sizes)
-fn analyze_page_images(doc: &Document, page_id: ObjectId) -> (bool, u64, bool) {
+/// Returns `(has_images, total_image_area, has_template_image)` for a page.
+/// `has_template_image` means a single large (>50% page coverage)
+/// background image — the signal `classify_pdf`/`detect_pdf_type` uses to
+/// route a page to OCR regardless of any incidental native text drawn over
+/// it. Exposed at crate visibility so extraction-side per-page `needs_ocr`
+/// computation (`extract_pages_markdown_mem`) can consult the same signal
+/// instead of maintaining its own, independent notion of "needs OCR" that
+/// can silently disagree with detection — see #227.
+pub(crate) fn analyze_page_images(doc: &Document, page_id: ObjectId) -> (bool, u64, bool) {
     // Threshold: image covering roughly half a page at 150+ DPI
     // 612 * 792 / 2 * (150/72)^2 ≈ 1M pixels, but we'll be conservative
     const TEMPLATE_IMAGE_THRESHOLD: u64 = 500_000; // 500K pixels
@@ -1690,6 +1749,62 @@ fn analyze_page_images(doc: &Document, page_id: ObjectId) -> (bool, u64, bool) {
     (has_images, total_area, has_template_image)
 }
 
+/// Computes both `(needs_ocr_for_template_image, has_vector_text)` for a
+/// page from a single shared `analyze_page_content` pass — that call
+/// decompresses and scans every content stream (page + XObjects) plus
+/// image coverage, so `extract_pages_markdown_mem` must not invoke it
+/// twice per page (once per signal) the way `detect_from_document` avoids
+/// by caching its per-page `PageAnalysis`.
+///
+/// `needs_ocr_for_template_image` is true when a page's template image
+/// should be treated as a scan needing OCR — a single full-page background
+/// image with little/no real text — rather than a text page that happens
+/// to carry a watermark, letterhead, or figure. Mirrors the two distinct
+/// signals classification uses to route a template-image page to OCR:
+///
+/// 1. `looks_like_scan`: image_count <= 1, few text operators (<50), and
+///    low alphanumeric diversity in raw string operands (unless decodable
+///    CID/ToUnicode fonts explain that away) — the gate used for
+///    `pages_with_template_images` and Mixed-type per-page routing.
+/// 2. Insufficient real text volume, using `DetectionConfig::default()`'s
+///    `min_text_ops_per_page` (3) — the same threshold Mixed-type per-page
+///    routing applies via `text_operator_count < config.min_text_ops_per_page
+///    && has_images` (simplified here since a template image implies
+///    `has_images`). Deliberately *not* the higher `effective_min_ops`
+///    floor (`min_text_ops_per_page.max(10)`) that whole-document
+///    `PdfType::ImageBased`/`Scanned` classification uses for
+///    `pages_with_text` — that's a cross-page aggregate decision this
+///    per-page function has no way to replicate exactly, and the lower
+///    per-page threshold is the one a single page's own signals can
+///    actually agree with.
+///
+/// `has_vector_text` is true when a page has vector-outlined text (glyphs
+/// drawn as paths rather than shown via text-showing operators) —
+/// `detect_from_document`'s Mixed-type per-page routing always sends
+/// these pages to OCR, independent of any template-image check, since
+/// outlined glyphs can't be extracted as text at all.
+///
+/// Exposed at crate visibility so `extract_pages_markdown_mem` can apply
+/// the same gates classification needs elsewhere instead of treating the
+/// raw signals alone as sufficient — see #227/#231.
+pub(crate) fn page_ocr_signals(doc: &Document, page_id: ObjectId) -> (bool, bool) {
+    let analysis = analyze_page_content(doc, page_id);
+
+    let needs_ocr_for_template_image = if !analysis.has_template_image {
+        false
+    } else {
+        let alphanum_low = analysis.unique_alphanum_chars < 10
+            && !(analysis.has_decodable_text_fonts && analysis.text_operator_count >= 10);
+        let looks_like_scan =
+            analysis.image_count <= 1 && analysis.text_operator_count < 50 && alphanum_low;
+        let insufficient_text =
+            analysis.text_operator_count < DetectionConfig::default().min_text_ops_per_page;
+        looks_like_scan || insufficient_text
+    };
+
+    (needs_ocr_for_template_image, analysis.has_vector_text)
+}
+
 /// Recursively collect image dimensions from XObject resources,
 /// including images nested inside Form XObjects.
 fn collect_images_from_resources(
@@ -1809,6 +1924,64 @@ fn get_document_title(doc: &Document) -> Option<String> {
 mod tests {
     use super::*;
 
+    #[test]
+    fn page_ocr_reasons_classify() {
+        // Scanned: no text, full-page image.
+        let scanned = PageAnalysis {
+            has_template_image: true,
+            ..Default::default()
+        };
+        assert_eq!(page_ocr_reasons(&scanned), vec![crate::OCR_REASON_SCANNED]);
+
+        // Image-only page (no template flag, but has an image).
+        let image_only = PageAnalysis {
+            has_images: true,
+            ..Default::default()
+        };
+        assert_eq!(
+            page_ocr_reasons(&image_only),
+            vec![crate::OCR_REASON_SCANNED]
+        );
+
+        // No text, no image → no_text.
+        let blank = PageAnalysis::default();
+        assert_eq!(page_ocr_reasons(&blank), vec![crate::OCR_REASON_NO_TEXT]);
+
+        // Vector-outlined text.
+        let vector = PageAnalysis {
+            has_vector_text: true,
+            ..Default::default()
+        };
+        assert_eq!(
+            page_ocr_reasons(&vector),
+            vec![crate::OCR_REASON_VECTOR_TEXT]
+        );
+
+        // Undecodable fonts → garbled, and it wins over the fall-through.
+        let garbled = PageAnalysis {
+            has_identity_h_no_tounicode: true,
+            has_images: true,
+            ..Default::default()
+        };
+        assert_eq!(
+            page_ocr_reasons(&garbled),
+            vec![crate::OCR_REASON_SUSPECTED_GARBLED_TEXT]
+        );
+
+        // A page with real extractable text and an image is not flagged here
+        // as scanned/no_text (only reached for pages already needing OCR).
+        let text_with_image = PageAnalysis {
+            text_operator_count: 40,
+            unique_text_chars: 120,
+            has_images: true,
+            ..Default::default()
+        };
+        assert_eq!(
+            page_ocr_reasons(&text_with_image),
+            vec![crate::OCR_REASON_SCANNED]
+        );
+    }
+
     #[test]
     fn test_scan_content_operators() {
         let mut uchars = HashSet::new();

src/extractor/base14.rs

@@ -0,0 +1,649 @@
+//! Built-in glyph metrics for the 14 standard PDF fonts.
+//!
+//! PDFs may omit `/Widths` for non-embedded base-14 fonts (Times, Helvetica,
+//! Courier, Symbol, ZapfDingbats); per the PDF spec the reader must supply
+//! the metrics. Without them every text item gets width 0, which breaks
+//! space synthesis, sub/superscript detection, and table column detection
+//! (common in 1990s dvips/Distiller output).
+//!
+//! Tables are generated from the Adobe Core 14 AFM files (via reportlab's
+//! `_fontdata`), keyed by Unicode char, sorted for binary search.
+//! Generator: scratchpad/gen_base14.py (session tooling, not checked in).
+
+/// Width in 1000ths of an em for `c` in the given base-14 font, or `None`
+/// if the font is not one of the base 14 (after name normalization) or the
+/// char has no glyph in its AFM.
+pub(crate) fn base14_char_width(base_font: &str, c: char) -> Option<u16> {
+    let table = base14_table(base_font)?;
+    // AFM tables key visible glyphs only; alias the invisible variants the
+    // cp1252 fallback can produce so they get the metric of their visible
+    // counterpart instead of the generic default.
+    let c = match c {
+        '\u{00A0}' => ' ', // no-break space -> space
+        '\u{00AD}' => '-', // soft hyphen -> hyphen
+        _ => c,
+    };
+    table
+        .binary_search_by_key(&c, |&(ch, _)| ch)
+        .ok()
+        .map(|i| table[i].1)
+}
+
+/// True when the base font name normalizes to one of the standard 14 fonts.
+pub(crate) fn is_base14_font(base_font: &str) -> bool {
+    base14_table(base_font).is_some()
+}
+
+/// Code → Unicode through the font's BUILT-IN encoding, for the base-14
+/// fonts whose repertoire is not Latin (Symbol, ZapfDingbats). Their glyphs
+/// live at byte positions that have nothing to do with cp1252 (Symbol 0x61
+/// renders α, Zapf 0x21 renders ✁), so advance widths must be resolved
+/// through this mapping — the renderer draws these glyphs regardless of how
+/// the text decoder transliterates them. Returns `None` for the Latin text
+/// fonts, which follow standard single-byte encodings.
+pub(crate) fn builtin_encoding_char(base_font: &str, code: u8) -> Option<char> {
+    let table = base14_table(base_font)?;
+    let enc: &[(u8, char)] = if std::ptr::eq(table, SYMBOL) {
+        SYMBOL_ENCODING
+    } else if std::ptr::eq(table, ZAPFDINGBATS) {
+        ZAPFDINGBATS_ENCODING
+    } else {
+        return None;
+    };
+    enc.binary_search_by_key(&code, |&(b, _)| b)
+        .ok()
+        .map(|i| enc[i].1)
+}
+
+/// Map a BaseFont name (possibly subset-prefixed, e.g. "ABCDEF+Times-Bold",
+/// or a common alias like "Arial" / "TimesNewRomanPSMT") to its width table.
+fn base14_table(base_font: &str) -> Option<&'static [(char, u16)]> {
+    // Strip subset prefix "ABCDEF+"
+    let name = match base_font.split_once('+') {
+        Some((prefix, rest))
+            if prefix.len() == 6 && prefix.chars().all(|c| c.is_ascii_uppercase()) =>
+        {
+            rest
+        }
+        _ => base_font,
+    };
+    let lower = name.to_ascii_lowercase();
+    let bold = lower.contains("bold");
+    let italic = lower.contains("italic") || lower.contains("oblique");
+    if lower.contains("courier") {
+        return Some(match (bold, italic) {
+            (false, false) => COURIER,
+            (true, false) => COURIER_BOLD,
+            (false, true) => COURIER_OBLIQUE,
+            (true, true) => COURIER_BOLDOBLIQUE,
+        });
+    }
+    if lower.contains("helvetica") || lower.contains("arial") {
+        return Some(match (bold, italic) {
+            (false, false) => HELVETICA,
+            (true, false) => HELVETICA_BOLD,
+            (false, true) => HELVETICA_OBLIQUE,
+            (true, true) => HELVETICA_BOLDOBLIQUE,
+        });
+    }
+    if lower.contains("times") {
+        return Some(match (bold, italic) {
+            (false, false) => TIMES_ROMAN,
+            (true, false) => TIMES_BOLD,
+            (false, true) => TIMES_ITALIC,
+            (true, true) => TIMES_BOLDITALIC,
+        });
+    }
+    // Symbol and ZapfDingbats have unique glyph repertoires, so only exact
+    // names (plus the common MT/ITC aliases) qualify — a custom font that
+    // merely mentions "Symbol" in its name must not get these metrics.
+    match lower.as_str() {
+        "zapfdingbats" | "dingbats" | "itczapfdingbats" | "zapfdingbatsitc" => {
+            return Some(ZAPFDINGBATS)
+        }
+        "symbol" | "symbolmt" | "symbolitc" => return Some(SYMBOL),
+        _ => {}
+    }
+    None
+}
+
+#[rustfmt::skip]
+static COURIER: &[(char, u16)] = &[
+    (' ', 600), ('!', 600), ('"', 600), ('#', 600), ('$', 600), ('%', 600),
+    ('&', 600), ('\'', 600), ('(', 600), (')', 600), ('*', 600), ('+', 600),
+    (',', 600), ('-', 600), ('.', 600), ('/', 600), ('0', 600), ('1', 600),
+    ('2', 600), ('3', 600), ('4', 600), ('5', 600), ('6', 600), ('7', 600),
+    ('8', 600), ('9', 600), (':', 600), (';', 600), ('<', 600), ('=', 600),
+    ('>', 600), ('?', 600), ('@', 600), ('A', 600), ('B', 600), ('C', 600),
+    ('D', 600), ('E', 600), ('F', 600), ('G', 600), ('H', 600), ('I', 600),
+    ('J', 600), ('K', 600), ('L', 600), ('M', 600), ('N', 600), ('O', 600),
+    ('P', 600), ('Q', 600), ('R', 600), ('S', 600), ('T', 600), ('U', 600),
+    ('V', 600), ('W', 600), ('X', 600), ('Y', 600), ('Z', 600), ('[', 600),
+    ('\\', 600), (']', 600), ('^', 600), ('_', 600), ('`', 600), ('a', 600),
+    ('b', 600), ('c', 600), ('d', 600), ('e', 600), ('f', 600), ('g', 600),
+    ('h', 600), ('i', 600), ('j', 600), ('k', 600), ('l', 600), ('m', 600),
+    ('n', 600), ('o', 600), ('p', 600), ('q', 600), ('r', 600), ('s', 600),
+    ('t', 600), ('u', 600), ('v', 600), ('w', 600), ('x', 600), ('y', 600),
+    ('z', 600), ('{', 600), ('|', 600), ('}', 600), ('~', 600), ('\u{00A1}', 600),
+    ('\u{00A2}', 600), ('\u{00A3}', 600), ('\u{00A4}', 600), ('\u{00A5}', 600), ('\u{00A6}', 600), ('\u{00A7}', 600),
+    ('\u{00A8}', 600), ('\u{00A9}', 600), ('\u{00AA}', 600), ('\u{00AB}', 600), ('\u{00AC}', 600), ('\u{00AE}', 600),
+    ('\u{00AF}', 600), ('\u{00B0}', 600), ('\u{00B1}', 600), ('\u{00B2}', 600), ('\u{00B3}', 600), ('\u{00B4}', 600),
+    ('\u{00B5}', 600), ('\u{00B6}', 600), ('\u{00B7}', 600), ('\u{00B8}', 600), ('\u{00B9}', 600), ('\u{00BA}', 600),
+    ('\u{00BB}', 600), ('\u{00BC}', 600), ('\u{00BD}', 600), ('\u{00BE}', 600), ('\u{00BF}', 600), ('\u{00C0}', 600),
+    ('\u{00C1}', 600), ('\u{00C2}', 600), ('\u{00C3}', 600), ('\u{00C4}', 600), ('\u{00C5}', 600), ('\u{00C6}', 600),
+    ('\u{00C7}', 600), ('\u{00C8}', 600), ('\u{00C9}', 600), ('\u{00CA}', 600), ('\u{00CB}', 600), ('\u{00CC}', 600),
+    ('\u{00CD}', 600), ('\u{00CE}', 600), ('\u{00CF}', 600), ('\u{00D0}', 600), ('\u{00D1}', 600), ('\u{00D2}', 600),
+    ('\u{00D3}', 600), ('\u{00D4}', 600), ('\u{00D5}', 600), ('\u{00D6}', 600), ('\u{00D7}', 600), ('\u{00D8}', 600),
+    ('\u{00D9}', 600), ('\u{00DA}', 600), ('\u{00DB}', 600), ('\u{00DC}', 600), ('\u{00DD}', 600), ('\u{00DE}', 600),
+    ('\u{00DF}', 600), ('\u{00E0}', 600), ('\u{00E1}', 600), ('\u{00E2}', 600), ('\u{00E3}', 600), ('\u{00E4}', 600),
+    ('\u{00E5}', 600), ('\u{00E6}', 600), ('\u{00E7}', 600), ('\u{00E8}', 600), ('\u{00E9}', 600), ('\u{00EA}', 600),
+    ('\u{00EB}', 600), ('\u{00EC}', 600), ('\u{00ED}', 600), ('\u{00EE}', 600), ('\u{00EF}', 600), ('\u{00F0}', 600),
+    ('\u{00F1}', 600), ('\u{00F2}', 600), ('\u{00F3}', 600), ('\u{00F4}', 600), ('\u{00F5}', 600), ('\u{00F6}', 600),
+    ('\u{00F7}', 600), ('\u{00F8}', 600), ('\u{00F9}', 600), ('\u{00FA}', 600), ('\u{00FB}', 600), ('\u{00FC}', 600),
+    ('\u{00FD}', 600), ('\u{00FE}', 600), ('\u{00FF}', 600), ('\u{0131}', 600), ('\u{0141}', 600), ('\u{0142}', 600),
+    ('\u{0152}', 600), ('\u{0153}', 600), ('\u{0160}', 600), ('\u{0161}', 600), ('\u{0178}', 600), ('\u{017D}', 600),
+    ('\u{017E}', 600), ('\u{0192}', 600), ('\u{02C6}', 600), ('\u{02C7}', 600), ('\u{02D8}', 600), ('\u{02D9}', 600),
+    ('\u{02DA}', 600), ('\u{02DB}', 600), ('\u{02DC}', 600), ('\u{02DD}', 600), ('\u{2013}', 600), ('\u{2014}', 600),
+    ('\u{2018}', 600), ('\u{2019}', 600), ('\u{201A}', 600), ('\u{201C}', 600), ('\u{201D}', 600), ('\u{201E}', 600),
+    ('\u{2020}', 600), ('\u{2021}', 600), ('\u{2022}', 600), ('\u{2026}', 600), ('\u{2030}', 600), ('\u{2039}', 600),
+    ('\u{203A}', 600), ('\u{2044}', 600), ('\u{20AC}', 600), ('\u{2122}', 600), ('\u{2212}', 600), ('\u{FB01}', 600),
+    ('\u{FB02}', 600),
+];
+
+static COURIER_BOLD: &[(char, u16)] = COURIER;
+
+static COURIER_OBLIQUE: &[(char, u16)] = COURIER;
+
+static COURIER_BOLDOBLIQUE: &[(char, u16)] = COURIER;
+
+#[rustfmt::skip]
+static HELVETICA: &[(char, u16)] = &[
+    (' ', 278), ('!', 278), ('"', 355), ('#', 556), ('$', 556), ('%', 889),
+    ('&', 667), ('\'', 191), ('(', 333), (')', 333), ('*', 389), ('+', 584),
+    (',', 278), ('-', 333), ('.', 278), ('/', 278), ('0', 556), ('1', 556),
+    ('2', 556), ('3', 556), ('4', 556), ('5', 556), ('6', 556), ('7', 556),
+    ('8', 556), ('9', 556), (':', 278), (';', 278), ('<', 584), ('=', 584),
+    ('>', 584), ('?', 556), ('@', 1015), ('A', 667), ('B', 667), ('C', 722),
+    ('D', 722), ('E', 667), ('F', 611), ('G', 778), ('H', 722), ('I', 278),
+    ('J', 500), ('K', 667), ('L', 556), ('M', 833), ('N', 722), ('O', 778),
+    ('P', 667), ('Q', 778), ('R', 722), ('S', 667), ('T', 611), ('U', 722),
+    ('V', 667), ('W', 944), ('X', 667), ('Y', 667), ('Z', 611), ('[', 278),
+    ('\\', 278), (']', 278), ('^', 469), ('_', 556), ('`', 333), ('a', 556),
+    ('b', 556), ('c', 500), ('d', 556), ('e', 556), ('f', 278), ('g', 556),
+    ('h', 556), ('i', 222), ('j', 222), ('k', 500), ('l', 222), ('m', 833),
+    ('n', 556), ('o', 556), ('p', 556), ('q', 556), ('r', 333), ('s', 500),
+    ('t', 278), ('u', 556), ('v', 500), ('w', 722), ('x', 500), ('y', 500),
+    ('z', 500), ('{', 334), ('|', 260), ('}', 334), ('~', 584), ('\u{00A1}', 333),
+    ('\u{00A2}', 556), ('\u{00A3}', 556), ('\u{00A4}', 556), ('\u{00A5}', 556), ('\u{00A6}', 260), ('\u{00A7}', 556),
+    ('\u{00A8}', 333), ('\u{00A9}', 737), ('\u{00AA}', 370), ('\u{00AB}', 556), ('\u{00AC}', 584), ('\u{00AE}', 737),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 584), ('\u{00B2}', 333), ('\u{00B3}', 333), ('\u{00B4}', 333),
+    ('\u{00B5}', 556), ('\u{00B6}', 537), ('\u{00B7}', 278), ('\u{00B8}', 333), ('\u{00B9}', 333), ('\u{00BA}', 365),
+    ('\u{00BB}', 556), ('\u{00BC}', 834), ('\u{00BD}', 834), ('\u{00BE}', 834), ('\u{00BF}', 611), ('\u{00C0}', 667),
+    ('\u{00C1}', 667), ('\u{00C2}', 667), ('\u{00C3}', 667), ('\u{00C4}', 667), ('\u{00C5}', 667), ('\u{00C6}', 1000),
+    ('\u{00C7}', 722), ('\u{00C8}', 667), ('\u{00C9}', 667), ('\u{00CA}', 667), ('\u{00CB}', 667), ('\u{00CC}', 278),
+    ('\u{00CD}', 278), ('\u{00CE}', 278), ('\u{00CF}', 278), ('\u{00D0}', 722), ('\u{00D1}', 722), ('\u{00D2}', 778),
+    ('\u{00D3}', 778), ('\u{00D4}', 778), ('\u{00D5}', 778), ('\u{00D6}', 778), ('\u{00D7}', 584), ('\u{00D8}', 778),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 667), ('\u{00DE}', 667),
+    ('\u{00DF}', 611), ('\u{00E0}', 556), ('\u{00E1}', 556), ('\u{00E2}', 556), ('\u{00E3}', 556), ('\u{00E4}', 556),
+    ('\u{00E5}', 556), ('\u{00E6}', 889), ('\u{00E7}', 500), ('\u{00E8}', 556), ('\u{00E9}', 556), ('\u{00EA}', 556),
+    ('\u{00EB}', 556), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 556),
+    ('\u{00F1}', 556), ('\u{00F2}', 556), ('\u{00F3}', 556), ('\u{00F4}', 556), ('\u{00F5}', 556), ('\u{00F6}', 556),
+    ('\u{00F7}', 584), ('\u{00F8}', 611), ('\u{00F9}', 556), ('\u{00FA}', 556), ('\u{00FB}', 556), ('\u{00FC}', 556),
+    ('\u{00FD}', 500), ('\u{00FE}', 556), ('\u{00FF}', 500), ('\u{0131}', 278), ('\u{0141}', 556), ('\u{0142}', 222),
+    ('\u{0152}', 1000), ('\u{0153}', 944), ('\u{0160}', 667), ('\u{0161}', 500), ('\u{0178}', 667), ('\u{017D}', 611),
+    ('\u{017E}', 500), ('\u{0192}', 556), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 556), ('\u{2014}', 1000),
+    ('\u{2018}', 222), ('\u{2019}', 222), ('\u{201A}', 222), ('\u{201C}', 333), ('\u{201D}', 333), ('\u{201E}', 333),
+    ('\u{2020}', 556), ('\u{2021}', 556), ('\u{2022}', 350), ('\u{2026}', 1000), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 556), ('\u{2122}', 1000), ('\u{2212}', 584), ('\u{FB01}', 500),
+    ('\u{FB02}', 500),
+];
+
+#[rustfmt::skip]
+static HELVETICA_BOLD: &[(char, u16)] = &[
+    (' ', 278), ('!', 333), ('"', 474), ('#', 556), ('$', 556), ('%', 889),
+    ('&', 722), ('\'', 238), ('(', 333), (')', 333), ('*', 389), ('+', 584),
+    (',', 278), ('-', 333), ('.', 278), ('/', 278), ('0', 556), ('1', 556),
+    ('2', 556), ('3', 556), ('4', 556), ('5', 556), ('6', 556), ('7', 556),
+    ('8', 556), ('9', 556), (':', 333), (';', 333), ('<', 584), ('=', 584),
+    ('>', 584), ('?', 611), ('@', 975), ('A', 722), ('B', 722), ('C', 722),
+    ('D', 722), ('E', 667), ('F', 611), ('G', 778), ('H', 722), ('I', 278),
+    ('J', 556), ('K', 722), ('L', 611), ('M', 833), ('N', 722), ('O', 778),
+    ('P', 667), ('Q', 778), ('R', 722), ('S', 667), ('T', 611), ('U', 722),
+    ('V', 667), ('W', 944), ('X', 667), ('Y', 667), ('Z', 611), ('[', 333),
+    ('\\', 278), (']', 333), ('^', 584), ('_', 556), ('`', 333), ('a', 556),
+    ('b', 611), ('c', 556), ('d', 611), ('e', 556), ('f', 333), ('g', 611),
+    ('h', 611), ('i', 278), ('j', 278), ('k', 556), ('l', 278), ('m', 889),
+    ('n', 611), ('o', 611), ('p', 611), ('q', 611), ('r', 389), ('s', 556),
+    ('t', 333), ('u', 611), ('v', 556), ('w', 778), ('x', 556), ('y', 556),
+    ('z', 500), ('{', 389), ('|', 280), ('}', 389), ('~', 584), ('\u{00A1}', 333),
+    ('\u{00A2}', 556), ('\u{00A3}', 556), ('\u{00A4}', 556), ('\u{00A5}', 556), ('\u{00A6}', 280), ('\u{00A7}', 556),
+    ('\u{00A8}', 333), ('\u{00A9}', 737), ('\u{00AA}', 370), ('\u{00AB}', 556), ('\u{00AC}', 584), ('\u{00AE}', 737),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 584), ('\u{00B2}', 333), ('\u{00B3}', 333), ('\u{00B4}', 333),
+    ('\u{00B5}', 611), ('\u{00B6}', 556), ('\u{00B7}', 278), ('\u{00B8}', 333), ('\u{00B9}', 333), ('\u{00BA}', 365),
+    ('\u{00BB}', 556), ('\u{00BC}', 834), ('\u{00BD}', 834), ('\u{00BE}', 834), ('\u{00BF}', 611), ('\u{00C0}', 722),
+    ('\u{00C1}', 722), ('\u{00C2}', 722), ('\u{00C3}', 722), ('\u{00C4}', 722), ('\u{00C5}', 722), ('\u{00C6}', 1000),
+    ('\u{00C7}', 722), ('\u{00C8}', 667), ('\u{00C9}', 667), ('\u{00CA}', 667), ('\u{00CB}', 667), ('\u{00CC}', 278),
+    ('\u{00CD}', 278), ('\u{00CE}', 278), ('\u{00CF}', 278), ('\u{00D0}', 722), ('\u{00D1}', 722), ('\u{00D2}', 778),
+    ('\u{00D3}', 778), ('\u{00D4}', 778), ('\u{00D5}', 778), ('\u{00D6}', 778), ('\u{00D7}', 584), ('\u{00D8}', 778),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 667), ('\u{00DE}', 667),
+    ('\u{00DF}', 611), ('\u{00E0}', 556), ('\u{00E1}', 556), ('\u{00E2}', 556), ('\u{00E3}', 556), ('\u{00E4}', 556),
+    ('\u{00E5}', 556), ('\u{00E6}', 889), ('\u{00E7}', 556), ('\u{00E8}', 556), ('\u{00E9}', 556), ('\u{00EA}', 556),
+    ('\u{00EB}', 556), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 611),
+    ('\u{00F1}', 611), ('\u{00F2}', 611), ('\u{00F3}', 611), ('\u{00F4}', 611), ('\u{00F5}', 611), ('\u{00F6}', 611),
+    ('\u{00F7}', 584), ('\u{00F8}', 611), ('\u{00F9}', 611), ('\u{00FA}', 611), ('\u{00FB}', 611), ('\u{00FC}', 611),
+    ('\u{00FD}', 556), ('\u{00FE}', 611), ('\u{00FF}', 556), ('\u{0131}', 278), ('\u{0141}', 611), ('\u{0142}', 278),
+    ('\u{0152}', 1000), ('\u{0153}', 944), ('\u{0160}', 667), ('\u{0161}', 556), ('\u{0178}', 667), ('\u{017D}', 611),
+    ('\u{017E}', 500), ('\u{0192}', 556), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 556), ('\u{2014}', 1000),
+    ('\u{2018}', 278), ('\u{2019}', 278), ('\u{201A}', 278), ('\u{201C}', 500), ('\u{201D}', 500), ('\u{201E}', 500),
+    ('\u{2020}', 556), ('\u{2021}', 556), ('\u{2022}', 350), ('\u{2026}', 1000), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 556), ('\u{2122}', 1000), ('\u{2212}', 584), ('\u{FB01}', 611),
+    ('\u{FB02}', 611),
+];
+
+static HELVETICA_OBLIQUE: &[(char, u16)] = HELVETICA;
+
+static HELVETICA_BOLDOBLIQUE: &[(char, u16)] = HELVETICA_BOLD;
+
+#[rustfmt::skip]
+static TIMES_ROMAN: &[(char, u16)] = &[
+    (' ', 250), ('!', 333), ('"', 408), ('#', 500), ('$', 500), ('%', 833),
+    ('&', 778), ('\'', 180), ('(', 333), (')', 333), ('*', 500), ('+', 564),
+    (',', 250), ('-', 333), ('.', 250), ('/', 278), ('0', 500), ('1', 500),
+    ('2', 500), ('3', 500), ('4', 500), ('5', 500), ('6', 500), ('7', 500),
+    ('8', 500), ('9', 500), (':', 278), (';', 278), ('<', 564), ('=', 564),
+    ('>', 564), ('?', 444), ('@', 921), ('A', 722), ('B', 667), ('C', 667),
+    ('D', 722), ('E', 611), ('F', 556), ('G', 722), ('H', 722), ('I', 333),
+    ('J', 389), ('K', 722), ('L', 611), ('M', 889), ('N', 722), ('O', 722),
+    ('P', 556), ('Q', 722), ('R', 667), ('S', 556), ('T', 611), ('U', 722),
+    ('V', 722), ('W', 944), ('X', 722), ('Y', 722), ('Z', 611), ('[', 333),
+    ('\\', 278), (']', 333), ('^', 469), ('_', 500), ('`', 333), ('a', 444),
+    ('b', 500), ('c', 444), ('d', 500), ('e', 444), ('f', 333), ('g', 500),
+    ('h', 500), ('i', 278), ('j', 278), ('k', 500), ('l', 278), ('m', 778),
+    ('n', 500), ('o', 500), ('p', 500), ('q', 500), ('r', 333), ('s', 389),
+    ('t', 278), ('u', 500), ('v', 500), ('w', 722), ('x', 500), ('y', 500),
+    ('z', 444), ('{', 480), ('|', 200), ('}', 480), ('~', 541), ('\u{00A1}', 333),
+    ('\u{00A2}', 500), ('\u{00A3}', 500), ('\u{00A4}', 500), ('\u{00A5}', 500), ('\u{00A6}', 200), ('\u{00A7}', 500),
+    ('\u{00A8}', 333), ('\u{00A9}', 760), ('\u{00AA}', 276), ('\u{00AB}', 500), ('\u{00AC}', 564), ('\u{00AE}', 760),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 564), ('\u{00B2}', 300), ('\u{00B3}', 300), ('\u{00B4}', 333),
+    ('\u{00B5}', 500), ('\u{00B6}', 453), ('\u{00B7}', 250), ('\u{00B8}', 333), ('\u{00B9}', 300), ('\u{00BA}', 310),
+    ('\u{00BB}', 500), ('\u{00BC}', 750), ('\u{00BD}', 750), ('\u{00BE}', 750), ('\u{00BF}', 444), ('\u{00C0}', 722),
+    ('\u{00C1}', 722), ('\u{00C2}', 722), ('\u{00C3}', 722), ('\u{00C4}', 722), ('\u{00C5}', 722), ('\u{00C6}', 889),
+    ('\u{00C7}', 667), ('\u{00C8}', 611), ('\u{00C9}', 611), ('\u{00CA}', 611), ('\u{00CB}', 611), ('\u{00CC}', 333),
+    ('\u{00CD}', 333), ('\u{00CE}', 333), ('\u{00CF}', 333), ('\u{00D0}', 722), ('\u{00D1}', 722), ('\u{00D2}', 722),
+    ('\u{00D3}', 722), ('\u{00D4}', 722), ('\u{00D5}', 722), ('\u{00D6}', 722), ('\u{00D7}', 564), ('\u{00D8}', 722),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 722), ('\u{00DE}', 556),
+    ('\u{00DF}', 500), ('\u{00E0}', 444), ('\u{00E1}', 444), ('\u{00E2}', 444), ('\u{00E3}', 444), ('\u{00E4}', 444),
+    ('\u{00E5}', 444), ('\u{00E6}', 667), ('\u{00E7}', 444), ('\u{00E8}', 444), ('\u{00E9}', 444), ('\u{00EA}', 444),
+    ('\u{00EB}', 444), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 500),
+    ('\u{00F1}', 500), ('\u{00F2}', 500), ('\u{00F3}', 500), ('\u{00F4}', 500), ('\u{00F5}', 500), ('\u{00F6}', 500),
+    ('\u{00F7}', 564), ('\u{00F8}', 500), ('\u{00F9}', 500), ('\u{00FA}', 500), ('\u{00FB}', 500), ('\u{00FC}', 500),
+    ('\u{00FD}', 500), ('\u{00FE}', 500), ('\u{00FF}', 500), ('\u{0131}', 278), ('\u{0141}', 611), ('\u{0142}', 278),
+    ('\u{0152}', 889), ('\u{0153}', 722), ('\u{0160}', 556), ('\u{0161}', 389), ('\u{0178}', 722), ('\u{017D}', 611),
+    ('\u{017E}', 444), ('\u{0192}', 500), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 500), ('\u{2014}', 1000),
+    ('\u{2018}', 333), ('\u{2019}', 333), ('\u{201A}', 333), ('\u{201C}', 444), ('\u{201D}', 444), ('\u{201E}', 444),
+    ('\u{2020}', 500), ('\u{2021}', 500), ('\u{2022}', 350), ('\u{2026}', 1000), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 500), ('\u{2122}', 980), ('\u{2212}', 564), ('\u{FB01}', 556),
+    ('\u{FB02}', 556),
+];
+
+#[rustfmt::skip]
+static TIMES_BOLD: &[(char, u16)] = &[
+    (' ', 250), ('!', 333), ('"', 555), ('#', 500), ('$', 500), ('%', 1000),
+    ('&', 833), ('\'', 278), ('(', 333), (')', 333), ('*', 500), ('+', 570),
+    (',', 250), ('-', 333), ('.', 250), ('/', 278), ('0', 500), ('1', 500),
+    ('2', 500), ('3', 500), ('4', 500), ('5', 500), ('6', 500), ('7', 500),
+    ('8', 500), ('9', 500), (':', 333), (';', 333), ('<', 570), ('=', 570),
+    ('>', 570), ('?', 500), ('@', 930), ('A', 722), ('B', 667), ('C', 722),
+    ('D', 722), ('E', 667), ('F', 611), ('G', 778), ('H', 778), ('I', 389),
+    ('J', 500), ('K', 778), ('L', 667), ('M', 944), ('N', 722), ('O', 778),
+    ('P', 611), ('Q', 778), ('R', 722), ('S', 556), ('T', 667), ('U', 722),
+    ('V', 722), ('W', 1000), ('X', 722), ('Y', 722), ('Z', 667), ('[', 333),
+    ('\\', 278), (']', 333), ('^', 581), ('_', 500), ('`', 333), ('a', 500),
+    ('b', 556), ('c', 444), ('d', 556), ('e', 444), ('f', 333), ('g', 500),
+    ('h', 556), ('i', 278), ('j', 333), ('k', 556), ('l', 278), ('m', 833),
+    ('n', 556), ('o', 500), ('p', 556), ('q', 556), ('r', 444), ('s', 389),
+    ('t', 333), ('u', 556), ('v', 500), ('w', 722), ('x', 500), ('y', 500),
+    ('z', 444), ('{', 394), ('|', 220), ('}', 394), ('~', 520), ('\u{00A1}', 333),
+    ('\u{00A2}', 500), ('\u{00A3}', 500), ('\u{00A4}', 500), ('\u{00A5}', 500), ('\u{00A6}', 220), ('\u{00A7}', 500),
+    ('\u{00A8}', 333), ('\u{00A9}', 747), ('\u{00AA}', 300), ('\u{00AB}', 500), ('\u{00AC}', 570), ('\u{00AE}', 747),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 570), ('\u{00B2}', 300), ('\u{00B3}', 300), ('\u{00B4}', 333),
+    ('\u{00B5}', 556), ('\u{00B6}', 540), ('\u{00B7}', 250), ('\u{00B8}', 333), ('\u{00B9}', 300), ('\u{00BA}', 330),
+    ('\u{00BB}', 500), ('\u{00BC}', 750), ('\u{00BD}', 750), ('\u{00BE}', 750), ('\u{00BF}', 500), ('\u{00C0}', 722),
+    ('\u{00C1}', 722), ('\u{00C2}', 722), ('\u{00C3}', 722), ('\u{00C4}', 722), ('\u{00C5}', 722), ('\u{00C6}', 1000),
+    ('\u{00C7}', 722), ('\u{00C8}', 667), ('\u{00C9}', 667), ('\u{00CA}', 667), ('\u{00CB}', 667), ('\u{00CC}', 389),
+    ('\u{00CD}', 389), ('\u{00CE}', 389), ('\u{00CF}', 389), ('\u{00D0}', 722), ('\u{00D1}', 722), ('\u{00D2}', 778),
+    ('\u{00D3}', 778), ('\u{00D4}', 778), ('\u{00D5}', 778), ('\u{00D6}', 778), ('\u{00D7}', 570), ('\u{00D8}', 778),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 722), ('\u{00DE}', 611),
+    ('\u{00DF}', 556), ('\u{00E0}', 500), ('\u{00E1}', 500), ('\u{00E2}', 500), ('\u{00E3}', 500), ('\u{00E4}', 500),
+    ('\u{00E5}', 500), ('\u{00E6}', 722), ('\u{00E7}', 444), ('\u{00E8}', 444), ('\u{00E9}', 444), ('\u{00EA}', 444),
+    ('\u{00EB}', 444), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 500),
+    ('\u{00F1}', 556), ('\u{00F2}', 500), ('\u{00F3}', 500), ('\u{00F4}', 500), ('\u{00F5}', 500), ('\u{00F6}', 500),
+    ('\u{00F7}', 570), ('\u{00F8}', 500), ('\u{00F9}', 556), ('\u{00FA}', 556), ('\u{00FB}', 556), ('\u{00FC}', 556),
+    ('\u{00FD}', 500), ('\u{00FE}', 556), ('\u{00FF}', 500), ('\u{0131}', 278), ('\u{0141}', 667), ('\u{0142}', 278),
+    ('\u{0152}', 1000), ('\u{0153}', 722), ('\u{0160}', 556), ('\u{0161}', 389), ('\u{0178}', 722), ('\u{017D}', 667),
+    ('\u{017E}', 444), ('\u{0192}', 500), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 500), ('\u{2014}', 1000),
+    ('\u{2018}', 333), ('\u{2019}', 333), ('\u{201A}', 333), ('\u{201C}', 500), ('\u{201D}', 500), ('\u{201E}', 500),
+    ('\u{2020}', 500), ('\u{2021}', 500), ('\u{2022}', 350), ('\u{2026}', 1000), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 500), ('\u{2122}', 1000), ('\u{2212}', 570), ('\u{FB01}', 556),
+    ('\u{FB02}', 556),
+];
+
+#[rustfmt::skip]
+static TIMES_ITALIC: &[(char, u16)] = &[
+    (' ', 250), ('!', 333), ('"', 420), ('#', 500), ('$', 500), ('%', 833),
+    ('&', 778), ('\'', 214), ('(', 333), (')', 333), ('*', 500), ('+', 675),
+    (',', 250), ('-', 333), ('.', 250), ('/', 278), ('0', 500), ('1', 500),
+    ('2', 500), ('3', 500), ('4', 500), ('5', 500), ('6', 500), ('7', 500),
+    ('8', 500), ('9', 500), (':', 333), (';', 333), ('<', 675), ('=', 675),
+    ('>', 675), ('?', 500), ('@', 920), ('A', 611), ('B', 611), ('C', 667),
+    ('D', 722), ('E', 611), ('F', 611), ('G', 722), ('H', 722), ('I', 333),
+    ('J', 444), ('K', 667), ('L', 556), ('M', 833), ('N', 667), ('O', 722),
+    ('P', 611), ('Q', 722), ('R', 611), ('S', 500), ('T', 556), ('U', 722),
+    ('V', 611), ('W', 833), ('X', 611), ('Y', 556), ('Z', 556), ('[', 389),
+    ('\\', 278), (']', 389), ('^', 422), ('_', 500), ('`', 333), ('a', 500),
+    ('b', 500), ('c', 444), ('d', 500), ('e', 444), ('f', 278), ('g', 500),
+    ('h', 500), ('i', 278), ('j', 278), ('k', 444), ('l', 278), ('m', 722),
+    ('n', 500), ('o', 500), ('p', 500), ('q', 500), ('r', 389), ('s', 389),
+    ('t', 278), ('u', 500), ('v', 444), ('w', 667), ('x', 444), ('y', 444),
+    ('z', 389), ('{', 400), ('|', 275), ('}', 400), ('~', 541), ('\u{00A1}', 389),
+    ('\u{00A2}', 500), ('\u{00A3}', 500), ('\u{00A4}', 500), ('\u{00A5}', 500), ('\u{00A6}', 275), ('\u{00A7}', 500),
+    ('\u{00A8}', 333), ('\u{00A9}', 760), ('\u{00AA}', 276), ('\u{00AB}', 500), ('\u{00AC}', 675), ('\u{00AE}', 760),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 675), ('\u{00B2}', 300), ('\u{00B3}', 300), ('\u{00B4}', 333),
+    ('\u{00B5}', 500), ('\u{00B6}', 523), ('\u{00B7}', 250), ('\u{00B8}', 333), ('\u{00B9}', 300), ('\u{00BA}', 310),
+    ('\u{00BB}', 500), ('\u{00BC}', 750), ('\u{00BD}', 750), ('\u{00BE}', 750), ('\u{00BF}', 500), ('\u{00C0}', 611),
+    ('\u{00C1}', 611), ('\u{00C2}', 611), ('\u{00C3}', 611), ('\u{00C4}', 611), ('\u{00C5}', 611), ('\u{00C6}', 889),
+    ('\u{00C7}', 667), ('\u{00C8}', 611), ('\u{00C9}', 611), ('\u{00CA}', 611), ('\u{00CB}', 611), ('\u{00CC}', 333),
+    ('\u{00CD}', 333), ('\u{00CE}', 333), ('\u{00CF}', 333), ('\u{00D0}', 722), ('\u{00D1}', 667), ('\u{00D2}', 722),
+    ('\u{00D3}', 722), ('\u{00D4}', 722), ('\u{00D5}', 722), ('\u{00D6}', 722), ('\u{00D7}', 675), ('\u{00D8}', 722),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 556), ('\u{00DE}', 611),
+    ('\u{00DF}', 500), ('\u{00E0}', 500), ('\u{00E1}', 500), ('\u{00E2}', 500), ('\u{00E3}', 500), ('\u{00E4}', 500),
+    ('\u{00E5}', 500), ('\u{00E6}', 667), ('\u{00E7}', 444), ('\u{00E8}', 444), ('\u{00E9}', 444), ('\u{00EA}', 444),
+    ('\u{00EB}', 444), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 500),
+    ('\u{00F1}', 500), ('\u{00F2}', 500), ('\u{00F3}', 500), ('\u{00F4}', 500), ('\u{00F5}', 500), ('\u{00F6}', 500),
+    ('\u{00F7}', 675), ('\u{00F8}', 500), ('\u{00F9}', 500), ('\u{00FA}', 500), ('\u{00FB}', 500), ('\u{00FC}', 500),
+    ('\u{00FD}', 444), ('\u{00FE}', 500), ('\u{00FF}', 444), ('\u{0131}', 278), ('\u{0141}', 556), ('\u{0142}', 278),
+    ('\u{0152}', 944), ('\u{0153}', 667), ('\u{0160}', 500), ('\u{0161}', 389), ('\u{0178}', 556), ('\u{017D}', 556),
+    ('\u{017E}', 389), ('\u{0192}', 500), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 500), ('\u{2014}', 889),
+    ('\u{2018}', 333), ('\u{2019}', 333), ('\u{201A}', 333), ('\u{201C}', 556), ('\u{201D}', 556), ('\u{201E}', 556),
+    ('\u{2020}', 500), ('\u{2021}', 500), ('\u{2022}', 350), ('\u{2026}', 889), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 500), ('\u{2122}', 980), ('\u{2212}', 675), ('\u{FB01}', 500),
+    ('\u{FB02}', 500),
+];
+
+#[rustfmt::skip]
+static TIMES_BOLDITALIC: &[(char, u16)] = &[
+    (' ', 250), ('!', 389), ('"', 555), ('#', 500), ('$', 500), ('%', 833),
+    ('&', 778), ('\'', 278), ('(', 333), (')', 333), ('*', 500), ('+', 570),
+    (',', 250), ('-', 333), ('.', 250), ('/', 278), ('0', 500), ('1', 500),
+    ('2', 500), ('3', 500), ('4', 500), ('5', 500), ('6', 500), ('7', 500),
+    ('8', 500), ('9', 500), (':', 333), (';', 333), ('<', 570), ('=', 570),
+    ('>', 570), ('?', 500), ('@', 832), ('A', 667), ('B', 667), ('C', 667),
+    ('D', 722), ('E', 667), ('F', 667), ('G', 722), ('H', 778), ('I', 389),
+    ('J', 500), ('K', 667), ('L', 611), ('M', 889), ('N', 722), ('O', 722),
+    ('P', 611), ('Q', 722), ('R', 667), ('S', 556), ('T', 611), ('U', 722),
+    ('V', 667), ('W', 889), ('X', 667), ('Y', 611), ('Z', 611), ('[', 333),
+    ('\\', 278), (']', 333), ('^', 570), ('_', 500), ('`', 333), ('a', 500),
+    ('b', 500), ('c', 444), ('d', 500), ('e', 444), ('f', 333), ('g', 500),
+    ('h', 556), ('i', 278), ('j', 278), ('k', 500), ('l', 278), ('m', 778),
+    ('n', 556), ('o', 500), ('p', 500), ('q', 500), ('r', 389), ('s', 389),
+    ('t', 278), ('u', 556), ('v', 444), ('w', 667), ('x', 500), ('y', 444),
+    ('z', 389), ('{', 348), ('|', 220), ('}', 348), ('~', 570), ('\u{00A1}', 389),
+    ('\u{00A2}', 500), ('\u{00A3}', 500), ('\u{00A4}', 500), ('\u{00A5}', 500), ('\u{00A6}', 220), ('\u{00A7}', 500),
+    ('\u{00A8}', 333), ('\u{00A9}', 747), ('\u{00AA}', 266), ('\u{00AB}', 500), ('\u{00AC}', 606), ('\u{00AE}', 747),
+    ('\u{00AF}', 333), ('\u{00B0}', 400), ('\u{00B1}', 570), ('\u{00B2}', 300), ('\u{00B3}', 300), ('\u{00B4}', 333),
+    ('\u{00B5}', 576), ('\u{00B6}', 500), ('\u{00B7}', 250), ('\u{00B8}', 333), ('\u{00B9}', 300), ('\u{00BA}', 300),
+    ('\u{00BB}', 500), ('\u{00BC}', 750), ('\u{00BD}', 750), ('\u{00BE}', 750), ('\u{00BF}', 500), ('\u{00C0}', 667),
+    ('\u{00C1}', 667), ('\u{00C2}', 667), ('\u{00C3}', 667), ('\u{00C4}', 667), ('\u{00C5}', 667), ('\u{00C6}', 944),
+    ('\u{00C7}', 667), ('\u{00C8}', 667), ('\u{00C9}', 667), ('\u{00CA}', 667), ('\u{00CB}', 667), ('\u{00CC}', 389),
+    ('\u{00CD}', 389), ('\u{00CE}', 389), ('\u{00CF}', 389), ('\u{00D0}', 722), ('\u{00D1}', 722), ('\u{00D2}', 722),
+    ('\u{00D3}', 722), ('\u{00D4}', 722), ('\u{00D5}', 722), ('\u{00D6}', 722), ('\u{00D7}', 570), ('\u{00D8}', 722),
+    ('\u{00D9}', 722), ('\u{00DA}', 722), ('\u{00DB}', 722), ('\u{00DC}', 722), ('\u{00DD}', 611), ('\u{00DE}', 611),
+    ('\u{00DF}', 500), ('\u{00E0}', 500), ('\u{00E1}', 500), ('\u{00E2}', 500), ('\u{00E3}', 500), ('\u{00E4}', 500),
+    ('\u{00E5}', 500), ('\u{00E6}', 722), ('\u{00E7}', 444), ('\u{00E8}', 444), ('\u{00E9}', 444), ('\u{00EA}', 444),
+    ('\u{00EB}', 444), ('\u{00EC}', 278), ('\u{00ED}', 278), ('\u{00EE}', 278), ('\u{00EF}', 278), ('\u{00F0}', 500),
+    ('\u{00F1}', 556), ('\u{00F2}', 500), ('\u{00F3}', 500), ('\u{00F4}', 500), ('\u{00F5}', 500), ('\u{00F6}', 500),
+    ('\u{00F7}', 570), ('\u{00F8}', 500), ('\u{00F9}', 556), ('\u{00FA}', 556), ('\u{00FB}', 556), ('\u{00FC}', 556),
+    ('\u{00FD}', 444), ('\u{00FE}', 500), ('\u{00FF}', 444), ('\u{0131}', 278), ('\u{0141}', 611), ('\u{0142}', 278),
+    ('\u{0152}', 944), ('\u{0153}', 722), ('\u{0160}', 556), ('\u{0161}', 389), ('\u{0178}', 611), ('\u{017D}', 611),
+    ('\u{017E}', 389), ('\u{0192}', 500), ('\u{02C6}', 333), ('\u{02C7}', 333), ('\u{02D8}', 333), ('\u{02D9}', 333),
+    ('\u{02DA}', 333), ('\u{02DB}', 333), ('\u{02DC}', 333), ('\u{02DD}', 333), ('\u{2013}', 500), ('\u{2014}', 1000),
+    ('\u{2018}', 333), ('\u{2019}', 333), ('\u{201A}', 333), ('\u{201C}', 500), ('\u{201D}', 500), ('\u{201E}', 500),
+    ('\u{2020}', 500), ('\u{2021}', 500), ('\u{2022}', 350), ('\u{2026}', 1000), ('\u{2030}', 1000), ('\u{2039}', 333),
+    ('\u{203A}', 333), ('\u{2044}', 167), ('\u{20AC}', 500), ('\u{2122}', 1000), ('\u{2212}', 606), ('\u{FB01}', 556),
+    ('\u{FB02}', 556),
+];
+
+#[rustfmt::skip]
+static SYMBOL: &[(char, u16)] = &[
+    (' ', 250), ('!', 333), ('#', 500), ('%', 833), ('&', 778), ('(', 333),
+    (')', 333), ('+', 549), (',', 250), ('.', 250), ('/', 278), ('0', 500),
+    ('1', 500), ('2', 500), ('3', 500), ('4', 500), ('5', 500), ('6', 500),
+    ('7', 500), ('8', 500), ('9', 500), (':', 278), (';', 278), ('<', 549),
+    ('=', 549), ('>', 549), ('?', 444), ('[', 333), (']', 333), ('_', 500),
+    ('{', 480), ('|', 200), ('}', 480), ('\u{00AC}', 713), ('\u{00B0}', 400), ('\u{00B1}', 549),
+    ('\u{00B5}', 576), ('\u{00D7}', 549), ('\u{00F7}', 549), ('\u{0192}', 500), ('\u{0391}', 722), ('\u{0392}', 667),
+    ('\u{0393}', 603), ('\u{0395}', 611), ('\u{0396}', 611), ('\u{0397}', 722), ('\u{0398}', 741), ('\u{0399}', 333),
+    ('\u{039A}', 722), ('\u{039B}', 686), ('\u{039C}', 889), ('\u{039D}', 722), ('\u{039E}', 645), ('\u{039F}', 722),
+    ('\u{03A0}', 768), ('\u{03A1}', 556), ('\u{03A3}', 592), ('\u{03A4}', 611), ('\u{03A5}', 690), ('\u{03A6}', 763),
+    ('\u{03A7}', 722), ('\u{03A8}', 795), ('\u{03B1}', 631), ('\u{03B2}', 549), ('\u{03B3}', 411), ('\u{03B4}', 494),
+    ('\u{03B5}', 439), ('\u{03B6}', 494), ('\u{03B7}', 603), ('\u{03B8}', 521), ('\u{03B9}', 329), ('\u{03BA}', 549),
+    ('\u{03BB}', 549), ('\u{03BD}', 521), ('\u{03BE}', 493), ('\u{03BF}', 549), ('\u{03C0}', 549), ('\u{03C1}', 549),
+    ('\u{03C2}', 439), ('\u{03C3}', 603), ('\u{03C4}', 439), ('\u{03C5}', 576), ('\u{03C6}', 521), ('\u{03C7}', 549),
+    ('\u{03C8}', 686), ('\u{03C9}', 686), ('\u{03D1}', 631), ('\u{03D2}', 620), ('\u{03D5}', 603), ('\u{03D6}', 713),
+    ('\u{2022}', 460), ('\u{2026}', 1000), ('\u{2032}', 247), ('\u{2033}', 411), ('\u{2044}', 167), ('\u{20AC}', 750),
+    ('\u{2111}', 686), ('\u{2118}', 987), ('\u{211C}', 795), ('\u{2126}', 768), ('\u{2135}', 823), ('\u{2190}', 987),
+    ('\u{2191}', 603), ('\u{2192}', 987), ('\u{2193}', 603), ('\u{2194}', 1042), ('\u{21B5}', 658), ('\u{21D0}', 987),
+    ('\u{21D1}', 603), ('\u{21D2}', 987), ('\u{21D3}', 603), ('\u{21D4}', 1042), ('\u{2200}', 713), ('\u{2202}', 494),
+    ('\u{2203}', 549), ('\u{2205}', 823), ('\u{2206}', 612), ('\u{2207}', 713), ('\u{2208}', 713), ('\u{2209}', 713),
+    ('\u{220B}', 439), ('\u{220F}', 823), ('\u{2211}', 713), ('\u{2212}', 549), ('\u{2217}', 500), ('\u{221A}', 549),
+    ('\u{221D}', 713), ('\u{221E}', 713), ('\u{2220}', 768), ('\u{2227}', 603), ('\u{2228}', 603), ('\u{2229}', 768),
+    ('\u{222A}', 768), ('\u{222B}', 274), ('\u{2234}', 863), ('\u{223C}', 549), ('\u{2245}', 549), ('\u{2248}', 549),
+    ('\u{2260}', 549), ('\u{2261}', 549), ('\u{2264}', 549), ('\u{2265}', 549), ('\u{2282}', 713), ('\u{2283}', 713),
+    ('\u{2284}', 713), ('\u{2286}', 713), ('\u{2287}', 713), ('\u{2295}', 768), ('\u{2297}', 768), ('\u{22A5}', 658),
+    ('\u{22C5}', 250), ('\u{2320}', 686), ('\u{2321}', 686), ('\u{2329}', 329), ('\u{232A}', 329), ('\u{25CA}', 494),
+    ('\u{2660}', 753), ('\u{2663}', 753), ('\u{2665}', 753), ('\u{2666}', 753), ('\u{F6D9}', 790), ('\u{F6DA}', 790),
+    ('\u{F6DB}', 890), ('\u{F8E5}', 500), ('\u{F8E6}', 603), ('\u{F8E7}', 1000), ('\u{F8E8}', 790), ('\u{F8E9}', 790),
+    ('\u{F8EA}', 786), ('\u{F8EB}', 384), ('\u{F8EC}', 384), ('\u{F8ED}', 384), ('\u{F8EE}', 384), ('\u{F8EF}', 384),
+    ('\u{F8F0}', 384), ('\u{F8F1}', 494), ('\u{F8F2}', 494), ('\u{F8F3}', 494), ('\u{F8F4}', 494), ('\u{F8F5}', 686),
+    ('\u{F8F6}', 384), ('\u{F8F7}', 384), ('\u{F8F8}', 384), ('\u{F8F9}', 384), ('\u{F8FA}', 384), ('\u{F8FB}', 384),
+    ('\u{F8FC}', 494), ('\u{F8FD}', 494), ('\u{F8FE}', 494), ('\u{F8FF}', 790),
+];
+
+#[rustfmt::skip]
+static ZAPFDINGBATS: &[(char, u16)] = &[
+    (' ', 278), ('\u{2192}', 838), ('\u{2194}', 1016), ('\u{2195}', 458), ('\u{2460}', 788), ('\u{2461}', 788),
+    ('\u{2462}', 788), ('\u{2463}', 788), ('\u{2464}', 788), ('\u{2465}', 788), ('\u{2466}', 788), ('\u{2467}', 788),
+    ('\u{2468}', 788), ('\u{2469}', 788), ('\u{25A0}', 761), ('\u{25B2}', 892), ('\u{25BC}', 892), ('\u{25C6}', 788),
+    ('\u{25CF}', 791), ('\u{25D7}', 438), ('\u{2605}', 816), ('\u{260E}', 719), ('\u{261B}', 960), ('\u{261E}', 939),
+    ('\u{2660}', 626), ('\u{2663}', 776), ('\u{2665}', 694), ('\u{2666}', 595), ('\u{2701}', 974), ('\u{2702}', 961),
+    ('\u{2703}', 974), ('\u{2704}', 980), ('\u{2706}', 789), ('\u{2707}', 790), ('\u{2708}', 791), ('\u{2709}', 690),
+    ('\u{270C}', 549), ('\u{270D}', 855), ('\u{270E}', 911), ('\u{270F}', 933), ('\u{2710}', 911), ('\u{2711}', 945),
+    ('\u{2712}', 974), ('\u{2713}', 755), ('\u{2714}', 846), ('\u{2715}', 762), ('\u{2716}', 761), ('\u{2717}', 571),
+    ('\u{2718}', 677), ('\u{2719}', 763), ('\u{271A}', 760), ('\u{271B}', 759), ('\u{271C}', 754), ('\u{271D}', 494),
+    ('\u{271E}', 552), ('\u{271F}', 537), ('\u{2720}', 577), ('\u{2721}', 692), ('\u{2722}', 786), ('\u{2723}', 788),
+    ('\u{2724}', 788), ('\u{2725}', 790), ('\u{2726}', 793), ('\u{2727}', 794), ('\u{2729}', 823), ('\u{272A}', 789),
+    ('\u{272B}', 841), ('\u{272C}', 823), ('\u{272D}', 833), ('\u{272E}', 816), ('\u{272F}', 831), ('\u{2730}', 923),
+    ('\u{2731}', 744), ('\u{2732}', 723), ('\u{2733}', 749), ('\u{2734}', 790), ('\u{2735}', 792), ('\u{2736}', 695),
+    ('\u{2737}', 776), ('\u{2738}', 768), ('\u{2739}', 792), ('\u{273A}', 759), ('\u{273B}', 707), ('\u{273C}', 708),
+    ('\u{273D}', 682), ('\u{273E}', 701), ('\u{273F}', 826), ('\u{2740}', 815), ('\u{2741}', 789), ('\u{2742}', 789),
+    ('\u{2743}', 707), ('\u{2744}', 687), ('\u{2745}', 696), ('\u{2746}', 689), ('\u{2747}', 786), ('\u{2748}', 787),
+    ('\u{2749}', 713), ('\u{274A}', 791), ('\u{274B}', 785), ('\u{274D}', 873), ('\u{274F}', 762), ('\u{2750}', 762),
+    ('\u{2751}', 759), ('\u{2752}', 759), ('\u{2756}', 784), ('\u{2758}', 138), ('\u{2759}', 277), ('\u{275A}', 415),
+    ('\u{275B}', 392), ('\u{275C}', 392), ('\u{275D}', 668), ('\u{275E}', 668), ('\u{2761}', 732), ('\u{2762}', 544),
+    ('\u{2763}', 544), ('\u{2764}', 910), ('\u{2765}', 667), ('\u{2766}', 760), ('\u{2767}', 760), ('\u{2768}', 390),
+    ('\u{2769}', 390), ('\u{276A}', 317), ('\u{276B}', 317), ('\u{276C}', 276), ('\u{276D}', 276), ('\u{276E}', 509),
+    ('\u{276F}', 509), ('\u{2770}', 410), ('\u{2771}', 410), ('\u{2772}', 234), ('\u{2773}', 234), ('\u{2774}', 334),
+    ('\u{2775}', 334), ('\u{2776}', 788), ('\u{2777}', 788), ('\u{2778}', 788), ('\u{2779}', 788), ('\u{277A}', 788),
+    ('\u{277B}', 788), ('\u{277C}', 788), ('\u{277D}', 788), ('\u{277E}', 788), ('\u{277F}', 788), ('\u{2780}', 788),
+    ('\u{2781}', 788), ('\u{2782}', 788), ('\u{2783}', 788), ('\u{2784}', 788), ('\u{2785}', 788), ('\u{2786}', 788),
+    ('\u{2787}', 788), ('\u{2788}', 788), ('\u{2789}', 788), ('\u{278A}', 788), ('\u{278B}', 788), ('\u{278C}', 788),
+    ('\u{278D}', 788), ('\u{278E}', 788), ('\u{278F}', 788), ('\u{2790}', 788), ('\u{2791}', 788), ('\u{2792}', 788),
+    ('\u{2793}', 788), ('\u{2794}', 894), ('\u{2798}', 748), ('\u{2799}', 924), ('\u{279A}', 748), ('\u{279B}', 918),
+    ('\u{279C}', 927), ('\u{279D}', 928), ('\u{279E}', 928), ('\u{279F}', 834), ('\u{27A0}', 873), ('\u{27A1}', 828),
+    ('\u{27A2}', 924), ('\u{27A3}', 924), ('\u{27A4}', 917), ('\u{27A5}', 930), ('\u{27A6}', 931), ('\u{27A7}', 463),
+    ('\u{27A8}', 883), ('\u{27A9}', 836), ('\u{27AA}', 836), ('\u{27AB}', 867), ('\u{27AC}', 867), ('\u{27AD}', 696),
+    ('\u{27AE}', 696), ('\u{27AF}', 874), ('\u{27B1}', 874), ('\u{27B2}', 760), ('\u{27B3}', 946), ('\u{27B4}', 771),
+    ('\u{27B5}', 865), ('\u{27B6}', 771), ('\u{27B7}', 888), ('\u{27B8}', 967), ('\u{27B9}', 888), ('\u{27BA}', 831),
+    ('\u{27BB}', 873), ('\u{27BC}', 927), ('\u{27BD}', 970), ('\u{27BE}', 918),
+];
+
+#[rustfmt::skip]
+static SYMBOL_ENCODING: &[(u8, char)] = &[
+    (0x20, ' '), (0x21, '!'), (0x22, '\u{2200}'), (0x23, '#'), (0x24, '\u{2203}'), (0x25, '%'),
+    (0x26, '&'), (0x27, '\u{220B}'), (0x28, '('), (0x29, ')'), (0x2A, '\u{2217}'), (0x2B, '+'),
+    (0x2C, ','), (0x2D, '\u{2212}'), (0x2E, '.'), (0x2F, '/'), (0x30, '0'), (0x31, '1'),
+    (0x32, '2'), (0x33, '3'), (0x34, '4'), (0x35, '5'), (0x36, '6'), (0x37, '7'),
+    (0x38, '8'), (0x39, '9'), (0x3A, ':'), (0x3B, ';'), (0x3C, '<'), (0x3D, '='),
+    (0x3E, '>'), (0x3F, '?'), (0x40, '\u{2245}'), (0x41, '\u{0391}'), (0x42, '\u{0392}'), (0x43, '\u{03A7}'),
+    (0x44, '\u{2206}'), (0x45, '\u{0395}'), (0x46, '\u{03A6}'), (0x47, '\u{0393}'), (0x48, '\u{0397}'), (0x49, '\u{0399}'),
+    (0x4A, '\u{03D1}'), (0x4B, '\u{039A}'), (0x4C, '\u{039B}'), (0x4D, '\u{039C}'), (0x4E, '\u{039D}'), (0x4F, '\u{039F}'),
+    (0x50, '\u{03A0}'), (0x51, '\u{0398}'), (0x52, '\u{03A1}'), (0x53, '\u{03A3}'), (0x54, '\u{03A4}'), (0x55, '\u{03A5}'),
+    (0x56, '\u{03C2}'), (0x57, '\u{2126}'), (0x58, '\u{039E}'), (0x59, '\u{03A8}'), (0x5A, '\u{0396}'), (0x5B, '['),
+    (0x5C, '\u{2234}'), (0x5D, ']'), (0x5E, '\u{22A5}'), (0x5F, '_'), (0x60, '\u{F8E5}'), (0x61, '\u{03B1}'),
+    (0x62, '\u{03B2}'), (0x63, '\u{03C7}'), (0x64, '\u{03B4}'), (0x65, '\u{03B5}'), (0x66, '\u{03C6}'), (0x67, '\u{03B3}'),
+    (0x68, '\u{03B7}'), (0x69, '\u{03B9}'), (0x6A, '\u{03D5}'), (0x6B, '\u{03BA}'), (0x6C, '\u{03BB}'), (0x6D, '\u{00B5}'),
+    (0x6E, '\u{03BD}'), (0x6F, '\u{03BF}'), (0x70, '\u{03C0}'), (0x71, '\u{03B8}'), (0x72, '\u{03C1}'), (0x73, '\u{03C3}'),
+    (0x74, '\u{03C4}'), (0x75, '\u{03C5}'), (0x76, '\u{03D6}'), (0x77, '\u{03C9}'), (0x78, '\u{03BE}'), (0x79, '\u{03C8}'),
+    (0x7A, '\u{03B6}'), (0x7B, '{'), (0x7C, '|'), (0x7D, '}'), (0x7E, '\u{223C}'), (0xA0, '\u{20AC}'),
+    (0xA1, '\u{03D2}'), (0xA2, '\u{2032}'), (0xA3, '\u{2264}'), (0xA4, '\u{2044}'), (0xA5, '\u{221E}'), (0xA6, '\u{0192}'),
+    (0xA7, '\u{2663}'), (0xA8, '\u{2666}'), (0xA9, '\u{2665}'), (0xAA, '\u{2660}'), (0xAB, '\u{2194}'), (0xAC, '\u{2190}'),
+    (0xAD, '\u{2191}'), (0xAE, '\u{2192}'), (0xAF, '\u{2193}'), (0xB0, '\u{00B0}'), (0xB1, '\u{00B1}'), (0xB2, '\u{2033}'),
+    (0xB3, '\u{2265}'), (0xB4, '\u{00D7}'), (0xB5, '\u{221D}'), (0xB6, '\u{2202}'), (0xB7, '\u{2022}'), (0xB8, '\u{00F7}'),
+    (0xB9, '\u{2260}'), (0xBA, '\u{2261}'), (0xBB, '\u{2248}'), (0xBC, '\u{2026}'), (0xBD, '\u{F8E6}'), (0xBE, '\u{F8E7}'),
+    (0xBF, '\u{21B5}'), (0xC0, '\u{2135}'), (0xC1, '\u{2111}'), (0xC2, '\u{211C}'), (0xC3, '\u{2118}'), (0xC4, '\u{2297}'),
+    (0xC5, '\u{2295}'), (0xC6, '\u{2205}'), (0xC7, '\u{2229}'), (0xC8, '\u{222A}'), (0xC9, '\u{2283}'), (0xCA, '\u{2287}'),
+    (0xCB, '\u{2284}'), (0xCC, '\u{2282}'), (0xCD, '\u{2286}'), (0xCE, '\u{2208}'), (0xCF, '\u{2209}'), (0xD0, '\u{2220}'),
+    (0xD1, '\u{2207}'), (0xD2, '\u{F6DA}'), (0xD3, '\u{F6D9}'), (0xD4, '\u{F6DB}'), (0xD5, '\u{220F}'), (0xD6, '\u{221A}'),
+    (0xD7, '\u{22C5}'), (0xD8, '\u{00AC}'), (0xD9, '\u{2227}'), (0xDA, '\u{2228}'), (0xDB, '\u{21D4}'), (0xDC, '\u{21D0}'),
+    (0xDD, '\u{21D1}'), (0xDE, '\u{21D2}'), (0xDF, '\u{21D3}'), (0xE0, '\u{25CA}'), (0xE1, '\u{2329}'), (0xE2, '\u{F8E8}'),
+    (0xE3, '\u{F8E9}'), (0xE4, '\u{F8EA}'), (0xE5, '\u{2211}'), (0xE6, '\u{F8EB}'), (0xE7, '\u{F8EC}'), (0xE8, '\u{F8ED}'),
+    (0xE9, '\u{F8EE}'), (0xEA, '\u{F8EF}'), (0xEB, '\u{F8F0}'), (0xEC, '\u{F8F1}'), (0xED, '\u{F8F2}'), (0xEE, '\u{F8F3}'),
+    (0xEF, '\u{F8F4}'), (0xF1, '\u{232A}'), (0xF2, '\u{222B}'), (0xF3, '\u{2320}'), (0xF4, '\u{F8F5}'), (0xF5, '\u{2321}'),
+    (0xF6, '\u{F8F6}'), (0xF7, '\u{F8F7}'), (0xF8, '\u{F8F8}'), (0xF9, '\u{F8F9}'), (0xFA, '\u{F8FA}'), (0xFB, '\u{F8FB}'),
+    (0xFC, '\u{F8FC}'), (0xFD, '\u{F8FD}'), (0xFE, '\u{F8FE}'),
+];
+
+#[rustfmt::skip]
+static ZAPFDINGBATS_ENCODING: &[(u8, char)] = &[
+    (0x20, ' '), (0x21, '\u{2701}'), (0x22, '\u{2702}'), (0x23, '\u{2703}'), (0x24, '\u{2704}'), (0x25, '\u{260E}'),
+    (0x26, '\u{2706}'), (0x27, '\u{2707}'), (0x28, '\u{2708}'), (0x29, '\u{2709}'), (0x2A, '\u{261B}'), (0x2B, '\u{261E}'),
+    (0x2C, '\u{270C}'), (0x2D, '\u{270D}'), (0x2E, '\u{270E}'), (0x2F, '\u{270F}'), (0x30, '\u{2710}'), (0x31, '\u{2711}'),
+    (0x32, '\u{2712}'), (0x33, '\u{2713}'), (0x34, '\u{2714}'), (0x35, '\u{2715}'), (0x36, '\u{2716}'), (0x37, '\u{2717}'),
+    (0x38, '\u{2718}'), (0x39, '\u{2719}'), (0x3A, '\u{271A}'), (0x3B, '\u{271B}'), (0x3C, '\u{271C}'), (0x3D, '\u{271D}'),
+    (0x3E, '\u{271E}'), (0x3F, '\u{271F}'), (0x40, '\u{2720}'), (0x41, '\u{2721}'), (0x42, '\u{2722}'), (0x43, '\u{2723}'),
+    (0x44, '\u{2724}'), (0x45, '\u{2725}'), (0x46, '\u{2726}'), (0x47, '\u{2727}'), (0x48, '\u{2605}'), (0x49, '\u{2729}'),
+    (0x4A, '\u{272A}'), (0x4B, '\u{272B}'), (0x4C, '\u{272C}'), (0x4D, '\u{272D}'), (0x4E, '\u{272E}'), (0x4F, '\u{272F}'),
+    (0x50, '\u{2730}'), (0x51, '\u{2731}'), (0x52, '\u{2732}'), (0x53, '\u{2733}'), (0x54, '\u{2734}'), (0x55, '\u{2735}'),
+    (0x56, '\u{2736}'), (0x57, '\u{2737}'), (0x58, '\u{2738}'), (0x59, '\u{2739}'), (0x5A, '\u{273A}'), (0x5B, '\u{273B}'),
+    (0x5C, '\u{273C}'), (0x5D, '\u{273D}'), (0x5E, '\u{273E}'), (0x5F, '\u{273F}'), (0x60, '\u{2740}'), (0x61, '\u{2741}'),
+    (0x62, '\u{2742}'), (0x63, '\u{2743}'), (0x64, '\u{2744}'), (0x65, '\u{2745}'), (0x66, '\u{2746}'), (0x67, '\u{2747}'),
+    (0x68, '\u{2748}'), (0x69, '\u{2749}'), (0x6A, '\u{274A}'), (0x6B, '\u{274B}'), (0x6C, '\u{25CF}'), (0x6D, '\u{274D}'),
+    (0x6E, '\u{25A0}'), (0x6F, '\u{274F}'), (0x70, '\u{2750}'), (0x71, '\u{2751}'), (0x72, '\u{2752}'), (0x73, '\u{25B2}'),
+    (0x74, '\u{25BC}'), (0x75, '\u{25C6}'), (0x76, '\u{2756}'), (0x77, '\u{25D7}'), (0x78, '\u{2758}'), (0x79, '\u{2759}'),
+    (0x7A, '\u{275A}'), (0x7B, '\u{275B}'), (0x7C, '\u{275C}'), (0x7D, '\u{275D}'), (0x7E, '\u{275E}'), (0x80, '\u{2768}'),
+    (0x81, '\u{2769}'), (0x82, '\u{276A}'), (0x83, '\u{276B}'), (0x84, '\u{276C}'), (0x85, '\u{276D}'), (0x86, '\u{276E}'),
+    (0x87, '\u{276F}'), (0x88, '\u{2770}'), (0x89, '\u{2771}'), (0x8A, '\u{2772}'), (0x8B, '\u{2773}'), (0x8C, '\u{2774}'),
+    (0x8D, '\u{2775}'), (0xA1, '\u{2761}'), (0xA2, '\u{2762}'), (0xA3, '\u{2763}'), (0xA4, '\u{2764}'), (0xA5, '\u{2765}'),
+    (0xA6, '\u{2766}'), (0xA7, '\u{2767}'), (0xA8, '\u{2663}'), (0xA9, '\u{2666}'), (0xAA, '\u{2665}'), (0xAB, '\u{2660}'),
+    (0xAC, '\u{2460}'), (0xAD, '\u{2461}'), (0xAE, '\u{2462}'), (0xAF, '\u{2463}'), (0xB0, '\u{2464}'), (0xB1, '\u{2465}'),
+    (0xB2, '\u{2466}'), (0xB3, '\u{2467}'), (0xB4, '\u{2468}'), (0xB5, '\u{2469}'), (0xB6, '\u{2776}'), (0xB7, '\u{2777}'),
+    (0xB8, '\u{2778}'), (0xB9, '\u{2779}'), (0xBA, '\u{277A}'), (0xBB, '\u{277B}'), (0xBC, '\u{277C}'), (0xBD, '\u{277D}'),
+    (0xBE, '\u{277E}'), (0xBF, '\u{277F}'), (0xC0, '\u{2780}'), (0xC1, '\u{2781}'), (0xC2, '\u{2782}'), (0xC3, '\u{2783}'),
+    (0xC4, '\u{2784}'), (0xC5, '\u{2785}'), (0xC6, '\u{2786}'), (0xC7, '\u{2787}'), (0xC8, '\u{2788}'), (0xC9, '\u{2789}'),
+    (0xCA, '\u{278A}'), (0xCB, '\u{278B}'), (0xCC, '\u{278C}'), (0xCD, '\u{278D}'), (0xCE, '\u{278E}'), (0xCF, '\u{278F}'),
+    (0xD0, '\u{2790}'), (0xD1, '\u{2791}'), (0xD2, '\u{2792}'), (0xD3, '\u{2793}'), (0xD4, '\u{2794}'), (0xD5, '\u{2192}'),
+    (0xD6, '\u{2194}'), (0xD7, '\u{2195}'), (0xD8, '\u{2798}'), (0xD9, '\u{2799}'), (0xDA, '\u{279A}'), (0xDB, '\u{279B}'),
+    (0xDC, '\u{279C}'), (0xDD, '\u{279D}'), (0xDE, '\u{279E}'), (0xDF, '\u{279F}'), (0xE0, '\u{27A0}'), (0xE1, '\u{27A1}'),
+    (0xE2, '\u{27A2}'), (0xE3, '\u{27A3}'), (0xE4, '\u{27A4}'), (0xE5, '\u{27A5}'), (0xE6, '\u{27A6}'), (0xE7, '\u{27A7}'),
+    (0xE8, '\u{27A8}'), (0xE9, '\u{27A9}'), (0xEA, '\u{27AA}'), (0xEB, '\u{27AB}'), (0xEC, '\u{27AC}'), (0xED, '\u{27AD}'),
+    (0xEE, '\u{27AE}'), (0xEF, '\u{27AF}'), (0xF1, '\u{27B1}'), (0xF2, '\u{27B2}'), (0xF3, '\u{27B3}'), (0xF4, '\u{27B4}'),
+    (0xF5, '\u{27B5}'), (0xF6, '\u{27B6}'), (0xF7, '\u{27B7}'), (0xF8, '\u{27B8}'), (0xF9, '\u{27B9}'), (0xFA, '\u{27BA}'),
+    (0xFB, '\u{27BB}'), (0xFC, '\u{27BC}'), (0xFD, '\u{27BD}'), (0xFE, '\u{27BE}'),
+];
+
+/// Every width table, for exhaustive testing.
+#[cfg(test)]
+static ALL_TABLES: &[(&str, &[(char, u16)])] = &[
+    ("COURIER", COURIER),
+    ("COURIER_BOLD", COURIER_BOLD),
+    ("COURIER_OBLIQUE", COURIER_OBLIQUE),
+    ("COURIER_BOLDOBLIQUE", COURIER_BOLDOBLIQUE),
+    ("HELVETICA", HELVETICA),
+    ("HELVETICA_BOLD", HELVETICA_BOLD),
+    ("HELVETICA_OBLIQUE", HELVETICA_OBLIQUE),
+    ("HELVETICA_BOLDOBLIQUE", HELVETICA_BOLDOBLIQUE),
+    ("TIMES_ROMAN", TIMES_ROMAN),
+    ("TIMES_BOLD", TIMES_BOLD),
+    ("TIMES_ITALIC", TIMES_ITALIC),
+    ("TIMES_BOLDITALIC", TIMES_BOLDITALIC),
+    ("SYMBOL", SYMBOL),
+    ("ZAPFDINGBATS", ZAPFDINGBATS),
+];
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn times_roman_ascii_widths() {
+        assert_eq!(base14_char_width("Times-Roman", ' '), Some(250));
+        assert_eq!(base14_char_width("Times-Roman", 'M'), Some(889));
+        assert_eq!(base14_char_width("Times-Roman", 'i'), Some(278));
+    }
+
+    #[test]
+    fn subset_prefix_and_aliases_normalize() {
+        assert_eq!(
+            base14_char_width("ABCDEF+Times-Bold", ' '),
+            base14_char_width("Times-Bold", ' ')
+        );
+        assert!(base14_char_width("ArialMT", 'a').is_some());
+        assert!(base14_char_width("TimesNewRomanPSMT", 'a').is_some());
+    }
+
+    #[test]
+    fn non_base14_returns_none() {
+        assert_eq!(base14_char_width("DejaVuSans", 'a'), None);
+        assert!(!is_base14_font("Garamond"));
+    }
+
+    #[test]
+    fn builtin_encoding_resolves_symbol_and_zapf_codes() {
+        // Symbol 0x61 renders alpha; Zapf 0x21 renders U+2701.
+        assert_eq!(builtin_encoding_char("Symbol", 0x61), Some('\u{03B1}'));
+        assert_eq!(builtin_encoding_char("Symbol", 0xA5), Some('\u{221E}'));
+        assert_eq!(
+            builtin_encoding_char("ZapfDingbats", 0x21),
+            Some('\u{2701}')
+        );
+        // Latin text fonts follow standard encodings — no builtin override.
+        assert_eq!(builtin_encoding_char("Times-Roman", 0x61), None);
+        // The resolved chars have real AFM widths.
+        let alpha_w = base14_char_width("Symbol", '\u{03B1}');
+        assert!(alpha_w.is_some() && alpha_w != Some(500));
+    }
+
+    #[test]
+    fn encoding_tables_are_sorted_for_binary_search() {
+        for table in [SYMBOL_ENCODING, ZAPFDINGBATS_ENCODING] {
+            assert!(table.windows(2).all(|w| w[0].0 < w[1].0));
+        }
+    }
+
+    #[test]
+    fn tables_are_sorted_for_binary_search() {
+        // Every table is queried by binary search, so all of them must be
+        // sorted — not just a sample.
+        for (name, table) in ALL_TABLES {
+            assert!(
+                table.windows(2).all(|w| w[0].0 < w[1].0),
+                "{name} is not sorted"
+            );
+        }
+    }
+}

src/extractor/content_stream.rs

@@ -14,9 +14,11 @@ use lopdf::{Document, Encoding, Object, ObjectId};
 use std::collections::HashMap;
 
 use super::fonts::{
-    build_font_encodings, build_font_widths, compute_string_width_ts, extract_text_from_operand,
-    get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    build_font_encodings, build_font_widths, build_type3_scales, compute_string_width_ts,
+    descriptor_style_flags, extract_text_from_operand, get_font_file2_obj_num, get_operand_bytes,
+    CMapDecisionCache, FontStyleCache,
 };
+use super::underline::UnderlineLine;
 use super::xobjects::{extract_form_xobject_text, get_page_xobjects, XObjectType};
 use super::{get_number, image_bbox_from_ctm, multiply_matrices};
 
@@ -39,6 +41,17 @@ fn strip_pdf_comments(data: &[u8]) -> Vec<u8> {
     while i < data.len() {
         let b = data[i];
         match b {
+            // Inside a string literal, a backslash escapes the next byte —
+            // `\(`, `\)`, and `\\` must not touch the nesting depth, or a
+            // later `%` glyph inside a string gets stripped as a comment,
+            // corrupting the stream.
+            b'\\' if in_string > 0 => {
+                result.push(b);
+                if let Some(&next) = data.get(i + 1) {
+                    result.push(next);
+                    i += 1;
+                }
+            }
             b'(' if !in_hex_string => {
                 in_string += 1;
                 result.push(b);
@@ -74,6 +87,56 @@ fn strip_pdf_comments(data: &[u8]) -> Vec<u8> {
     result
 }
 
+fn transform_path_point(x: f32, y: f32, ctm: &[f32; 6]) -> (f32, f32) {
+    (
+        x * ctm[0] + y * ctm[2] + ctm[4],
+        x * ctm[1] + y * ctm[3] + ctm[5],
+    )
+}
+
+fn transformed_stroke_width(
+    line_width: f32,
+    ctm: &[f32; 6],
+    x1: f32,
+    y1: f32,
+    x2: f32,
+    y2: f32,
+) -> f32 {
+    let user_width = line_width.abs();
+    let dx = x2 - x1;
+    let dy = y2 - y1;
+    let len = (dx * dx + dy * dy).sqrt();
+    if len <= f32::EPSILON {
+        return user_width;
+    }
+
+    // PDF stroke width scales perpendicular to the path direction.
+    let nx = -dy / len;
+    let ny = dx / len;
+    let ndx = nx * ctm[0] + ny * ctm[2];
+    let ndy = nx * ctm[1] + ny * ctm[3];
+    user_width * (ndx * ndx + ndy * ndy).sqrt()
+}
+
+/// Text rise (Ts) displaces the glyph origin by (0, rise) in unscaled text
+/// space — per the rendering-matrix definition it sits left of Tm, so the
+/// offset maps through the text matrix's y column. Rise never contributes
+/// to the advance, so callers apply it only to the rendering position and
+/// keep advancing the unshifted text matrix.
+fn rise_adjusted(tm: &[f32; 6], rise: f32) -> [f32; 6] {
+    if rise == 0.0 {
+        return *tm;
+    }
+    [
+        tm[0],
+        tm[1],
+        tm[2],
+        tm[3],
+        tm[4] + rise * tm[2],
+        tm[5] + rise * tm[3],
+    ]
+}
+
 /// Returns `(page_extraction, has_gid_fonts)` where `has_gid_fonts` indicates
 /// the page uses fonts with unresolvable gid-encoded glyphs.
 pub(crate) fn extract_page_text_items(
@@ -82,6 +145,7 @@ pub(crate) fn extract_page_text_items(
     page_num: u32,
     font_cmaps: &FontCMaps,
     include_invisible: bool,
+    style_cache: &mut FontStyleCache,
 ) -> Result<(PageExtraction, bool, bool), PdfError> {
     use lopdf::content::Content;
 
@@ -89,6 +153,7 @@ pub(crate) fn extract_page_text_items(
     let mut rects: Vec<PdfRect> = Vec::new();
     let mut clip_rects: Vec<PdfRect> = Vec::new();
     let mut lines: Vec<PdfLine> = Vec::new();
+    let mut underline_lines: Vec<UnderlineLine> = Vec::new();
 
     // Path construction state for m/l/h → S/s line extraction
     let mut path_subpath_start: Option<(f32, f32)> = None;
@@ -97,15 +162,22 @@ pub(crate) fn extract_page_text_items(
     // Completed subpaths (each a vec of line segments) for f/f* rect extraction
     let mut pending_subpaths: Vec<Vec<(f32, f32, f32, f32)>> = Vec::new();
     let mut fill_rects: Vec<PdfRect> = Vec::new();
+    // `re` rects awaiting a paint operator. Underline detection must only
+    // see painted rects: a `re W n` clip path or `re n` no-op draws nothing
+    // on the page, so treating every `re` as ink would underline text that
+    // merely sits near an invisible clip boundary.
+    let mut pending_re_rects: Vec<PdfRect> = Vec::new();
+    let mut painted_rects: Vec<PdfRect> = Vec::new();
 
     // Get fonts for encoding
     let fonts = doc.get_page_fonts(page_id).unwrap_or_default();
 
     // Build font encoding maps from Differences arrays
-    let (font_encodings, has_gid_fonts) = build_font_encodings(doc, &fonts);
+    let (font_encodings, has_gid_fonts) = build_font_encodings(doc, &fonts, font_cmaps);
 
     // Build font width info for accurate text positioning
     let font_widths = build_font_widths(doc, &fonts);
+    let type3_scales = build_type3_scales(doc, &fonts);
 
     // Build maps of font resource names to their base font names and ToUnicode object refs
     let mut font_base_names: std::collections::HashMap<String, String> =
@@ -114,6 +186,8 @@ pub(crate) fn extract_page_text_items(
         std::collections::HashMap::new();
     let mut inline_cmaps: std::collections::HashMap<String, crate::tounicode::CMapEntry> =
         std::collections::HashMap::new();
+    let mut font_style_flags: std::collections::HashMap<String, (bool, bool)> =
+        std::collections::HashMap::new();
     for (font_name, font_dict) in &fonts {
         let resource_name = String::from_utf8_lossy(font_name).to_string();
         if let Ok(base_font) = font_dict.get(b"BaseFont") {
@@ -122,6 +196,12 @@ pub(crate) fn extract_page_text_items(
                 font_base_names.insert(resource_name.clone(), base_name);
             }
         }
+        // Descriptor style flags rescue subset fonts whose BaseFont names
+        // are opaque tags the name heuristics can't read.
+        let style = descriptor_style_flags(doc, font_dict, style_cache);
+        if style != (false, false) {
+            font_style_flags.insert(resource_name.clone(), style);
+        }
         // Track ToUnicode object reference, with FontFile2 fallback for Identity-H/V.
         // Also handle inline ToUnicode streams.
         match font_dict.get(b"ToUnicode") {
@@ -188,7 +268,20 @@ pub(crate) fn extract_page_text_items(
     // Graphics state tracking
     let mut ctm = [1.0f32, 0.0, 0.0, 1.0, 0.0, 0.0]; // Current Transformation Matrix
     let mut text_rendering_mode: i32 = 0; // 0=fill, 1=stroke, 2=fill+stroke, 3=invisible
-    let mut gstate_stack: Vec<([f32; 6], i32, f32, f32)> = Vec::new();
+    let mut line_width: f32 = 1.0;
+    #[derive(Clone)]
+    struct SavedGraphicsState {
+        ctm: [f32; 6],
+        text_rendering_mode: i32,
+        line_width: f32,
+        char_spacing: f32,
+        word_spacing: f32,
+        text_rise: f32,
+        text_leading: f32,
+        current_font: String,
+        current_font_size: f32,
+    }
+    let mut gstate_stack: Vec<SavedGraphicsState> = Vec::new();
 
     // Text state tracking
     let mut current_font = String::new();
@@ -196,6 +289,7 @@ pub(crate) fn extract_page_text_items(
     let mut text_leading: f32 = 0.0; // TL parameter (in text-space units)
     let mut char_spacing: f32 = 0.0; // Tc parameter (extra spacing per character, unscaled)
     let mut word_spacing: f32 = 0.0; // Tw parameter (extra spacing per space char, unscaled)
+    let mut text_rise: f32 = 0.0; // Ts parameter (baseline shift for super/subscripts, unscaled)
     let mut text_matrix = [1.0f32, 0.0, 0.0, 1.0, 0.0, 0.0];
     let mut line_matrix = [1.0f32, 0.0, 0.0, 1.0, 0.0, 0.0];
     let mut in_text_block = false;
@@ -217,6 +311,10 @@ pub(crate) fn extract_page_text_items(
     let mut suppress_glyph_extraction = false;
     let mut actual_text_start_tm: Option<[f32; 6]> = None; // text matrix at BDC entry
     let mut actual_text_glyph_tm: Option<[f32; 6]> = None; // text matrix at first glyph inside BDC
+                                                           // Text rise in effect at each captured matrix — the item must render at
+                                                           // the rise of its GLYPHS, not whatever rise is set by EMC time.
+    let mut actual_text_start_rise: f32 = 0.0;
+    let mut actual_text_glyph_rise: Option<f32> = None;
     /// Get the innermost MCID from the marked content stack.
     fn current_mcid(stack: &[MarkedContentEntry]) -> Option<i64> {
         stack.iter().rev().find_map(|e| e.mcid)
@@ -227,15 +325,30 @@ pub(crate) fn extract_page_text_items(
         match op.operator.as_str() {
             "q" => {
                 // Save graphics state
-                gstate_stack.push((ctm, text_rendering_mode, char_spacing, word_spacing));
+                gstate_stack.push(SavedGraphicsState {
+                    ctm,
+                    text_rendering_mode,
+                    line_width,
+                    char_spacing,
+                    word_spacing,
+                    text_rise,
+                    text_leading,
+                    current_font: current_font.clone(),
+                    current_font_size,
+                });
             }
             "Q" => {
                 // Restore graphics state
-                if let Some((saved_ctm, saved_tr, saved_tc, saved_tw)) = gstate_stack.pop() {
-                    ctm = saved_ctm;
-                    text_rendering_mode = saved_tr;
-                    char_spacing = saved_tc;
-                    word_spacing = saved_tw;
+                if let Some(saved) = gstate_stack.pop() {
+                    ctm = saved.ctm;
+                    text_rendering_mode = saved.text_rendering_mode;
+                    line_width = saved.line_width;
+                    char_spacing = saved.char_spacing;
+                    word_spacing = saved.word_spacing;
+                    text_rise = saved.text_rise;
+                    text_leading = saved.text_leading;
+                    current_font = saved.current_font;
+                    current_font_size = saved.current_font_size;
                 }
             }
             "cm" => {
@@ -252,6 +365,11 @@ pub(crate) fn extract_page_text_items(
                     ctm = multiply_matrices(&new_matrix, &ctm);
                 }
             }
+            "w" => {
+                if let Some(width) = op.operands.first().and_then(get_number) {
+                    line_width = width;
+                }
+            }
             "BT" => {
                 // Begin text block
                 in_text_block = true;
@@ -300,6 +418,12 @@ pub(crate) fn extract_page_text_items(
                     word_spacing = tw;
                 }
             }
+            "Ts" => {
+                // Set text rise (baseline shift for superscripts/subscripts)
+                if let Some(ts) = op.operands.first().and_then(get_number) {
+                    text_rise = ts;
+                }
+            }
             "Td" | "TD" => {
                 // Move text position: TLM = T(tx,ty) × TLM; Tm = TLM
                 // tx,ty are in text space — must be scaled by the text line matrix
@@ -358,6 +482,7 @@ pub(crate) fn extract_page_text_items(
                     if suppress_glyph_extraction {
                         if actual_text_glyph_tm.is_none() {
                             actual_text_glyph_tm = Some(text_matrix);
+                            actual_text_glyph_rise = Some(text_rise);
                         }
                         if let Some(w_ts) = w_ts_opt {
                             text_matrix[4] += w_ts * text_matrix[0];
@@ -387,8 +512,10 @@ pub(crate) fn extract_page_text_items(
                         &mut cmap_decisions,
                         &font_widths,
                     ) {
-                        let combined = multiply_matrices(&text_matrix, &ctm);
-                        let rendered_size = effective_font_size(current_font_size, &combined);
+                        let combined =
+                            multiply_matrices(&rise_adjusted(&text_matrix, text_rise), &ctm);
+                        let rendered_size = effective_font_size(current_font_size, &combined)
+                            * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                         let (x, y) = (combined[4], combined[5]);
                         if combined[0].abs() >= combined[1].abs() {
                             rotation_votes.horizontal += 1;
@@ -409,6 +536,10 @@ pub(crate) fn extract_page_text_items(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
@@ -418,8 +549,10 @@ pub(crate) fn extract_page_text_items(
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
+                                is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: current_mcid(&marked_content_stack),
                             });
@@ -437,6 +570,7 @@ pub(crate) fn extract_page_text_items(
                         // Capture first-glyph position for ActualText
                         if suppress_glyph_extraction && actual_text_glyph_tm.is_none() {
                             actual_text_glyph_tm = Some(text_matrix);
+                            actual_text_glyph_rise = Some(text_rise);
                         }
 
                         // Compute space threshold based on font metrics when available
@@ -552,11 +686,16 @@ pub(crate) fn extract_page_text_items(
                             } else {
                                 rotation_votes.rotated += 1;
                             }
-                            let rendered_size = effective_font_size(current_font_size, &combined);
+                            let rendered_size = effective_font_size(current_font_size, &combined)
+                                * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                             let base_font = font_base_names
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             let scale_x = text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2];
                             for (text, start_w, end_w) in &sub_items {
                                 let offset_tm = [
@@ -567,7 +706,8 @@ pub(crate) fn extract_page_text_items(
                                     text_matrix[4] + start_w * text_matrix[0],
                                     text_matrix[5] + start_w * text_matrix[1],
                                 ];
-                                let combined = multiply_matrices(&offset_tm, &ctm);
+                                let combined =
+                                    multiply_matrices(&rise_adjusted(&offset_tm, text_rise), &ctm);
                                 let (x, y) = (combined[4], combined[5]);
                                 let width = if font_info.is_some() {
                                     ((end_w - start_w) * scale_x).abs()
@@ -583,8 +723,10 @@ pub(crate) fn extract_page_text_items(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
+                                    is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: current_mcid(&marked_content_stack),
                                 });
@@ -608,6 +750,26 @@ pub(crate) fn extract_page_text_items(
                 line_matrix[4] += (-tl) * line_matrix[2];
                 line_matrix[5] += (-tl) * line_matrix[3];
                 text_matrix = line_matrix;
+                // Capture first-glyph position for ActualText AFTER the
+                // line move — the BDC-entry matrix is on the previous line.
+                if suppress_glyph_extraction && actual_text_glyph_tm.is_none() {
+                    actual_text_glyph_tm = Some(text_matrix);
+                    actual_text_glyph_rise = Some(text_rise);
+                }
+                // Advance width, as for Tj — without it the item stays
+                // zero-width and geometric underline/strikeout detection
+                // rejects it (`is_underline_candidate` needs width > 0).
+                let w_ts_opt = font_widths.get(&current_font).and_then(|fi| {
+                    op.operands.first().and_then(get_operand_bytes).map(|raw| {
+                        compute_string_width_ts(
+                            raw,
+                            fi,
+                            current_font_size,
+                            char_spacing,
+                            word_spacing,
+                        )
+                    })
+                });
                 if !((text_rendering_mode == 3 && !include_invisible)
                     || suppress_glyph_extraction
                     || op.operands.is_empty())
@@ -625,35 +787,55 @@ pub(crate) fn extract_page_text_items(
                         &font_widths,
                     ) {
                         if !text.trim().is_empty() {
-                            let combined = multiply_matrices(&text_matrix, &ctm);
+                            let combined =
+                                multiply_matrices(&rise_adjusted(&text_matrix, text_rise), &ctm);
                             if combined[0].abs() >= combined[1].abs() {
                                 rotation_votes.horizontal += 1;
                             } else {
                                 rotation_votes.rotated += 1;
                             }
-                            let rendered_size = effective_font_size(current_font_size, &combined);
+                            let rendered_size = effective_font_size(current_font_size, &combined)
+                                * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                             let (x, y) = (combined[4], combined[5]);
+                            let width = w_ts_opt
+                                .map(|w_ts| {
+                                    (w_ts * (text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2]))
+                                        .abs()
+                                })
+                                .unwrap_or(0.0);
                             let base_font = font_base_names
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
                                 y,
-                                width: 0.0,
+                                width,
                                 height: rendered_size,
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
+                                is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: current_mcid(&marked_content_stack),
                             });
                         }
                     }
                 }
+                // Advance regardless of visibility so later show-text
+                // operators on the same line stay positioned (as for Tj).
+                if let Some(w_ts) = w_ts_opt {
+                    text_matrix[4] += w_ts * text_matrix[0];
+                    text_matrix[5] += w_ts * text_matrix[1];
+                }
             }
             "Do" => {
                 // XObject invocation - could be an image or form
@@ -684,6 +866,8 @@ pub(crate) fn extract_page_text_items(
                                         page: page_num,
                                         is_bold: false,
                                         is_italic: false,
+                                        is_underline: false,
+                                        is_strikeout: false,
                                         item_type: ItemType::Image,
                                         mcid: current_mcid(&marked_content_stack),
                                     });
@@ -697,6 +881,7 @@ pub(crate) fn extract_page_text_items(
                                         font_cmaps,
                                         &ctm,
                                         &mut cmap_decisions,
+                                        style_cache,
                                     );
                                     items.extend(form_items);
                                 }
@@ -737,7 +922,9 @@ pub(crate) fn extract_page_text_items(
                 if actual_text.is_some() {
                     suppress_glyph_extraction = true;
                     actual_text_start_tm = Some(text_matrix);
+                    actual_text_start_rise = text_rise;
                     actual_text_glyph_tm = None; // reset — will be captured at first Tj/TJ
+                    actual_text_glyph_rise = None;
                 }
                 marked_content_stack.push(MarkedContentEntry { actual_text, mcid });
             }
@@ -750,15 +937,18 @@ pub(crate) fn extract_page_text_items(
                         // Tj may have moved the text position to the correct line —
                         // the BDC-entry position can be on the previous line.
                         let glyph_tm = actual_text_glyph_tm.take();
+                        let glyph_rise = actual_text_glyph_rise.take();
                         let entry_tm = actual_text_start_tm.take();
                         if let Some(start_tm) = glyph_tm.or(entry_tm) {
-                            let combined = multiply_matrices(&start_tm, &ctm);
+                            let rise = glyph_rise.unwrap_or(actual_text_start_rise);
+                            let combined = multiply_matrices(&rise_adjusted(&start_tm, rise), &ctm);
                             if combined[0].abs() >= combined[1].abs() {
                                 rotation_votes.horizontal += 1;
                             } else {
                                 rotation_votes.rotated += 1;
                             }
-                            let rendered_size = effective_font_size(current_font_size, &combined);
+                            let rendered_size = effective_font_size(current_font_size, &combined)
+                                * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                             let (x, y) = (combined[4], combined[5]);
                             // Width in device space from text matrix delta
                             let delta_ts = text_matrix[4] - start_tm[4];
@@ -769,6 +959,10 @@ pub(crate) fn extract_page_text_items(
                                     .get(&current_font)
                                     .map(|s| s.as_str())
                                     .unwrap_or(&current_font);
+                                let (desc_italic, desc_bold) = font_style_flags
+                                    .get(&current_font)
+                                    .copied()
+                                    .unwrap_or((false, false));
                                 items.push(TextItem {
                                     text: expand_ligatures(&at),
                                     x,
@@ -778,8 +972,10 @@ pub(crate) fn extract_page_text_items(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
+                                    is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: entry
                                         .mcid
@@ -804,13 +1000,19 @@ pub(crate) fn extract_page_text_items(
                     let y_dev = rx * ctm[1] + ry * ctm[3] + ctm[5];
                     let w_dev = rw * ctm[0];
                     let h_dev = rh * ctm[3];
-                    rects.push(PdfRect {
+                    let rect = PdfRect {
                         x: x_dev,
                         y: y_dev,
                         width: w_dev,
                         height: h_dev,
                         page: page_num,
-                    });
+                    };
+                    // Underline detection must only see rects that are
+                    // actually painted — a `re` used purely as a clip path
+                    // (`re W n`) or discarded (`re n`) draws nothing. Hold
+                    // the rect as pending until a paint operator confirms it.
+                    pending_re_rects.push(rect.clone());
+                    rects.push(rect);
                 }
             }
             // ── Path construction operators ──────────────────────
@@ -860,10 +1062,8 @@ pub(crate) fn extract_page_text_items(
                     }
                 }
                 for (x1, y1, x2, y2) in pending_lines.drain(..) {
-                    let x1d = x1 * ctm[0] + y1 * ctm[2] + ctm[4];
-                    let y1d = x1 * ctm[1] + y1 * ctm[3] + ctm[5];
-                    let x2d = x2 * ctm[0] + y2 * ctm[2] + ctm[4];
-                    let y2d = x2 * ctm[1] + y2 * ctm[3] + ctm[5];
+                    let (x1d, y1d) = transform_path_point(x1, y1, &ctm);
+                    let (x2d, y2d) = transform_path_point(x2, y2, &ctm);
                     lines.push(PdfLine {
                         x1: x1d,
                         y1: y1d,
@@ -871,7 +1071,16 @@ pub(crate) fn extract_page_text_items(
                         y2: y2d,
                         page: page_num,
                     });
+                    underline_lines.push(UnderlineLine {
+                        x1: x1d,
+                        y1: y1d,
+                        x2: x2d,
+                        y2: y2d,
+                        stroke_width: transformed_stroke_width(line_width, &ctm, x1, y1, x2, y2),
+                        page: page_num,
+                    });
                 }
+                painted_rects.append(&mut pending_re_rects);
                 pending_subpaths.clear();
                 path_subpath_start = None;
                 path_current = None;
@@ -887,10 +1096,8 @@ pub(crate) fn extract_page_text_items(
                     }
                 }
                 for (x1, y1, x2, y2) in pending_lines.drain(..) {
-                    let x1d = x1 * ctm[0] + y1 * ctm[2] + ctm[4];
-                    let y1d = x1 * ctm[1] + y1 * ctm[3] + ctm[5];
-                    let x2d = x2 * ctm[0] + y2 * ctm[2] + ctm[4];
-                    let y2d = x2 * ctm[1] + y2 * ctm[3] + ctm[5];
+                    let (x1d, y1d) = transform_path_point(x1, y1, &ctm);
+                    let (x2d, y2d) = transform_path_point(x2, y2, &ctm);
                     lines.push(PdfLine {
                         x1: x1d,
                         y1: y1d,
@@ -898,7 +1105,16 @@ pub(crate) fn extract_page_text_items(
                         y2: y2d,
                         page: page_num,
                     });
+                    underline_lines.push(UnderlineLine {
+                        x1: x1d,
+                        y1: y1d,
+                        x2: x2d,
+                        y2: y2d,
+                        stroke_width: transformed_stroke_width(line_width, &ctm, x1, y1, x2, y2),
+                        page: page_num,
+                    });
                 }
+                painted_rects.append(&mut pending_re_rects);
                 pending_subpaths.clear();
                 path_subpath_start = None;
                 path_current = None;
@@ -956,6 +1172,7 @@ pub(crate) fn extract_page_text_items(
                         }
                     }
                 }
+                painted_rects.append(&mut pending_re_rects);
                 pending_lines.clear();
                 path_subpath_start = None;
                 path_current = None;
@@ -1021,7 +1238,10 @@ pub(crate) fn extract_page_text_items(
                 // Do NOT clear pending_lines — the following `n` does that
             }
             "n" => {
-                // end path (no-op): discard
+                // end path (no-op): discard — including any `re` rects that
+                // were only ever part of a clip path (`re W n`), which draw
+                // no ink and must not feed underline detection.
+                pending_re_rects.clear();
                 pending_lines.clear();
                 pending_subpaths.clear();
                 path_subpath_start = None;
@@ -1031,6 +1251,12 @@ pub(crate) fn extract_page_text_items(
         }
     }
 
+    // Underline detection reads only painted ink: `re` rects confirmed by
+    // a paint operator plus filled-subpath rects — never clip-only rects,
+    // which draw nothing.
+    let mut underline_rects = painted_rects;
+    underline_rects.extend(fill_rects.iter().cloned());
+
     // Only use clip/fill rects when no `re` rects exist on this page.
     // Clip rects take priority over fill rects, but first we deduplicate
     // them: some PDFs wrap every text block in a full-page W* clip path,
@@ -1060,8 +1286,17 @@ pub(crate) fn extract_page_text_items(
     // Some PDFs embed landscape content in portrait pages using a rotated text
     // matrix (e.g. [0, b, -b, 0, tx, ty] for 90° CCW).  The layout engine
     // assumes x=horizontal, y=vertical — so we swap coordinates to match.
-    let (items, rects, lines, coords_rotated) =
+    let (mut items, rects, lines, coords_rotated) =
         correct_rotated_page(items, rects, lines, &rotation_votes);
+    if coords_rotated {
+        rotate_underline_graphics(&mut underline_rects, &mut underline_lines);
+    }
+    super::underline::mark_underlined_items(
+        &mut items,
+        &underline_rects,
+        &underline_lines,
+        page_num,
+    );
 
     let items = super::merge_text_items(items);
     let items = super::merge_subscript_items(items);
@@ -1147,6 +1382,27 @@ fn correct_rotated_page(
     (items, rects, lines, true)
 }
 
+fn rotate_underline_graphics(rects: &mut [PdfRect], lines: &mut [UnderlineLine]) {
+    for rect in rects {
+        let new_x = rect.y;
+        let new_y = -(rect.x + rect.width.abs());
+        rect.x = new_x;
+        rect.y = new_y;
+        std::mem::swap(&mut rect.width, &mut rect.height);
+    }
+
+    for line in lines {
+        let new_x1 = line.y1;
+        let new_y1 = -line.x1;
+        let new_x2 = line.y2;
+        let new_y2 = -line.x2;
+        line.x1 = new_x1;
+        line.y1 = new_y1;
+        line.x2 = new_x2;
+        line.y2 = new_y2;
+    }
+}
+
 /// Remove near-duplicate rects (same coordinates within 0.5 pt tolerance).
 /// Some PDFs emit a full-page clip path for every text block, producing
 /// thousands of identical rects. After dedup these collapse to one rect,
@@ -1196,6 +1452,64 @@ mod tests {
         }
     }
 
+    fn simple_doc_with_content(content: &[u8]) -> (lopdf::Document, lopdf::ObjectId) {
+        use lopdf::{dictionary, Object, Stream};
+
+        let mut doc = lopdf::Document::new();
+        let widths: Vec<Object> = (0..=255).map(|_| 600.into()).collect();
+        let font_id = doc.add_object(dictionary! {
+            "Type" => "Font",
+            "Subtype" => "Type1",
+            "BaseFont" => "Helvetica",
+            "FirstChar" => 0,
+            "LastChar" => 255,
+            "Widths" => Object::Array(widths),
+        });
+        let content_id = doc.add_object(Object::Stream(Stream::new(
+            dictionary! {},
+            content.to_vec(),
+        )));
+        let page_id = doc.add_object(dictionary! {
+            "Type" => "Page",
+            "Contents" => Object::Reference(content_id),
+            "Resources" => dictionary! {
+                "Font" => dictionary! {
+                    "F1" => Object::Reference(font_id),
+                },
+            },
+            "MediaBox" => vec![0.into(), 0.into(), 612.into(), 792.into()],
+        });
+        let pages_id = doc.add_object(dictionary! {
+            "Type" => "Pages",
+            "Count" => Object::Integer(1),
+            "Kids" => vec![Object::Reference(page_id)],
+        });
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "Pages" => Object::Reference(pages_id),
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        (doc, page_id)
+    }
+
+    fn extract_simple_items(content: &[u8]) -> Vec<TextItem> {
+        use crate::tounicode::FontCMaps;
+
+        let (doc, page_id) = simple_doc_with_content(content);
+        let font_cmaps = FontCMaps::from_doc(&doc);
+        let ((items, _, _), _, _) = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
+        items
+    }
+
     #[test]
     fn test_dedup_rects_identical() {
         let mut rects = vec![rect(0.0, 0.0, 612.0, 792.0, 1); 3759];
@@ -1245,6 +1559,140 @@ mod tests {
         assert_eq!(single.len(), 1);
     }
 
+    #[test]
+    fn thick_stroked_rule_does_not_mark_underline() {
+        let content = b"BT /F1 12 Tf 1 0 0 1 100 500 Tm (THICK) Tj ET
+4 w
+100 498 m 170 498 l S
+BT /F1 12 Tf 1 0 0 1 100 480 Tm (THIN) Tj ET
+1 w
+100 478 m 160 478 l S";
+
+        let items = extract_simple_items(content);
+        let thick = items.iter().find(|item| item.text == "THICK").unwrap();
+        let thin = items.iter().find(|item| item.text == "THIN").unwrap();
+
+        assert!(!thick.is_underline);
+        assert!(thin.is_underline);
+    }
+
+    #[test]
+    fn rotated_page_underline_is_detected_after_coordinate_correction() {
+        let content = b"BT /F1 12 Tf 0 1 -1 0 200 100 Tm (HELLO) Tj ET
+BT /F1 12 Tf 0 1 -1 0 240 100 Tm (WORLD) Tj ET
+1 w
+202 100 m 202 170 l S";
+
+        let items = extract_simple_items(content);
+        let hello = items.iter().find(|item| item.text == "HELLO").unwrap();
+        let world = items.iter().find(|item| item.text == "WORLD").unwrap();
+
+        assert!(hello.is_underline);
+        assert!(!world.is_underline);
+    }
+
+    #[test]
+    fn quote_operator_text_carries_advance_width() {
+        // `'` (move-to-next-line-and-show-text) must retain the string's
+        // advance width like Tj — zero-width items are invisible to
+        // geometric underline/strikeout detection.
+        let content = b"BT /F1 12 Tf 12 TL 1 0 0 1 100 512 Tm (first) Tj (struck) ' ET
+1 w
+99 503 m 145 503 l S";
+
+        let items = extract_simple_items(content);
+        let struck = items.iter().find(|item| item.text == "struck").unwrap();
+
+        // 6 glyphs x 600/1000 x 12pt = 43.2pt, drawn one leading below Tm.
+        assert!((struck.width - 43.2).abs() < 0.1);
+        assert!((struck.y - 500.0).abs() < 0.1);
+        assert!(struck.is_strikeout);
+        assert!(!struck.is_underline);
+    }
+
+    #[test]
+    fn quote_operator_advances_text_matrix() {
+        // Text shown after `'` on the same line must start past the shown
+        // string: "CD" lands at x=114.4 (2 glyphs x 600/1000 x 12pt after
+        // x=100), flush against "AB", so the merge pass joins them. Without
+        // the advance "CD" overlaps "AB" at x=100 and the items stay apart.
+        let content = b"BT /F1 12 Tf 12 TL 1 0 0 1 100 512 Tm (AB) ' (CD) Tj ET";
+
+        let items = extract_simple_items(content);
+        let merged = items.iter().find(|item| item.text == "ABCD").unwrap();
+
+        assert!((merged.x - 100.0).abs() < 0.1);
+        assert!((merged.width - 28.8).abs() < 0.1);
+        assert!((merged.y - 500.0).abs() < 0.1);
+    }
+
+    #[test]
+    fn text_rise_shifts_item_baseline() {
+        // Ts displaces the glyph origin vertically without touching the
+        // advance; the next run at rise 0 must return to the original
+        // baseline and follow the raised run horizontally.
+        let content =
+            b"BT /F1 12 Tf 1 0 0 1 100 500 Tm (base) Tj 5 Ts (super) Tj 0 Ts (after) Tj ET";
+
+        let items = extract_simple_items(content);
+        let base = items.iter().find(|item| item.text == "base").unwrap();
+        let raised = items.iter().find(|item| item.text == "super").unwrap();
+        let after = items.iter().find(|item| item.text == "after").unwrap();
+
+        assert!((base.y - 500.0).abs() < 0.1);
+        assert!((raised.y - 505.0).abs() < 0.1);
+        assert!((after.y - 500.0).abs() < 0.1);
+        assert!(after.x > raised.x);
+    }
+
+    #[test]
+    fn actual_text_item_uses_glyph_rise() {
+        // The ActualText replacement item must render at the rise in
+        // effect when its glyphs were drawn — not the unshifted BDC
+        // baseline, and not whatever rise is set by EMC time.
+        let content = b"BT /F1 12 Tf 1 0 0 1 100 500 Tm \
+/Span <</ActualText (super) >> BDC 5 Ts (sup) Tj 0 Ts EMC (after) Tj ET";
+
+        let items = extract_simple_items(content);
+        let sup = items.iter().find(|item| item.text == "super").unwrap();
+        let after = items.iter().find(|item| item.text == "after").unwrap();
+
+        assert!((sup.y - 505.0).abs() < 0.1);
+        assert!((after.y - 500.0).abs() < 0.1);
+    }
+
+    #[test]
+    fn actual_text_shown_with_quote_op_uses_moved_risen_baseline() {
+        // When the tagged span's show op is `'`, the glyph position is
+        // only known AFTER its line move — falling back to the BDC-entry
+        // matrix would place the item on the previous line, unrisen.
+        let content = b"BT /F1 12 Tf 14 TL 1 0 0 1 100 500 Tm \
+/Span <</ActualText (replaced) >> BDC 3 Ts (raw) ' 0 Ts EMC ET";
+
+        let items = extract_simple_items(content);
+        let item = items.iter().find(|item| item.text == "replaced").unwrap();
+
+        // Line move: 500 - 14 = 486; rise: +3 -> 489.
+        assert!((item.y - 489.0).abs() < 0.1);
+        assert!(item.width > 0.0);
+    }
+
+    #[test]
+    fn strikeout_detected_on_risen_text() {
+        // The rule crosses the glyphs at their risen position; without the
+        // rise in item.y the strike window sits 4pt too low and misses.
+        let content = b"BT /F1 12 Tf 1 0 0 1 100 500 Tm 4 Ts (struck) Tj ET
+1 w
+99 507 m 145 507 l S";
+
+        let items = extract_simple_items(content);
+        let struck = items.iter().find(|item| item.text == "struck").unwrap();
+
+        assert!((struck.y - 504.0).abs() < 0.1);
+        assert!(struck.is_strikeout);
+        assert!(!struck.is_underline);
+    }
+
     #[test]
     fn test_skip_excessive_operations() {
         use crate::tounicode::FontCMaps;
@@ -1279,13 +1727,113 @@ mod tests {
         doc.add_object(catalog);
 
         let font_cmaps = FontCMaps::from_doc(&doc);
-        let result = extract_page_text_items(&doc, page_id, 1, &font_cmaps, false).unwrap();
+        let result = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
         let ((items, rects, lines), _has_gid, _coords_rotated) = result;
         assert!(items.is_empty());
         assert!(rects.is_empty());
         assert!(lines.is_empty());
     }
 
+    #[test]
+    fn test_q_restores_current_font_for_text_decoding() {
+        use crate::tounicode::FontCMaps;
+        use lopdf::{dictionary, Object, Stream};
+
+        fn cmap_stream(dst_hex: &str) -> Stream {
+            let cmap = format!(
+                r#"/CIDInit /ProcSet findresource begin
+12 dict begin
+begincmap
+/CIDSystemInfo << /Registry (Adobe) /Ordering (UCS) /Supplement 0 >> def
+/CMapName /Test-UCS def
+/CMapType 2 def
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+1 beginbfchar
+<41> <{dst_hex}>
+endbfchar
+endcmap
+CMapName currentdict /CMap defineresource pop
+end
+end"#
+            );
+            Stream::new(dictionary! {}, cmap.into_bytes())
+        }
+
+        let mut doc = lopdf::Document::new();
+        let f1_cmap = doc.add_object(Object::Stream(cmap_stream("0058"))); // X
+        let f2_cmap = doc.add_object(Object::Stream(cmap_stream("0059"))); // Y
+        let f1 = doc.add_object(dictionary! {
+            "Type" => "Font",
+            "Subtype" => "Type1",
+            "BaseFont" => "Helvetica",
+            "ToUnicode" => Object::Reference(f1_cmap),
+        });
+        let f2 = doc.add_object(dictionary! {
+            "Type" => "Font",
+            "Subtype" => "Type1",
+            "BaseFont" => "Helvetica",
+            "ToUnicode" => Object::Reference(f2_cmap),
+        });
+
+        let content = b"BT /F1 12 Tf 10 700 Tm <41> Tj ET
+q
+BT /F2 12 Tf 20 700 Tm <41> Tj ET
+Q
+BT 30 700 Tm <41> Tj ET";
+        let content_id = doc.add_object(Object::Stream(Stream::new(
+            dictionary! {},
+            content.to_vec(),
+        )));
+        let page_id = doc.add_object(dictionary! {
+            "Type" => "Page",
+            "Contents" => Object::Reference(content_id),
+            "Resources" => dictionary! {
+                "Font" => dictionary! {
+                    "F1" => Object::Reference(f1),
+                    "F2" => Object::Reference(f2),
+                },
+            },
+            "MediaBox" => vec![0.into(), 0.into(), 612.into(), 792.into()],
+        });
+        let pages_id = doc.add_object(dictionary! {
+            "Type" => "Pages",
+            "Count" => Object::Integer(1),
+            "Kids" => vec![Object::Reference(page_id)],
+        });
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "Pages" => Object::Reference(pages_id),
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let font_cmaps = FontCMaps::from_doc(&doc);
+        let ((items, _, _), _, _) = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
+        let text = items
+            .iter()
+            .map(|item| item.text.as_str())
+            .collect::<String>();
+
+        assert_eq!(text, "XYX");
+    }
+
     #[test]
     fn test_strip_pdf_comments() {
         // Basic comment stripping
@@ -1317,4 +1865,26 @@ mod tests {
             "ET should be preserved after comment stripping"
         );
     }
+
+    #[test]
+    fn test_strip_pdf_comments_escaped_parens() {
+        // An escaped `\)` must not close the string: the `%` after it is
+        // still string content, not a comment (subset fonts routinely map
+        // glyphs to `%` and to escaped parens in the same TJ array).
+        let input = b"[ (a\\)b) 1 (%) 1 (c) ] TJ\n";
+        let output = strip_pdf_comments(input);
+        assert_eq!(output, input.to_vec());
+
+        // Same for an escaped `\(` — must not open a phantom string that
+        // shields a real comment.
+        let input = b"(x\\(y) Tj % real comment\nET\n";
+        let output = strip_pdf_comments(input);
+        assert_eq!(output, b"(x\\(y) Tj  \nET\n");
+
+        // Escaped backslash before a real close-paren: `\\` ends the escape,
+        // the `)` does close the string, and the comment is stripped.
+        let input = b"(x\\\\) Tj % comment\nET\n";
+        let output = strip_pdf_comments(input);
+        assert_eq!(output, b"(x\\\\) Tj  \nET\n");
+    }
 }

src/extractor/links.rs

@@ -2,11 +2,51 @@
 
 use crate::types::{ItemType, TextItem};
 use lopdf::{Document, Object, ObjectId};
-use std::collections::HashMap;
+use std::collections::{HashMap, HashSet};
 
 use super::fonts::{resolve_array, resolve_dict};
 use super::get_number;
 
+/// Upper bound on the number of form-field nodes visited during a single
+/// `extract_form_fields` pass. A crafted PDF can chain thousands of distinct
+/// `/Kids` fields to blow the stack even without an outright reference cycle,
+/// so we cap total traversal work in addition to detecting cycles.
+const MAX_FORM_FIELD_NODES: usize = 100_000;
+
+/// Upper bound on `/Kids` recursion depth. Real AcroForm hierarchies are only
+/// a few levels deep (fields → child fields → widgets); a crafted PDF can chain
+/// tens of thousands of distinct fields into a linear `/Kids` list that would
+/// overflow the stack via depth-first recursion long before the node budget is
+/// reached. This depth cap bounds the stack independently of total node count.
+const MAX_FORM_FIELD_DEPTH: usize = 100;
+
+/// Traversal budget for the AcroForm field walk. Bounds both the number of
+/// distinct nodes visited *and* the total number of `/Fields`/`/Kids` entries
+/// examined.
+///
+/// Counting `visited` alone is not enough: invalid entries (non-references) and
+/// duplicate references never grow `visited`, so an oversized array full of them
+/// would iterate to completion no matter how large. Charging every examined
+/// entry against the same budget makes it a real cap on traversal work.
+pub(crate) struct FieldWalkBudget {
+    visited: HashSet<ObjectId>,
+    examined: usize,
+}
+
+impl FieldWalkBudget {
+    fn new() -> Self {
+        Self {
+            visited: HashSet::new(),
+            examined: 0,
+        }
+    }
+
+    /// True once the budget is spent; callers must stop iterating and recursing.
+    fn exhausted(&self) -> bool {
+        self.visited.len() >= MAX_FORM_FIELD_NODES || self.examined >= MAX_FORM_FIELD_NODES
+    }
+}
+
 pub fn extract_page_links(doc: &Document, page_id: ObjectId, page_num: u32) -> Vec<TextItem> {
     let mut links = Vec::new();
 
@@ -78,6 +118,8 @@ pub fn extract_page_links(doc: &Document, page_id: ObjectId, page_num: u32) -> V
                             page: page_num,
                             is_bold: false,
                             is_italic: false,
+                            is_underline: false,
+                            is_strikeout: false,
                             item_type: ItemType::Link(url),
                             mcid: None,
                         });
@@ -144,32 +186,104 @@ pub(crate) fn extract_form_fields(
         Err(_) => return items,
     };
 
+    // Borrow the array rather than cloning it: a crafted `/Fields` can be huge,
+    // and cloning would pay an O(n) allocation/copy before the budget check
+    // below can stop the work.
     let fields = match acroform.get(b"Fields") {
         Ok(obj) => match resolve_array(doc, obj) {
-            Some(arr) => arr.clone(),
+            Some(arr) => arr,
             None => return items,
         },
         Err(_) => return items,
     };
+    if fields.is_empty() {
+        return items;
+    }
+    let annotation_pages = annotation_page_map(doc, page_map);
 
-    for field_obj in &fields {
+    // Bound the walk so a crafted PDF cannot send us into unbounded recursion
+    // via a `/Kids` cycle, a deep chain, or an oversized array of invalid or
+    // duplicate entries.
+    let mut budget = FieldWalkBudget::new();
+
+    for field_obj in fields {
+        // Stop once the budget is spent so a `/Fields` array wider than the
+        // budget can't burn CPU iterating entries whose walk would no-op. Charge
+        // every entry (including invalid ones) against the budget.
+        if budget.exhausted() {
+            break;
+        }
+        budget.examined += 1;
         if let Ok(field_ref) = field_obj.as_reference() {
-            walk_form_fields(doc, field_ref, None, "", page_map, &mut items);
+            walk_form_fields(
+                doc,
+                field_ref,
+                None,
+                "",
+                page_map,
+                &annotation_pages,
+                &mut items,
+                &mut budget,
+                0,
+            );
         }
     }
 
     items
 }
 
+/// Map widget annotation objects back to the page whose `/Annots` array owns
+/// them. Some valid widgets omit `/P`, so the page tree is the only reliable
+/// ownership signal available for page-filtered extraction.
+fn annotation_page_map(
+    doc: &Document,
+    page_map: &HashMap<ObjectId, u32>,
+) -> HashMap<ObjectId, u32> {
+    let mut annotation_pages = HashMap::new();
+    for (&page_id, &page_num) in page_map {
+        let Some(annotations) = doc
+            .get_dictionary(page_id)
+            .ok()
+            .and_then(|page| page.get(b"Annots").ok())
+            .and_then(|annotations| resolve_array(doc, annotations))
+        else {
+            continue;
+        };
+        for annotation in annotations {
+            if let Ok(annotation_id) = annotation.as_reference() {
+                annotation_pages.insert(annotation_id, page_num);
+            }
+        }
+    }
+    annotation_pages
+}
+
 /// Recursively walk the form field tree, extracting leaf field values.
+#[allow(clippy::too_many_arguments)]
 pub(crate) fn walk_form_fields(
     doc: &Document,
     field_id: ObjectId,
     parent_ft: Option<&[u8]>,
     parent_name: &str,
     page_map: &HashMap<ObjectId, u32>,
+    annotation_pages: &HashMap<ObjectId, u32>,
     items: &mut Vec<TextItem>,
+    budget: &mut FieldWalkBudget,
+    depth: usize,
 ) {
+    // Guard against `/Kids` cycles and pathologically large field trees.
+    // Exceeding the depth cap means the chain is too deep to be a legitimate
+    // form (and would overflow the stack); an exhausted budget means the tree is
+    // too large. Both checks run *before* inserting so the visited set can never
+    // grow past the budget.
+    if depth > MAX_FORM_FIELD_DEPTH || budget.exhausted() {
+        return;
+    }
+    // Revisiting an object ID means we hit a `/Kids` cycle.
+    if !budget.visited.insert(field_id) {
+        return;
+    }
+
     let field_dict = match doc.get_dictionary(field_id) {
         Ok(d) => d,
         Err(_) => return,
@@ -200,11 +314,31 @@ pub(crate) fn walk_form_fields(
 
     // Check for /Kids — if present, recurse into children
     if let Ok(kids_obj) = field_dict.get(b"Kids") {
+        // Iterate the borrowed array directly — cloning a crafted, oversized
+        // `/Kids` would allocate and copy every entry before the budget check
+        // below could stop the work.
         if let Some(kids) = resolve_array(doc, kids_obj) {
-            let kids = kids.clone();
-            for kid in &kids {
+            for kid in kids {
+                // Stop once the budget is spent so a `/Kids` array wider than the
+                // budget can't burn CPU iterating entries whose walk would no-op.
+                // Charge every entry (including invalid/duplicate ones) against
+                // the budget so this is a true traversal-work cap.
+                if budget.exhausted() {
+                    break;
+                }
+                budget.examined += 1;
                 if let Ok(kid_ref) = kid.as_reference() {
-                    walk_form_fields(doc, kid_ref, ft, &full_name, page_map, items);
+                    walk_form_fields(
+                        doc,
+                        kid_ref,
+                        ft,
+                        &full_name,
+                        page_map,
+                        annotation_pages,
+                        items,
+                        budget,
+                        depth + 1,
+                    );
                 }
             }
             return;
@@ -297,6 +431,7 @@ pub(crate) fn walk_form_fields(
         .ok()
         .and_then(|o| o.as_reference().ok())
         .and_then(|p| page_map.get(&p).copied())
+        .or_else(|| annotation_pages.get(&field_id).copied())
         .unwrap_or(1);
 
     let text = if full_name.is_empty() {
@@ -316,7 +451,270 @@ pub(crate) fn walk_form_fields(
         page: page_num,
         is_bold: false,
         is_italic: false,
+        is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::FormField,
         mcid: None,
     });
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use lopdf::{dictionary, Object};
+
+    #[test]
+    fn widget_without_page_reference_uses_owning_page_annotation() {
+        let mut doc = Document::new();
+        let widget_id = doc.add_object(dictionary! {
+            "Type" => "Annot",
+            "Subtype" => "Widget",
+            "FT" => "Tx",
+            "T" => Object::string_literal("customer"),
+            "V" => Object::string_literal("Alice"),
+            "Rect" => vec![10.into(), 20.into(), 110.into(), 40.into()],
+        });
+        let page_one_id = doc.add_object(dictionary! {
+            "Type" => "Page",
+        });
+        let page_two_id = doc.add_object(dictionary! {
+            "Type" => "Page",
+            "Annots" => vec![Object::Reference(widget_id)],
+        });
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(widget_id)],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::from([(page_one_id, 1), (page_two_id, 2)]);
+        let items = extract_form_fields(&doc, &page_map);
+
+        assert_eq!(items.len(), 1);
+        assert_eq!(items[0].page, 2);
+        assert_eq!(items[0].text, "customer: Alice");
+    }
+
+    #[test]
+    fn kids_self_cycle_does_not_overflow_stack() {
+        // A crafted AcroForm field that lists itself in `/Kids` must not send
+        // the traversal into unbounded recursion.
+        let mut doc = Document::new();
+        let field_id = doc.new_object_id();
+        doc.set_object(
+            field_id,
+            dictionary! {
+                "FT" => "Tx",
+                "T" => Object::string_literal("loop"),
+                "Kids" => vec![Object::Reference(field_id)],
+            },
+        );
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(field_id)],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        // Completes (rather than overflowing the stack) and yields no items.
+        let items = extract_form_fields(&doc, &page_map);
+        assert!(items.is_empty());
+    }
+
+    #[test]
+    fn kids_mutual_cycle_terminates() {
+        // Two fields that reference each other via `/Kids` form a cycle that
+        // must also terminate.
+        let mut doc = Document::new();
+        let field_a = doc.new_object_id();
+        let field_b = doc.new_object_id();
+        doc.set_object(
+            field_a,
+            dictionary! {
+                "T" => Object::string_literal("a"),
+                "Kids" => vec![Object::Reference(field_b)],
+            },
+        );
+        doc.set_object(
+            field_b,
+            dictionary! {
+                "T" => Object::string_literal("b"),
+                "Kids" => vec![Object::Reference(field_a)],
+            },
+        );
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(field_a)],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        let items = extract_form_fields(&doc, &page_map);
+        assert!(items.is_empty());
+    }
+
+    #[test]
+    fn deep_acyclic_kids_chain_does_not_overflow_stack() {
+        // A long chain of *distinct* fields (no cycle) must also terminate:
+        // the visited set alone would still recurse to the chain length, so
+        // the depth cap is what prevents a stack overflow here.
+        let mut doc = Document::new();
+        let n = MAX_FORM_FIELD_DEPTH * 500;
+        let ids: Vec<ObjectId> = (0..=n).map(|_| doc.new_object_id()).collect();
+        for i in 0..n {
+            doc.set_object(
+                ids[i],
+                dictionary! {
+                    "FT" => "Tx",
+                    "Kids" => vec![Object::Reference(ids[i + 1])],
+                },
+            );
+        }
+        // Leaf carries a value; it sits far below the depth cap so it is never
+        // reached, proving traversal stops early rather than crashing.
+        doc.set_object(
+            ids[n],
+            dictionary! {
+                "FT" => "Tx",
+                "T" => Object::string_literal("leaf"),
+                "V" => Object::string_literal("x"),
+                "Rect" => vec![10.into(), 20.into(), 110.into(), 40.into()],
+            },
+        );
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(ids[0])],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        let items = extract_form_fields(&doc, &page_map);
+        assert!(items.is_empty());
+    }
+
+    #[test]
+    fn wide_tree_traversal_stops_at_node_budget() {
+        // A single field with a `/Kids` array wider than the node budget must
+        // stop traversal at the cap rather than growing `visited` (and the work)
+        // without bound. Each processed leaf emits one item, so the item count
+        // is bounded by the budget and reaches right up to it (a couple of
+        // slots go to the root and the boundary node charged against the cap).
+        let mut doc = Document::new();
+        let fanout = MAX_FORM_FIELD_NODES + 50;
+        let leaf_ids: Vec<ObjectId> = (0..fanout).map(|_| doc.new_object_id()).collect();
+        for &leaf in &leaf_ids {
+            doc.set_object(
+                leaf,
+                dictionary! {
+                    "FT" => "Tx",
+                    "V" => Object::string_literal("v"),
+                    "Rect" => vec![10.into(), 20.into(), 110.into(), 40.into()],
+                },
+            );
+        }
+        let kids: Vec<Object> = leaf_ids.iter().map(|&id| Object::Reference(id)).collect();
+        let root_id = doc.add_object(dictionary! {
+            "T" => Object::string_literal("root"),
+            "Kids" => kids,
+        });
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(root_id)],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        let items = extract_form_fields(&doc, &page_map);
+        // Extraction stops at the budget: bounded above by the cap, and it gets
+        // right up to it (allowing a small delta for the root/boundary nodes
+        // charged against the budget).
+        assert!(items.len() <= MAX_FORM_FIELD_NODES);
+        assert!(items.len() >= MAX_FORM_FIELD_NODES - 3);
+    }
+
+    #[test]
+    fn wide_top_level_fields_stop_at_node_budget() {
+        // A top-level `/Fields` array wider than the budget must also stop at
+        // the cap: the item count is bounded by the budget and reaches right up
+        // to it.
+        let mut doc = Document::new();
+        let fanout = MAX_FORM_FIELD_NODES + 50;
+        let leaf_ids: Vec<ObjectId> = (0..fanout).map(|_| doc.new_object_id()).collect();
+        for &leaf in &leaf_ids {
+            doc.set_object(
+                leaf,
+                dictionary! {
+                    "FT" => "Tx",
+                    "V" => Object::string_literal("v"),
+                    "Rect" => vec![10.into(), 20.into(), 110.into(), 40.into()],
+                },
+            );
+        }
+        let fields: Vec<Object> = leaf_ids.iter().map(|&id| Object::Reference(id)).collect();
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => fields,
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        let items = extract_form_fields(&doc, &page_map);
+        assert!(items.len() <= MAX_FORM_FIELD_NODES);
+        assert!(items.len() >= MAX_FORM_FIELD_NODES - 3);
+    }
+
+    #[test]
+    fn duplicate_and_invalid_kids_entries_stop_at_budget() {
+        // Duplicate references and non-reference junk never grow `visited`, so
+        // without charging examined entries against the budget an oversized
+        // array of them would iterate to completion. The walk must still
+        // terminate and extract the single real leaf exactly once.
+        let mut doc = Document::new();
+        let leaf_id = doc.new_object_id();
+        doc.set_object(
+            leaf_id,
+            dictionary! {
+                "FT" => "Tx",
+                "V" => Object::string_literal("v"),
+                "Rect" => vec![10.into(), 20.into(), 110.into(), 40.into()],
+            },
+        );
+        // A `/Kids` array far wider than the budget: half duplicate references
+        // to the same leaf, half invalid (null) entries.
+        let mut kids: Vec<Object> = Vec::new();
+        for i in 0..(MAX_FORM_FIELD_NODES * 2) {
+            if i % 2 == 0 {
+                kids.push(Object::Reference(leaf_id));
+            } else {
+                kids.push(Object::Null);
+            }
+        }
+        let root_id = doc.add_object(dictionary! {
+            "T" => Object::string_literal("root"),
+            "Kids" => kids,
+        });
+        let catalog_id = doc.add_object(dictionary! {
+            "Type" => "Catalog",
+            "AcroForm" => dictionary! {
+                "Fields" => vec![Object::Reference(root_id)],
+            },
+        });
+        doc.trailer.set("Root", Object::Reference(catalog_id));
+
+        let page_map = HashMap::new();
+        let items = extract_form_fields(&doc, &page_map);
+        assert_eq!(items.len(), 1);
+    }
+}

src/extractor/reading_order.rs

@@ -0,0 +1,591 @@
+//! Region-graph evidence for page reading order.
+//!
+//! Whole-page column histograms fail when images or spanning captions occupy
+//! only part of a page. This module turns image geometry and repeated row
+//! gutters into a small directed acyclic graph: content above a local column
+//! band, the left flow, the right flow, and content below it. The graph is
+//! deliberately evidence-gated; ordinary pages keep the established layout
+//! path.
+
+use crate::text_utils::{effective_width, is_cjk_char, is_rtl_text};
+use crate::types::TextItem;
+
+const MIN_IMAGE_WIDTH: f32 = 60.0;
+const MIN_IMAGE_HEIGHT: f32 = 40.0;
+const MIN_ROW_GUTTER: f32 = 8.0;
+const SPLIT_CLUSTER_TOLERANCE: f32 = 20.0;
+const MIN_ALIGNED_ROWS: usize = 4;
+
+pub(crate) type ImageRegion = (f32, f32, f32, f32);
+
+#[derive(Debug, Clone, Copy, PartialEq)]
+pub(crate) struct ColumnFlowBand {
+    pub(crate) split_x: f32,
+    pub(crate) y_bottom: f32,
+    pub(crate) y_top: f32,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+pub(crate) enum RegionKind {
+    FullWidth,
+    Column,
+}
+
+#[derive(Debug)]
+pub(crate) struct RegionNode {
+    pub(crate) kind: RegionKind,
+    pub(crate) items: Vec<TextItem>,
+}
+
+#[derive(Debug)]
+struct Row<'a> {
+    y: f32,
+    items: Vec<&'a TextItem>,
+}
+
+fn page_x_bounds(items: &[TextItem], images: &[ImageRegion]) -> Option<(f32, f32)> {
+    let text_min = items
+        .iter()
+        .map(|item| item.x)
+        .fold(f32::INFINITY, f32::min);
+    let text_max = items
+        .iter()
+        .map(|item| item.x + effective_width(item))
+        .fold(f32::NEG_INFINITY, f32::max);
+    let image_min = images
+        .iter()
+        .map(|region| region.0.min(region.2))
+        .fold(f32::INFINITY, f32::min);
+    let image_max = images
+        .iter()
+        .map(|region| region.0.max(region.2))
+        .fold(f32::NEG_INFINITY, f32::max);
+    let x_min = text_min.min(image_min);
+    let x_max = text_max.max(image_max);
+    (x_min.is_finite() && x_max.is_finite() && x_max > x_min).then_some((x_min, x_max))
+}
+
+fn group_rows(items: &[TextItem]) -> Vec<Row<'_>> {
+    const Y_TOLERANCE: f32 = 3.0;
+    let mut sorted: Vec<&TextItem> = items.iter().collect();
+    sorted.sort_by(|left, right| right.y.total_cmp(&left.y));
+    let mut rows: Vec<Row<'_>> = Vec::new();
+    for item in sorted {
+        if let Some(row) = rows
+            .last_mut()
+            .filter(|row| (row.y - item.y).abs() <= Y_TOLERANCE)
+        {
+            row.items.push(item);
+            row.y = row.items.iter().map(|member| member.y).sum::<f32>() / row.items.len() as f32;
+        } else {
+            rows.push(Row {
+                y: item.y,
+                items: vec![item],
+            });
+        }
+    }
+    for row in &mut rows {
+        row.items.sort_by(|left, right| left.x.total_cmp(&right.x));
+    }
+    rows
+}
+
+fn side_is_prose(items: &[&TextItem]) -> bool {
+    let text = items
+        .iter()
+        .map(|item| item.text.trim())
+        .collect::<Vec<_>>()
+        .join(" ");
+    let alphabetic_count = text
+        .chars()
+        .filter(|character| character.is_alphabetic())
+        .count();
+    let cjk_count = text
+        .chars()
+        .filter(|character| is_cjk_char(*character))
+        .count();
+    (text.split_whitespace().count() >= 3 || cjk_count >= 10) && alphabetic_count >= 10
+}
+
+fn aligned_row_split(row: &Row<'_>, x_min: f32, x_max: f32) -> Option<f32> {
+    if row.items.len() < 2 {
+        return None;
+    }
+    let page_width = x_max - x_min;
+    let center_low = x_min + page_width * 0.25;
+    let center_high = x_min + page_width * 0.75;
+    row.items
+        .windows(2)
+        .filter_map(|pair| {
+            let left_end = pair[0].x + effective_width(pair[0]);
+            let right_start = pair[1].x;
+            let gap = right_start - left_end;
+            let split_x = (left_end + right_start) / 2.0;
+            if gap < MIN_ROW_GUTTER || split_x < center_low || split_x > center_high {
+                return None;
+            }
+            let left: Vec<&TextItem> = row
+                .items
+                .iter()
+                .copied()
+                .filter(|item| item.x + effective_width(item) / 2.0 < split_x)
+                .collect();
+            let right: Vec<&TextItem> = row
+                .items
+                .iter()
+                .copied()
+                .filter(|item| item.x + effective_width(item) / 2.0 >= split_x)
+                .collect();
+            (side_is_prose(&left) && side_is_prose(&right)).then_some((split_x, gap))
+        })
+        .max_by(|left, right| left.1.total_cmp(&right.1))
+        .map(|candidate| candidate.0)
+}
+
+fn local_flow_below_full_width_image(
+    items: &[TextItem],
+    images: &[ImageRegion],
+    x_min: f32,
+    x_max: f32,
+) -> Option<ColumnFlowBand> {
+    let page_width = x_max - x_min;
+    let full_width_images: Vec<ImageRegion> = images
+        .iter()
+        .copied()
+        .filter(|&(x0, y0, x1, y1)| {
+            let width = (x1 - x0).abs();
+            let height = (y1 - y0).abs();
+            width >= page_width * 0.65 && height >= 60.0
+        })
+        .collect();
+    // A local column flow below an image is only unambiguous for a single,
+    // nearly square hero/figure. Wide report banners and full-page artwork
+    // frequently sit above unrelated page furniture whose aligned labels can
+    // mimic prose columns.
+    if full_width_images.len() != 1 {
+        return None;
+    }
+    let (image_x0, _, image_x1, _) = full_width_images[0];
+    let anchor_width = (image_x1 - image_x0).abs();
+    let anchor_height = (full_width_images[0].3 - full_width_images[0].1).abs();
+    if anchor_width < page_width * 0.85
+        || anchor_height < anchor_width * 0.85
+        || anchor_height > anchor_width * 1.2
+    {
+        return None;
+    }
+    let image_bottom = full_width_images
+        .iter()
+        .map(|&(_, y0, _, y1)| y0.min(y1))
+        .fold(f32::NEG_INFINITY, f32::max);
+    if !image_bottom.is_finite() {
+        return None;
+    }
+
+    let below: Vec<TextItem> = items
+        .iter()
+        .filter(|item| item.y < image_bottom && item.y >= image_bottom - 220.0)
+        .cloned()
+        .collect();
+    let candidates: Vec<(f32, f32)> = group_rows(&below)
+        .into_iter()
+        .filter_map(|row| aligned_row_split(&row, x_min, x_max).map(|split| (split, row.y)))
+        .collect();
+    if candidates.len() < MIN_ALIGNED_ROWS {
+        return None;
+    }
+
+    let mut clusters: Vec<Vec<(f32, f32)>> = Vec::new();
+    for candidate in candidates {
+        if let Some(cluster) = clusters.iter_mut().find(|cluster| {
+            let mean = cluster.iter().map(|entry| entry.0).sum::<f32>() / cluster.len() as f32;
+            (mean - candidate.0).abs() <= SPLIT_CLUSTER_TOLERANCE
+        }) {
+            cluster.push(candidate);
+        } else {
+            clusters.push(vec![candidate]);
+        }
+    }
+    let dominant = clusters.into_iter().max_by_key(Vec::len)?;
+    if dominant.len() < MIN_ALIGNED_ROWS {
+        return None;
+    }
+    let split_x = dominant.iter().map(|entry| entry.0).sum::<f32>() / dominant.len() as f32;
+    let y_top = dominant
+        .iter()
+        .map(|entry| entry.1)
+        .fold(f32::NEG_INFINITY, f32::max)
+        + 3.0;
+    let image_gap = image_bottom - y_top;
+    if !(60.0..=120.0).contains(&image_gap) {
+        return None;
+    }
+    let y_bottom = dominant
+        .iter()
+        .map(|entry| entry.1)
+        .fold(f32::INFINITY, f32::min)
+        - 3.0;
+    if y_top - y_bottom > 130.0 {
+        return None;
+    }
+    log::debug!(
+        "page {}: full-width image flow images={} aligned_rows={} split={:.1} page=[{:.1}..{:.1}] image_bottom={:.1} y=[{:.1}..{:.1}] full_width={:?}",
+        items.first().map_or(0, |item| item.page),
+        images.len(),
+        dominant.len(),
+        split_x,
+        x_min,
+        x_max,
+        image_bottom,
+        y_bottom,
+        y_top,
+        full_width_images
+    );
+    Some(ColumnFlowBand {
+        split_x,
+        y_bottom,
+        y_top,
+    })
+}
+
+fn paired_column_images(
+    items: &[TextItem],
+    images: &[ImageRegion],
+    split_x: f32,
+    x_min: f32,
+    x_max: f32,
+) -> Option<ColumnFlowBand> {
+    let page_width = x_max - x_min;
+    if split_x < x_min + page_width * 0.4 || split_x > x_min + page_width * 0.6 {
+        return None;
+    }
+    let qualifying: Vec<ImageRegion> = images
+        .iter()
+        .copied()
+        .filter(|&(x0, y0, x1, y1)| {
+            let image_left = x0.min(x1);
+            let image_right = x0.max(x1);
+            let confined_to_one_column = image_right <= split_x || image_left >= split_x;
+            confined_to_one_column
+                && (x1 - x0).abs() >= MIN_IMAGE_WIDTH
+                && (y1 - y0).abs() >= MIN_IMAGE_HEIGHT
+        })
+        .collect();
+    let wide_images: Vec<ImageRegion> = qualifying
+        .iter()
+        .copied()
+        .filter(|(x0, _, x1, _)| (x1 - x0).abs() >= page_width * 0.35)
+        .collect();
+    if qualifying.len() < 3 || wide_images.len() < 3 {
+        return None;
+    }
+    let has_left = qualifying
+        .iter()
+        .any(|&(x0, _, x1, _)| (x0 + x1) / 2.0 < split_x);
+    let has_right = qualifying
+        .iter()
+        .any(|&(x0, _, x1, _)| (x0 + x1) / 2.0 >= split_x);
+    if !has_left || !has_right {
+        return None;
+    }
+    // A meaningful image-backed column flow spans multiple vertical panels.
+    // Three same-row header/logo images can otherwise satisfy the image count
+    // and send an ordinary asymmetric page through sequential column order.
+    let image_y_min = wide_images
+        .iter()
+        .map(|region| region.1.min(region.3))
+        .fold(f32::INFINITY, f32::min);
+    let image_y_max = wide_images
+        .iter()
+        .map(|region| region.1.max(region.3))
+        .fold(f32::NEG_INFINITY, f32::max);
+    let has_vertical_stack = wide_images.iter().enumerate().any(|(index, left)| {
+        wide_images.iter().skip(index + 1).any(|right| {
+            let same_side =
+                ((left.0 + left.2) / 2.0 < split_x) == ((right.0 + right.2) / 2.0 < split_x);
+            let left_center = (left.1 + left.3) / 2.0;
+            let right_center = (right.1 + right.3) / 2.0;
+            let left_height = (left.3 - left.1).abs();
+            let right_height = (right.3 - right.1).abs();
+            let vertical_gap = if left.1.max(left.3) < right.1.min(right.3) {
+                right.1.min(right.3) - left.1.max(left.3)
+            } else if right.1.max(right.3) < left.1.min(left.3) {
+                left.1.min(left.3) - right.1.max(right.3)
+            } else {
+                0.0
+            };
+            same_side
+                && (left_center - right_center).abs() >= left_height.min(right_height) * 0.5
+                && vertical_gap <= left_height.max(right_height) * 0.5
+        })
+    });
+    if image_y_max - image_y_min < page_width * 0.45 || !has_vertical_stack {
+        return None;
+    }
+    let y_top = qualifying
+        .iter()
+        .map(|region| region.1.max(region.3))
+        .fold(f32::NEG_INFINITY, f32::max)
+        + 3.0;
+    // Only column-confined text proves the lower extent of the flow. A
+    // spanning heading or caption below the columns must become the trailing
+    // full-width node rather than stretching the column band to the page foot.
+    let y_bottom = items
+        .iter()
+        .filter(|item| {
+            let item_right = item.x + effective_width(item);
+            item.y <= y_top && (item_right <= split_x || item.x >= split_x)
+        })
+        .map(|item| item.y)
+        .fold(f32::INFINITY, f32::min)
+        - 3.0;
+    if !y_bottom.is_finite() {
+        return None;
+    }
+    let distinct_rows = |right: bool| {
+        let mut ys: Vec<f32> = items
+            .iter()
+            .filter(|item| {
+                item.y <= y_top && (item.x + effective_width(item) / 2.0 >= split_x) == right
+            })
+            .map(|item| item.y)
+            .collect();
+        ys.sort_by(|left, right| left.total_cmp(right));
+        ys.dedup_by(|left, right| (*left - *right).abs() <= 3.0);
+        ys.len()
+    };
+    let left_rows = distinct_rows(false);
+    let right_rows = distinct_rows(true);
+    let line_balance = left_rows.min(right_rows) as f32 / left_rows.max(right_rows).max(1) as f32;
+    (left_rows >= 5 && right_rows >= 5 && line_balance < 0.55).then(|| {
+        log::debug!(
+            "page {}: paired-image flow qualifying_images={} rows={}/{} split={:.1} page=[{:.1}..{:.1}] y=[{:.1}..{:.1}] images={:?}",
+            items.first().map_or(0, |item| item.page),
+            qualifying.len(),
+            left_rows,
+            right_rows,
+            split_x,
+            x_min,
+            x_max,
+            y_bottom,
+            y_top,
+            qualifying
+        );
+        ColumnFlowBand {
+            split_x,
+            y_bottom,
+            y_top,
+        }
+    })
+}
+
+pub(crate) fn infer_image_anchored_flow(
+    items: &[TextItem],
+    images: &[ImageRegion],
+    detected_split: Option<f32>,
+) -> Option<ColumnFlowBand> {
+    if items.is_empty() || images.is_empty() {
+        return None;
+    }
+    let (x_min, x_max) = page_x_bounds(items, images)?;
+    detected_split
+        .and_then(|split_x| paired_column_images(items, images, split_x, x_min, x_max))
+        .or_else(|| local_flow_below_full_width_image(items, images, x_min, x_max))
+}
+
+/// Partition a page into the topological order `above -> left -> right -> below`.
+/// These edges encode the reading-order DAG; empty nodes are omitted.
+pub(crate) fn build_region_graph(items: Vec<TextItem>, band: ColumnFlowBand) -> Vec<RegionNode> {
+    let mut above = Vec::new();
+    let mut left = Vec::new();
+    let mut right = Vec::new();
+    let mut below = Vec::new();
+    for item in items {
+        if item.y > band.y_top {
+            above.push(item);
+        } else if item.y < band.y_bottom {
+            below.push(item);
+        } else if item.x + effective_width(&item) / 2.0 < band.split_x {
+            left.push(item);
+        } else {
+            right.push(item);
+        }
+    }
+    let rtl = is_rtl_text(left.iter().chain(right.iter()).map(|item| &item.text));
+    let mut ordered = vec![(RegionKind::FullWidth, above)];
+    if rtl {
+        ordered.push((RegionKind::Column, right));
+        ordered.push((RegionKind::Column, left));
+    } else {
+        ordered.push((RegionKind::Column, left));
+        ordered.push((RegionKind::Column, right));
+    }
+    ordered.push((RegionKind::FullWidth, below));
+    ordered
+        .into_iter()
+        .filter_map(|(kind, items)| (!items.is_empty()).then_some(RegionNode { kind, items }))
+        .collect()
+}
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+    use crate::types::ItemType;
+
+    fn item(text: &str, x: f32, y: f32, width: f32) -> TextItem {
+        TextItem {
+            text: text.into(),
+            x,
+            y,
+            width,
+            height: 11.0,
+            font: "F1".into(),
+            font_size: 11.0,
+            page: 1,
+            is_bold: false,
+            is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
+            item_type: ItemType::Text,
+            mcid: None,
+        }
+    }
+
+    #[test]
+    fn full_width_image_anchors_local_two_column_flow() {
+        let mut items = vec![
+            item("A full width caption", 55.0, 230.0, 430.0),
+            item("A trailing full width heading", 55.0, 80.0, 430.0),
+        ];
+        for index in 0..5 {
+            let y = 170.0 - index as f32 * 14.0;
+            items.push(item("left column prose words", 55.0, y, 210.0));
+            items.push(item("right column prose words", 280.0, y, 210.0));
+        }
+        let images = vec![(55.0, 250.0, 490.0, 680.0)];
+        let band = infer_image_anchored_flow(&items, &images, None).unwrap();
+        assert!((band.split_x - 272.5).abs() < 2.0);
+        let graph = build_region_graph(items, band);
+        assert_eq!(graph.len(), 4);
+        assert_eq!(graph[0].kind, RegionKind::FullWidth);
+        assert_eq!(graph[1].kind, RegionKind::Column);
+        assert_eq!(graph[2].kind, RegionKind::Column);
+        assert_eq!(graph[3].kind, RegionKind::FullWidth);
+        assert_eq!(graph[3].items[0].text, "A trailing full width heading");
+    }
+
+    #[test]
+    fn full_width_image_anchors_cjk_column_flow() {
+        let mut items = Vec::new();
+        for index in 0..5 {
+            let y = 170.0 - index as f32 * 14.0;
+            items.push(item("左栏这是没有空格的正文内容", 55.0, y, 210.0));
+            items.push(item("右栏这是没有空格的正文内容", 280.0, y, 210.0));
+        }
+        let images = vec![(55.0, 250.0, 490.0, 680.0)];
+
+        assert!(infer_image_anchored_flow(&items, &images, None).is_some());
+    }
+
+    #[test]
+    fn paired_images_anchor_unbalanced_column_flows() {
+        let mut items = vec![
+            item("running header", 55.0, 700.0, 430.0),
+            item("trailing full width caption", 55.0, 300.0, 430.0),
+        ];
+        for index in 0..5 {
+            items.push(item(
+                "left prose words",
+                55.0,
+                500.0 - index as f32 * 14.0,
+                200.0,
+            ));
+            items.push(item(
+                "right prose words",
+                280.0,
+                520.0 - index as f32 * 14.0,
+                200.0,
+            ));
+        }
+        for index in 5..12 {
+            items.push(item(
+                "right continuation prose words",
+                280.0,
+                520.0 - index as f32 * 14.0,
+                200.0,
+            ));
+        }
+        let images = vec![
+            (55.0, 530.0, 255.0, 680.0),
+            (55.0, 380.0, 255.0, 530.0),
+            (280.0, 560.0, 490.0, 680.0),
+        ];
+        let band = infer_image_anchored_flow(&items, &images, Some(270.0)).unwrap();
+        let graph = build_region_graph(items, band);
+        assert_eq!(graph[0].kind, RegionKind::FullWidth);
+        assert_eq!(graph[1].kind, RegionKind::Column);
+        assert_eq!(graph[2].kind, RegionKind::Column);
+        assert_eq!(graph[3].kind, RegionKind::FullWidth);
+        assert_eq!(graph[3].items[0].text, "trailing full width caption");
+    }
+
+    #[test]
+    fn rtl_region_graph_reads_right_column_first() {
+        let items = vec![
+            item("A long English report header", 55.0, 250.0, 430.0),
+            item("نص العمود الأيسر", 55.0, 150.0, 180.0),
+            item("نص العمود الأيمن", 300.0, 150.0, 180.0),
+        ];
+        let graph = build_region_graph(
+            items,
+            ColumnFlowBand {
+                split_x: 270.0,
+                y_bottom: 100.0,
+                y_top: 200.0,
+            },
+        );
+
+        assert_eq!(graph.len(), 3);
+        assert_eq!(graph[0].kind, RegionKind::FullWidth);
+        assert!(graph[1].items[0].x > graph[2].items[0].x);
+    }
+
+    #[test]
+    fn paired_header_logos_do_not_anchor_page_columns() {
+        let mut items = Vec::new();
+        for index in 0..7 {
+            items.push(item(
+                "left prose words",
+                55.0,
+                700.0 - index as f32 * 14.0,
+                200.0,
+            ));
+        }
+        for index in 0..30 {
+            items.push(item(
+                "right prose words",
+                280.0,
+                700.0 - index as f32 * 14.0,
+                200.0,
+            ));
+        }
+        let images = vec![
+            (55.0, 720.0, 205.0, 770.0),
+            (60.0, 718.0, 210.0, 768.0),
+            (280.0, 720.0, 450.0, 770.0),
+        ];
+        assert!(infer_image_anchored_flow(&items, &images, Some(270.0)).is_none());
+    }
+
+    #[test]
+    fn wide_banner_does_not_anchor_local_columns() {
+        let mut items = Vec::new();
+        for index in 0..7 {
+            let y = 270.0 - index as f32 * 14.0;
+            items.push(item("left column prose words", 55.0, y, 210.0));
+            items.push(item("right column prose words", 280.0, y, 210.0));
+        }
+        let images = vec![(55.0, 310.0, 490.0, 550.0)];
+        assert!(infer_image_anchored_flow(&items, &images, None).is_none());
+    }
+}

src/extractor/xobjects.rs

@@ -1,5 +1,6 @@
 //! Form XObject and image XObject extraction.
 
+use super::fonts::descriptor_style_flags;
 use crate::text_utils::{effective_font_size, expand_ligatures, is_bold_font, is_italic_font};
 use crate::tounicode::FontCMaps;
 use crate::types::{ItemType, TextItem};
@@ -7,8 +8,9 @@ use lopdf::{Document, Encoding, Object, ObjectId};
 use std::collections::HashMap;
 
 use super::fonts::{
-    build_font_encodings, build_font_widths, compute_string_width_ts, extract_text_from_operand,
-    get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    build_font_encodings, build_font_widths, build_type3_scales, compute_string_width_ts,
+    extract_text_from_operand, get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    FontStyleCache,
 };
 use super::{get_number, image_bbox_from_ctm, multiply_matrices};
 
@@ -114,6 +116,7 @@ pub(crate) fn extract_form_xobject_text(
     font_cmaps: &FontCMaps,
     parent_ctm: &[f32; 6],
     cmap_decisions: &mut CMapDecisionCache,
+    style_cache: &mut FontStyleCache,
 ) -> Vec<TextItem> {
     extract_form_xobject_text_inner(
         doc,
@@ -122,10 +125,12 @@ pub(crate) fn extract_form_xobject_text(
         font_cmaps,
         parent_ctm,
         cmap_decisions,
+        style_cache,
         0,
     )
 }
 
+#[allow(clippy::too_many_arguments)]
 fn extract_form_xobject_text_inner(
     doc: &Document,
     form_id: ObjectId,
@@ -133,6 +138,7 @@ fn extract_form_xobject_text_inner(
     font_cmaps: &FontCMaps,
     parent_ctm: &[f32; 6],
     cmap_decisions: &mut CMapDecisionCache,
+    style_cache: &mut FontStyleCache,
     depth: u8,
 ) -> Vec<TextItem> {
     use lopdf::content::Content;
@@ -157,16 +163,18 @@ fn extract_form_xobject_text_inner(
 
     // Get fonts from the Form's Resources
     let form_fonts = get_form_fonts(doc, &stream.dict);
-    let (font_encodings, _has_gid_fonts) = build_font_encodings(doc, &form_fonts);
+    let (font_encodings, _has_gid_fonts) = build_font_encodings(doc, &form_fonts, font_cmaps);
 
     // Build font width info for the form
     let font_widths = build_font_widths(doc, &form_fonts);
+    let type3_scales = build_type3_scales(doc, &form_fonts);
 
     // Build font base names and ToUnicode refs for the form
     let mut font_base_names: HashMap<String, String> = HashMap::new();
     let mut font_tounicode_refs: HashMap<String, u32> = HashMap::new();
     let mut inline_cmaps: HashMap<String, crate::tounicode::CMapEntry> = HashMap::new();
 
+    let mut font_style_flags: HashMap<String, (bool, bool)> = HashMap::new();
     for (font_name, font_dict) in &form_fonts {
         let resource_name = String::from_utf8_lossy(font_name).to_string();
         if let Ok(base_font) = font_dict.get(b"BaseFont") {
@@ -175,6 +183,10 @@ fn extract_form_xobject_text_inner(
                 font_base_names.insert(resource_name.clone(), base_name);
             }
         }
+        let style = descriptor_style_flags(doc, font_dict, style_cache);
+        if style != (false, false) {
+            font_style_flags.insert(resource_name.clone(), style);
+        }
         match font_dict.get(b"ToUnicode") {
             Ok(tounicode) => {
                 if let Ok(obj_ref) = tounicode.as_reference() {
@@ -272,6 +284,7 @@ fn extract_form_xobject_text_inner(
                                         font_cmaps,
                                         &ctm,
                                         cmap_decisions,
+                                        style_cache,
                                         depth + 1,
                                     );
                                     items.extend(nested_items);
@@ -294,6 +307,8 @@ fn extract_form_xobject_text_inner(
                                     page: page_num,
                                     is_bold: false,
                                     is_italic: false,
+                                    is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Image,
                                     mcid: None,
                                 });
@@ -400,7 +415,8 @@ fn extract_form_xobject_text_inner(
                         &font_widths,
                     ) {
                         let combined = multiply_matrices(&text_matrix, &ctm);
-                        let rendered_size = effective_font_size(current_font_size, &combined);
+                        let rendered_size = effective_font_size(current_font_size, &combined)
+                            * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                         let (x, y) = (combined[4], combined[5]);
                         let width = if let Some(font_info) = font_widths.get(&current_font) {
                             if let Some(raw_bytes) = get_operand_bytes(&op.operands[0]) {
@@ -427,6 +443,10 @@ fn extract_form_xobject_text_inner(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
@@ -436,8 +456,10 @@ fn extract_form_xobject_text_inner(
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
+                                is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: None,
                             });
@@ -553,11 +575,16 @@ fn extract_form_xobject_text_inner(
                         }
                         if !sub_items.is_empty() {
                             let combined = multiply_matrices(&text_matrix, &ctm);
-                            let rendered_size = effective_font_size(current_font_size, &combined);
+                            let rendered_size = effective_font_size(current_font_size, &combined)
+                                * type3_scales.get(&current_font).copied().unwrap_or(1.0);
                             let base_font = font_base_names
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             let scale_x = text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2];
                             for (text, start_w, end_w) in &sub_items {
                                 let offset_tm = [
@@ -584,8 +611,10 @@ fn extract_form_xobject_text_inner(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
+                                    is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: None,
                                 });

src/glyph_names.rs

@@ -4566,9 +4566,13 @@ pub fn glyph_to_char(name: &str) -> Option<char> {
         }
     }
 
-    // Try to parse uniXXXX format
-    if name.starts_with("uni") && name.len() >= 7 {
-        if let Ok(code) = u32::from_str_radix(&name[3..7], 16) {
+    // Try to parse uniXXXX format.
+    // Use `get` rather than a byte-length check + slice: `name` can contain
+    // non-ASCII bytes (e.g. U+FFFD from lossy UTF-8 decoding of an attacker
+    // controlled /Differences name), so byte index 7 may not be a char
+    // boundary and `&name[3..7]` would panic.
+    if let Some(hex) = name.strip_prefix("uni").and_then(|rest| rest.get(..4)) {
+        if let Ok(code) = u32::from_str_radix(hex, 16) {
             // Strip PUA F000 offset: uniF0XX → U+00XX (Windows Symbol encoding convention)
             let code = if (0xF000..=0xF0FF).contains(&code) {
                 code - 0xF000
@@ -4588,3 +4592,36 @@ pub fn glyph_to_char(name: &str) -> Option<char> {
 
     None
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn uni_hex_parsing() {
+        assert_eq!(glyph_to_char("uni0041"), Some('A'));
+        assert_eq!(glyph_to_char("uni00e9"), Some('\u{00e9}'));
+        // PUA F0xx symbol-encoding offset is stripped.
+        assert_eq!(glyph_to_char("uniF041"), Some('A'));
+    }
+
+    #[test]
+    fn u_hex_parsing() {
+        assert_eq!(glyph_to_char("u0041"), Some('A'));
+        assert_eq!(glyph_to_char("u1F600"), Some('\u{1F600}'));
+    }
+
+    #[test]
+    fn non_ascii_uni_name_does_not_panic() {
+        // A crafted /Differences name like `/uni#80#80#80#80` decodes via
+        // from_utf8_lossy into "uni" followed by four U+FFFD replacements.
+        // Byte index 7 lands mid-character, so a naive `&name[3..7]` slice
+        // would panic. It must be handled gracefully instead.
+        let crafted = format!("uni{0}{0}{0}{0}", '\u{FFFD}');
+        assert_eq!(glyph_to_char(&crafted), None);
+
+        // Assorted non-ASCII bytes right after the "uni" prefix.
+        assert_eq!(glyph_to_char("uni\u{FFFD}bc"), None);
+        assert_eq!(glyph_to_char("uni\u{00e9}00"), None);
+    }
+}

src/markdown/analysis.rs

@@ -130,6 +130,315 @@ pub(crate) fn has_dot_leaders(text: &str) -> bool {
     dot_groups >= 2
 }
 
+/// Detect a table-of-contents entry: a line ending in a page number preceded by
+/// a dot-leader group (e.g. "Measurement Lab worksheet ... 3"). `has_dot_leaders`
+/// misses single-group leaders ("..."), but a trailing "<dots> <number>" is a
+/// strong TOC signal on its own. Such lines must never be promoted to headings.
+pub(crate) fn is_toc_entry_line(text: &str) -> bool {
+    let trimmed = text.trim_end();
+    let digits = trimmed
+        .chars()
+        .rev()
+        .take_while(|c| c.is_ascii_digit())
+        .count();
+    if digits == 0 || digits > 4 {
+        return false;
+    }
+    let before_number = trimmed[..trimmed.len() - digits].trim_end();
+    let dots = before_number
+        .chars()
+        .rev()
+        .take_while(|c| *c == '.')
+        .count();
+    dots >= 3
+}
+
+/// A heading that announces a table of contents ("Contents", "Table of
+/// Contents"). Lines after it on the same page are ToC entries — section
+/// titles that look exactly like headings but must not be promoted.
+pub(crate) fn is_toc_marker_heading(text: &str) -> bool {
+    let t = text.trim().trim_end_matches(':').trim().to_lowercase();
+    matches!(t.as_str(), "contents" | "table of contents")
+}
+
+/// Lines that resemble headings structurally but are display-math fragments:
+/// equations ending in an equation number ("S = kB ln W, (2)") or equation
+/// lead-ins ("Rearranging Equation (8) gives:"). Both carry an "(N)" equation
+/// reference — but a trailing "(N)" alone is not enough: real headings end
+/// with parenthesized numbers too ("Nicaea (325)", appendix numbering), so
+/// the suffix form additionally requires math evidence — an "=" in the line
+/// or a comma immediately before the number, both present in every display
+/// equation and absent from name-plus-number headings. A bare trailing colon
+/// is NOT a fragment signal either: real headings frequently end with colons
+/// ("Procedure:", "Steps for Using the Microscope:").
+/// True when the line opens with a section number ("3.", "2.1.4", "IV)").
+///
+/// Mirrors the acceptance of `heading::parse_numbering` rather than the
+/// stricter `convert::starts_with_section_number`, which deliberately
+/// requires two components because it bypasses isolation checks. Here a
+/// single "1." counts: numbering is independent evidence of a heading, and
+/// `heading.rs` applies its numbered-prefix allowance *after* consulting
+/// `is_heading_fragment`, so without this exemption a numbered
+/// sentence-case heading would be vetoed before that allowance can run.
+fn starts_with_numbering_prefix(t: &str) -> bool {
+    let Some(first) = t.split_whitespace().next() else {
+        return false;
+    };
+    let has_delimiter = first.ends_with(['.', ')', ':']);
+    let token = first.trim_end_matches(['.', ')', ':']);
+    if token.is_empty() {
+        return false;
+    }
+    let parts: Vec<&str> = token.split('.').collect();
+    let decimal = parts
+        .iter()
+        .all(|p| !p.is_empty() && p.len() <= 3 && p.chars().all(|c| c.is_ascii_digit()));
+    if decimal {
+        // "1." / "2.1." carry a delimiter; "2.3 Title" is written without
+        // one, so a multi-component number is accepted bare. A bare single
+        // number ("3 apples") is not — that is ordinary prose.
+        return has_delimiter || parts.len() >= 2;
+    }
+    // Roman numerals go through the heading parser's own grammar so the two
+    // agree: uppercase I/V/X/L/C only, at most 8 characters. A looser rule
+    // here would exempt markers the parser rejects — "iv)" or "d)" from an
+    // alphabetical list — letting an ordinary list item bypass the veto and
+    // reach heading promotion.
+    //
+    // A delimiter is also required: a bare leading "I" is the pronoun far
+    // more often than a section number.
+    has_delimiter && crate::markdown::heading::roman_value(token).is_some()
+}
+
+/// True when the line reads as a title rather than a sentence: every
+/// content word (ignoring minor words) starts uppercase. Used to spare real
+/// headings from the dangling-verb veto — "Bond Yields" is a section title,
+/// "the method yields" is a stranded clause, and only the casing tells them
+/// apart.
+fn looks_title_case(t: &str) -> bool {
+    const MINOR: &[&str] = &[
+        "a", "an", "the", "of", "and", "or", "for", "to", "in", "on", "at", "by", "with", "from",
+        "as", "is", "are", "that", "than", "into",
+    ];
+    let mut content = 0usize;
+    let mut capitalized = 0usize;
+    for w in t.split_whitespace() {
+        let cleaned: String = w.chars().filter(|c| c.is_alphabetic()).collect();
+        if cleaned.is_empty() {
+            continue;
+        }
+        if MINOR.contains(&cleaned.to_lowercase().as_str()) {
+            continue;
+        }
+        content += 1;
+        if cleaned.chars().next().is_some_and(char::is_uppercase) {
+            capitalized += 1;
+        }
+    }
+    // A single content word ("Yields") is a title by default.
+    content == 0 || capitalized == content
+}
+
+pub(crate) fn is_heading_fragment(text: &str) -> bool {
+    let t = text.trim_end();
+
+    // A lowercase-initial one-or-two-word "heading" is a mid-sentence
+    // fragment beside display math ("or inversely", "and therefore") —
+    // real headings that short start uppercase. Measured as spurious
+    // headings on academic docs (fire-pdf ENG-5029 / opendataloader MHS).
+    {
+        let words: Vec<&str> = t.split_whitespace().collect();
+        if words.len() <= 2 {
+            if let Some(first_alpha) = t.chars().find(|c| c.is_alphabetic()) {
+                if first_alpha.is_lowercase() {
+                    return true;
+                }
+            }
+        }
+    }
+
+    fn is_equation_number(s: &str) -> bool {
+        s.strip_prefix('(')
+            .and_then(|r| r.strip_suffix(')'))
+            .is_some_and(|inner| {
+                !inner.is_empty() && inner.len() <= 3 && inner.chars().all(|c| c.is_ascii_digit())
+            })
+    }
+
+    // Equation-number suffix with math evidence: "S = kB ln W, (2)"
+    let mut rev = t.rsplit(' ');
+    let last = rev.next().unwrap_or("");
+    if is_equation_number(last) {
+        // Page-of-total running headers: "LIVSMEDELSVERKET PM 2 (10)"
+        if let Some(prev_word) = t.rsplit(' ').nth(1) {
+            if let (Ok(page), Some(total)) = (
+                prev_word.parse::<u32>(),
+                last.trim_start_matches('(')
+                    .trim_end_matches(')')
+                    .parse::<u32>()
+                    .ok(),
+            ) {
+                if page <= total {
+                    return true;
+                }
+            }
+        }
+        let punct_before = rev
+            .next()
+            .is_some_and(|w| w.ends_with(',') || w.ends_with(':'));
+        let has_math_op = t.chars().any(|c| {
+            matches!(
+                c,
+                '=' | '<'
+                    | '>'
+                    | '≤'
+                    | '≥'
+                    | '≪'
+                    | '≫'
+                    | '≈'
+                    | '≠'
+                    | '±'
+                    | '∑'
+                    | '∫'
+                    | '√'
+                    | '∝'
+            )
+        });
+        if punct_before || has_math_op {
+            return true;
+        }
+    }
+    // Lead-in: ends with a colon AND references an equation number inline
+    if t.ends_with(':') && t.split_whitespace().any(is_equation_number) {
+        return true;
+    }
+
+    // Dangling clause: a stranded sentence lead-in ends on a relational
+    // verb with no terminal punctuation — "Note that the exact error equals"
+    // left ahead of its formula when a phantom table dissolved.
+    //
+    // Gated on the line reading as prose rather than a title. Case is the
+    // discriminator the trailing word alone cannot provide: a heading is
+    // title case ("Bond Yields", "The Method Yields") while a stranded
+    // lead-in is sentence case ("the method yields"). Without this gate the
+    // veto eats real headings — "Bond Yields", "Crop Yields" and any wrapped
+    // title-case heading the preprocessor failed to merge.
+    if !t.ends_with(['.', '!', '?', ':', ';', ')', ']'])
+        && !looks_title_case(t)
+        && !starts_with_numbering_prefix(t)
+    {
+        if let Some(last) = t.split_whitespace().next_back() {
+            let word: String = last
+                .trim_matches(|c: char| !c.is_alphanumeric())
+                .to_lowercase();
+            // Relational verbs only, and only those with no common noun
+            // sense. "yields" was dropped for exactly that reason: "Bond
+            // Yields" is a real section title. Function words, copulas and
+            // auxiliaries were measured and rejected outright — a heading
+            // that wraps across lines ends on those, and suppressing them
+            // destroyed real IRS Publication 17 headings.
+            const DANGLING_TAIL: &[&str] =
+                &["equals", "denotes", "implies", "satisfies", "signifies"];
+            if DANGLING_TAIL.contains(&word.as_str()) {
+                return true;
+            }
+        }
+    }
+    false
+}
+
+#[cfg(test)]
+mod fragment_heading_tests {
+    use super::is_heading_fragment;
+
+    #[test]
+    fn dangling_tail_marks_stranded_clause() {
+        // opendataloader 01030000000144: left behind when a phantom table
+        // dissolved, ahead of its formula on the next line.
+        assert!(is_heading_fragment("Note that the exact error equals"));
+        assert!(is_heading_fragment("The remainder term satisfies"));
+        assert!(is_heading_fragment("we conclude that the sum equals"));
+    }
+
+    #[test]
+    fn real_headings_survive() {
+        assert!(!is_heading_fragment("Introduction"));
+        assert!(!is_heading_fragment("Error Analysis"));
+        assert!(!is_heading_fragment("Materials and Methods"));
+        assert!(!is_heading_fragment("Results"));
+        assert!(!is_heading_fragment("3.2 Richardson Extrapolation"));
+        assert!(!is_heading_fragment("Discussion and Conclusions"));
+        // Terminal punctuation means the clause is complete.
+        assert!(!is_heading_fragment("What is a Derivative?"));
+        assert!(!is_heading_fragment("Procedure:"));
+        assert!(!is_heading_fragment("Note that this is important."));
+    }
+
+    #[test]
+    fn title_case_headings_ending_in_a_verb_survive() {
+        // "yields" is also a plural noun; these are real section titles.
+        assert!(!is_heading_fragment("Bond Yields"));
+        assert!(!is_heading_fragment("Crop Yields"));
+        assert!(!is_heading_fragment("Dividend Yields"));
+        assert!(!is_heading_fragment("Yields"));
+        // A wrapped title-case heading whose first line ends on a listed
+        // verb must survive even if the preprocessor failed to merge it.
+        assert!(!is_heading_fragment("The Theorem Implies"));
+        assert!(!is_heading_fragment("What This Denotes"));
+    }
+
+    #[test]
+    fn numbered_sentence_case_headings_survive() {
+        // heading.rs consults is_heading_fragment BEFORE applying its
+        // numbered-prefix allowance, so the veto must not pre-empt it.
+        assert!(!is_heading_fragment("1. What the model implies"));
+        assert!(!is_heading_fragment("2.3 How the estimator satisfies"));
+        assert!(!is_heading_fragment("IV) What this denotes"));
+        // Without numbering the same wording is still a stranded clause.
+        assert!(is_heading_fragment("What the model implies"));
+        // A bare leading number or pronoun is prose, not numbering.
+        assert!(is_heading_fragment("3 apples and what that implies"));
+        assert!(is_heading_fragment("I think the model implies"));
+        // Markers heading::parse_numbering rejects must not be exempted
+        // either, or an ordinary list item bypasses the veto: lowercase
+        // roman, alphabetical markers, and over-long tokens.
+        assert!(is_heading_fragment("iv) the estimator satisfies"));
+        assert!(is_heading_fragment("d) the value implies"));
+        // Unsupported character (M is outside the parser's I/V/X/L/C set).
+        assert!(is_heading_fragment("MMMM. the value implies"));
+        // Over-long token: nine valid characters, so this exercises the
+        // 8-character bound rather than the character set.
+        assert!(is_heading_fragment("IIIIIIIII. the value implies"));
+        // Eight is still within the bound and stays exempt.
+        assert!(!is_heading_fragment("IIIIIIII. What this implies"));
+        // Uppercase roman within the parser's grammar is still exempt.
+        assert!(!is_heading_fragment("IV. What this denotes"));
+        assert!(!is_heading_fragment("XII) What this implies"));
+    }
+
+    #[test]
+    fn wrapped_headings_are_not_fragments() {
+        // A heading that wraps across lines ends on a function word. These
+        // are real headings from IRS Publication 17 and must survive.
+        assert!(!is_heading_fragment("Casualty and"));
+        assert!(!is_heading_fragment("Rule 10. You Must Be at"));
+        assert!(!is_heading_fragment("Higher Standard Deduction for"));
+        assert!(!is_heading_fragment("Qualifying Child of"));
+        assert!(!is_heading_fragment("When Can I Withdraw or"));
+        // Copulas and auxiliaries also end real wrapped headings.
+        assert!(!is_heading_fragment("Rule 15. Your AGI Must Be"));
+        assert!(!is_heading_fragment("What Medical Expenses Are"));
+        assert!(!is_heading_fragment("Rule 13. You Must Have"));
+        assert!(!is_heading_fragment("When Can a Roth IRA Be"));
+    }
+
+    #[test]
+    fn dangling_check_is_case_insensitive() {
+        // All-caps is not sentence case, so the veto must not fire there.
+        assert!(!is_heading_fragment("THE REMAINDER EQUALS"));
+    }
+}
+
 /// Compute the Y-gap threshold for paragraph break detection.
 ///
 /// Instead of using a fixed multiple of base_size (which fails for double-spaced
@@ -257,6 +566,15 @@ pub(crate) fn compute_heading_tiers(lines: &[TextLine], base_size: f32) -> Vec<f
     for line in lines {
         if let Some(first) = line.items.first() {
             if first.font_size / base_size >= 1.2 {
+                // Digit-only lines (page numbers, issue numbers) must not
+                // define heading tiers: a large bold folio claims tier 0 and
+                // blocks the bold-size fallback for the document's real
+                // same-size headings.
+                let text = line.text();
+                let t = text.trim();
+                if !t.is_empty() && t.chars().all(|c| !c.is_alphabetic()) {
+                    continue;
+                }
                 heading_sizes.push(first.font_size);
             }
         }
@@ -274,11 +592,45 @@ pub(crate) fn compute_heading_tiers(lines: &[TextLine], base_size: f32) -> Vec<f
         }
     }
 
+    // Books often set section headings barely above body size (e.g. 11pt
+    // bold over 10pt text). When nothing clears the 1.2x ratio gate, fall
+    // back to bold lines modestly larger than body so those documents still
+    // get an H1 instead of every bold heading defaulting to H2.
+    if tiers.is_empty() {
+        let mut bold_sizes: Vec<f32> = lines
+            .iter()
+            .filter(|line| {
+                let text = line.text();
+                let t = text.trim();
+                !t.is_empty() && t.chars().any(|c| c.is_alphabetic())
+            })
+            .filter_map(|line| line.items.first())
+            .filter(|it| it.is_bold && it.font_size / base_size >= 1.05)
+            .map(|it| it.font_size)
+            .collect();
+        bold_sizes.sort_by(|a, b| b.total_cmp(a));
+        for size in bold_sizes {
+            if !tiers.iter().any(|&t| (t - size).abs() < 0.5) {
+                tiers.push(size);
+            }
+        }
+    }
+
     // Cap at 4 tiers
     tiers.truncate(4);
     tiers
 }
 
+/// Boldness of a line judged by character mass, so a heading with an
+/// unbold section-number prefix ("4. " + bold title) still counts as bold.
+pub(crate) fn line_is_mostly_bold(line: &TextLine) -> bool {
+    let (bold, total) = line.items.iter().fold((0usize, 0usize), |(b, t), it| {
+        let n = it.text.trim().chars().count();
+        (b + if it.is_bold { n } else { 0 }, t + n)
+    });
+    total > 0 && bold * 2 >= total
+}
+
 /// Detect header level from font size using document-specific heading tiers.
 /// When tiers are available, maps tier 0→H1, tier 1→H2, etc.
 /// Falls back to ratio-based thresholds when no tiers exist.
@@ -286,9 +638,21 @@ pub(crate) fn detect_header_level(
     font_size: f32,
     base_size: f32,
     heading_tiers: &[f32],
+    is_bold: bool,
 ) -> Option<usize> {
     let ratio = font_size / base_size;
 
+    // Tier matches are trusted below the 1.2x gate (down to 1.05x) only for
+    // bold lines: sub-gate tiers come from the bold fallback, and honoring
+    // them for non-bold text at the same size would promote captions.
+    if (1.05..1.2).contains(&ratio) && is_bold && !heading_tiers.is_empty() {
+        for (i, &tier_size) in heading_tiers.iter().enumerate() {
+            if (font_size - tier_size).abs() < 0.5 {
+                return Some(i + 1); // tier 0 → H1, tier 1 → H2, etc.
+            }
+        }
+    }
+
     if ratio < 1.2 {
         return None; // Regular text
     }
@@ -320,3 +684,138 @@ pub(crate) fn detect_header_level(
         Some(4)
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    fn line_of(text: &str, font_size: f32, bold: bool, y: f32) -> crate::types::TextLine {
+        let item = crate::types::TextItem {
+            text: text.into(),
+            x: 72.0,
+            y,
+            width: text.len() as f32 * font_size * 0.5,
+            height: font_size,
+            font: "Test".into(),
+            font_size,
+            page: 1,
+            is_bold: bold,
+            is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
+            item_type: crate::types::ItemType::Text,
+            mcid: None,
+        };
+        crate::types::TextLine {
+            items: vec![item],
+            y,
+            page: 1,
+            adaptive_threshold: 0.10,
+        }
+    }
+
+    #[test]
+    fn digit_only_lines_do_not_define_tiers() {
+        // A 14pt bold page number must not claim tier 0 — that both demotes
+        // every real heading a level and blocks the bold-size fallback.
+        let lines = vec![
+            line_of("76", 14.0, true, 760.0),
+            line_of("Replace", 11.0, true, 700.0),
+            line_of("body text at eleven points", 11.0, false, 680.0),
+        ];
+        let tiers = compute_heading_tiers(&lines, 11.0);
+        assert!(tiers.is_empty(), "page number claimed a tier: {tiers:?}");
+    }
+
+    #[test]
+    fn bold_fallback_tiers_when_nothing_clears_ratio_gate() {
+        // 10pt body, 11pt bold section headings (book-style): no size clears
+        // 1.2x, so bold sizes modestly above body form the tiers.
+        let lines = vec![
+            line_of("4. Entropy", 11.0, true, 700.0),
+            line_of("body text about entropy", 10.0, false, 680.0),
+            line_of("5. The dynamics", 11.0, true, 500.0),
+        ];
+        let tiers = compute_heading_tiers(&lines, 10.0);
+        assert_eq!(tiers, vec![11.0]);
+        assert_eq!(detect_header_level(11.0, 10.0, &tiers, true), Some(1));
+        // Non-bold text at the fallback size must not become a heading.
+        assert_eq!(detect_header_level(11.0, 10.0, &tiers, false), None);
+        // Non-tier body text stays regular.
+        assert_eq!(detect_header_level(10.0, 10.0, &tiers, true), None);
+    }
+
+    #[test]
+    fn bold_fallback_skipped_when_real_tiers_exist() {
+        let lines = vec![
+            line_of("Chapter One", 18.0, false, 700.0),
+            line_of("bold label", 11.0, true, 600.0),
+            line_of("body", 10.0, false, 580.0),
+        ];
+        let tiers = compute_heading_tiers(&lines, 10.0);
+        assert_eq!(tiers, vec![18.0]);
+        // The 11pt bold label does not match any tier and stays non-heading.
+        assert_eq!(detect_header_level(11.0, 10.0, &tiers, true), None);
+    }
+
+    #[test]
+    fn toc_entry_with_single_dot_group() {
+        assert!(is_toc_entry_line("Measurement Lab worksheet ... 3"));
+        assert!(is_toc_entry_line("Results ........ 12"));
+        assert!(is_toc_entry_line("Appendix B...42"));
+    }
+
+    #[test]
+    fn non_toc_lines_pass() {
+        assert!(!is_toc_entry_line(
+            "6.2. Expectations for Re-Hiring Employees"
+        ));
+        assert!(!is_toc_entry_line("What happened in 2020"));
+        assert!(!is_toc_entry_line("IMPLEMENTATION"));
+        // Ellipsis without a trailing page number
+        assert!(!is_toc_entry_line("and so it goes ..."));
+        // Long numbers are data, not page refs
+        assert!(!is_toc_entry_line("ISBN ... 97814"));
+    }
+
+    #[test]
+    fn toc_marker_headings() {
+        assert!(is_toc_marker_heading("Contents"));
+        assert!(is_toc_marker_heading("CONTENTS"));
+        assert!(is_toc_marker_heading("Table of Contents"));
+        assert!(is_toc_marker_heading("Table of contents:"));
+        assert!(!is_toc_marker_heading("Contents of the Shipment"));
+        assert!(!is_toc_marker_heading("Introduction"));
+    }
+
+    #[test]
+    fn heading_fragments() {
+        // Equation lead-ins: colon ending + inline equation reference
+        assert!(is_heading_fragment("or inversely"));
+        assert!(is_heading_fragment("and therefore"));
+        assert!(!is_heading_fragment("Introduction"));
+        assert!(!is_heading_fragment("iPhone Sales Strategy Overview")); // 4 words, exempt
+        assert!(is_heading_fragment("Rearranging Equation (8) gives:"));
+        // Display-equation neighbours ending in an equation number
+        assert!(is_heading_fragment("S = kB ln W, (2)"));
+        assert!(is_heading_fragment("E = mc2 (12)"));
+        assert!(is_heading_fragment("x + y = z, (3)"));
+        // Page-of-total running headers
+        assert!(is_heading_fragment("LIVSMEDELSVERKET PM 2 (10)"));
+        // Comparison-operator evidence and colon-before-number
+        assert!(is_heading_fragment(
+            "PLL\u{fe} PHH\u{226a} PLH\u{fe} PHL: (12)"
+        ));
+        // Real headings pass — including name-plus-number and colon-ended ones
+        assert!(!is_heading_fragment("Nicaea (325)"));
+        assert!(!is_heading_fragment(
+            "\u{627}\u{644}\u{645}\u{644}\u{62d}\u{642} \u{631}\u{642}\u{645} (1)"
+        ));
+        assert!(!is_heading_fragment("4. Entropy"));
+        assert!(!is_heading_fragment("Procedure:"));
+        assert!(!is_heading_fragment("Steps for Using the Microscope:"));
+        assert!(!is_heading_fragment("Changing objectives:"));
+        assert!(!is_heading_fragment("Sales by Region (2024)"));
+        assert!(!is_heading_fragment("Results (preliminary)"));
+    }
+}

src/markdown/classify.rs

@@ -131,10 +131,11 @@ pub(crate) fn format_list_item(text: &str) -> String {
         if let Some(rest) = trimmed.strip_prefix(*bullet) {
             return format!("- {}", rest.trim_start());
         }
-        // Bullet inside a leading bold/italic run (e.g. "**● Label:** rest").
-        // The run wraps both the marker and the following label because both
-        // use a bold font in the PDF.
-        for wrapper in ["**", "*"] {
+        // Bullet inside a leading style run (e.g. "**● Label:** rest" or
+        // "<u>● Label</u>"). The run wraps both the marker and the following
+        // label because both carry the style in the PDF. The marker must move
+        // outside the wrapper so markdown still sees a list item.
+        for wrapper in ["**", "*", "<u>"] {
             if let Some(after_open) = trimmed.strip_prefix(wrapper) {
                 if let Some(rest) = after_open.strip_prefix(*bullet) {
                     return format!("- {}{}", wrapper, rest.trim_start());
@@ -235,6 +236,13 @@ mod tests {
         assert_eq!(format_list_item("• Item"), "- Item");
     }
 
+    #[test]
+    fn format_list_item_bullet_inside_underline() {
+        // Fully-underlined bullet line: the marker must move outside the
+        // <u> wrapper so markdown still renders a list item.
+        assert_eq!(format_list_item("<u>● Item text</u>"), "- <u>Item text</u>");
+    }
+
     #[test]
     fn format_list_item_bullet_inside_bold() {
         // PDF that uses bold font for both the marker and the label produces

src/markdown/postprocess.rs

@@ -2,12 +2,16 @@
 
 use regex::Regex;
 
-use super::MarkdownOptions;
+use super::{MarkdownOptions, MarkdownProfile};
+use crate::text_utils::is_page_number_line;
 
 /// Clean up markdown output with post-processing
 pub(crate) fn clean_markdown(mut text: String, options: &MarkdownOptions) -> String {
-    // Collapse dot leaders (e.g. TOC entries: "Introduction...............................1")
-    text = collapse_dot_leaders(&text);
+    if options.profile == MarkdownProfile::Compact {
+        // Dot-leader collapse saves tokens but changes source text, so it is
+        // reserved for the explicit compact profile.
+        text = collapse_dot_leaders(&text);
+    }
 
     // Fix hyphenation first (before other processing)
     if options.fix_hyphenation {
@@ -29,6 +33,8 @@ pub(crate) fn clean_markdown(mut text: String, options: &MarkdownOptions) -> Str
     // text item, which combine with gap-based space insertion to produce
     // double spaces ("Vice  President" instead of "Vice President").
     collapse_consecutive_spaces(&mut text);
+    remove_spaces_before_closing_brackets(&mut text);
+    remove_spaces_before_sentence_punctuation(&mut text);
 
     // Remove excessive newlines (more than 2 in a row)
     while text.contains("\n\n\n") {
@@ -71,6 +77,46 @@ fn collapse_consecutive_spaces(text: &mut String) {
     *text = result;
 }
 
+/// Remove spaces before closing square brackets.
+/// Unit markers and markdown links occasionally pick up a gap-inserted space
+/// before `]` (e.g. `[kg/m3 ]`), which is cosmetic padding.
+fn remove_spaces_before_closing_brackets(text: &mut String) {
+    let mut result = String::with_capacity(text.len());
+    for ch in text.chars() {
+        if ch == ']' && result.ends_with(' ') {
+            result.pop();
+        }
+        result.push(ch);
+    }
+    *text = result;
+}
+
+/// Remove a stray space before sentence punctuation ("word ." → "word.").
+/// Style-boundary item splits (bold/italic/underline runs) can strand a
+/// trailing period or comma in its own fragment, and several assembly paths
+/// join fragments with spaces. Only fires when the punctuation ends the
+/// token (followed by whitespace or end of text), so decimals ("3 .14" stays
+/// untouched — no such input exists, but the guard is cheap) and dot leaders
+/// (" ... ") are unaffected.
+fn remove_spaces_before_sentence_punctuation(text: &mut String) {
+    let chars: Vec<char> = text.chars().collect();
+    let mut result = String::with_capacity(text.len());
+    for (i, &ch) in chars.iter().enumerate() {
+        if matches!(ch, '.' | ',' | ';') && result.ends_with(' ') {
+            let next = chars.get(i + 1);
+            // `|` counts as a token end so table cells get the same fix.
+            let token_ends = next.is_none_or(|c| c.is_whitespace() || *c == '|');
+            // Never touch runs of dots (ellipsis / dot leaders).
+            let in_dot_run = ch == '.' && next == Some(&'.');
+            if token_ends && !in_dot_run {
+                result.pop();
+            }
+        }
+        result.push(ch);
+    }
+    *text = result;
+}
+
 /// Collapse dot leaders (runs of 4+ dots) into " ... "
 /// Common in tables of contents: "Introduction...............................1" -> "Introduction ... 1"
 fn collapse_dot_leaders(text: &str) -> String {
@@ -100,7 +146,7 @@ fn fix_hyphenation(text: &str) -> String {
     result
 }
 
-/// Remove standalone page numbers (lines that are just 1-4 digit numbers)
+/// Remove isolated page-number expressions from Markdown.
 fn remove_page_numbers(text: &str) -> String {
     let mut result = Vec::new();
     let lines: Vec<&str> = text.lines().collect();
@@ -138,69 +184,6 @@ fn remove_page_numbers(text: &str) -> String {
     result.join("\n")
 }
 
-/// Check if a line looks like a page number
-fn is_page_number_line(trimmed: &str) -> bool {
-    // Empty lines are not page numbers
-    if trimmed.is_empty() {
-        return false;
-    }
-
-    // Pattern 1: Just a number (1-4 digits)
-    if trimmed.len() <= 4 && trimmed.chars().all(|c| c.is_ascii_digit()) {
-        return true;
-    }
-
-    // Pattern 2: "Page X of Y" or "Page X" or "Page   of" (placeholder)
-    let lower = trimmed.to_lowercase();
-    if let Some(rest) = lower.strip_prefix("page") {
-        let rest = rest.trim();
-        // "Page   of" (empty page numbers)
-        if rest == "of" || rest.starts_with("of ") {
-            return true;
-        }
-        // "Page X" or "Page X of Y"
-        if rest
-            .chars()
-            .next()
-            .map(|c| c.is_ascii_digit())
-            .unwrap_or(false)
-        {
-            return true;
-        }
-        // Just "Page" followed by whitespace and maybe "of"
-        if rest.is_empty()
-            || rest
-                .split_whitespace()
-                .all(|w| w == "of" || w.chars().all(|c| c.is_ascii_digit()))
-        {
-            return true;
-        }
-    }
-
-    // Pattern 3: "X of Y" where X and Y are numbers
-    if let Some(of_idx) = trimmed.find(" of ") {
-        let before = trimmed[..of_idx].trim();
-        let after = trimmed[of_idx + 4..].trim();
-        if before.chars().all(|c| c.is_ascii_digit())
-            && after.chars().all(|c| c.is_ascii_digit())
-            && !before.is_empty()
-            && !after.is_empty()
-        {
-            return true;
-        }
-    }
-
-    // Pattern 4: "- X -" centered page number
-    if trimmed.len() >= 3 && trimmed.starts_with('-') && trimmed.ends_with('-') {
-        let inner = trimmed[1..trimmed.len() - 1].trim();
-        if inner.chars().all(|c| c.is_ascii_digit()) && !inner.is_empty() {
-            return true;
-        }
-    }
-
-    false
-}
-
 /// Convert URLs to markdown links
 fn format_urls(text: &str) -> String {
     use once_cell::sync::Lazy;
@@ -313,6 +296,23 @@ fn format_urls(text: &str) -> String {
 mod tests {
     use super::*;
 
+    #[test]
+    fn fidelity_profile_preserves_dot_leaders() {
+        let input = "Introduction............................1".to_string();
+        let result = clean_markdown(input.clone(), &MarkdownOptions::default());
+        assert_eq!(result, format!("{input}\n"));
+    }
+
+    #[test]
+    fn compact_profile_collapses_dot_leaders() {
+        let input = "Introduction............................1".to_string();
+        let options = MarkdownOptions {
+            profile: MarkdownProfile::Compact,
+            ..MarkdownOptions::default()
+        };
+        assert_eq!(clean_markdown(input, &options), "Introduction ... 1\n");
+    }
+
     // --- collapse_dot_leaders ---
 
     #[test]
@@ -342,6 +342,48 @@ mod tests {
         assert!(result.contains("Chapter 2 ... 20"));
     }
 
+    // --- remove_spaces_before_closing_brackets ---
+
+    #[test]
+    fn test_remove_spaces_before_closing_brackets() {
+        let mut input = "Density [kg/m3 ] and [linked text ](https://example.com)".to_string();
+        remove_spaces_before_closing_brackets(&mut input);
+        assert_eq!(
+            input,
+            "Density [kg/m3] and [linked text](https://example.com)"
+        );
+    }
+
+    // --- remove_spaces_before_sentence_punctuation ---
+
+    #[test]
+    fn strips_space_before_trailing_period() {
+        let mut t = "Foreign insurance companies . The provisions".to_string();
+        remove_spaces_before_sentence_punctuation(&mut t);
+        assert_eq!(t, "Foreign insurance companies. The provisions");
+    }
+
+    #[test]
+    fn strips_space_before_period_at_cell_boundary() {
+        let mut t = "|Applicability date .|This section|".to_string();
+        remove_spaces_before_sentence_punctuation(&mut t);
+        assert_eq!(t, "|Applicability date.|This section|");
+    }
+
+    #[test]
+    fn keeps_dot_leaders_and_ellipses() {
+        let mut t = "Introduction ... 1".to_string();
+        remove_spaces_before_sentence_punctuation(&mut t);
+        assert_eq!(t, "Introduction ... 1");
+    }
+
+    #[test]
+    fn keeps_mid_token_periods() {
+        let mut t = "version 3 .14 released".to_string();
+        remove_spaces_before_sentence_punctuation(&mut t);
+        assert_eq!(t, "version 3 .14 released");
+    }
+
     // --- fix_hyphenation ---
 
     #[test]
@@ -385,12 +427,14 @@ mod tests {
     fn test_is_page_number_page_x() {
         assert!(is_page_number_line("Page 5"));
         assert!(is_page_number_line("page 12"));
+        assert!(is_page_number_line("Page123"));
     }
 
     #[test]
     fn test_is_page_number_page_x_of_y() {
         assert!(is_page_number_line("Page 3 of 10"));
         assert!(is_page_number_line("page 1 of 5"));
+        assert!(is_page_number_line("Page 3 of 10 Report header"));
     }
 
     #[test]
@@ -420,6 +464,13 @@ mod tests {
         assert!(!is_page_number_line("Hello World"));
         assert!(!is_page_number_line("Chapter 1"));
         assert!(!is_page_number_line("Total: 500"));
+        assert!(!is_page_number_line("PAGE0-PARA2-END-MARKER-0"));
+    }
+
+    #[test]
+    fn test_is_page_number_labeled_running_header() {
+        assert!(is_page_number_line("Page 42 Chapter 5"));
+        assert!(is_page_number_line("Page 42 explains the result"));
     }
 
     // --- remove_page_numbers ---
@@ -447,6 +498,24 @@ mod tests {
         assert!(result.contains("42"));
     }
 
+    #[test]
+    fn test_remove_page_numbers_labeled_header_with_content() {
+        let input = "Content\n\nPage 42 explains the result\n---\nEnd";
+        let result = remove_page_numbers(input);
+
+        assert!(!result.contains("Page 42 explains the result"));
+        assert!(result.contains("Content"));
+        assert!(result.contains("End"));
+    }
+
+    #[test]
+    fn test_remove_page_numbers_preserves_page_prefixed_content() {
+        let input = "PAGE0-PARA2-START substantive report text PAGE0-PARA2-END-MARKER-0";
+        let result = remove_page_numbers(input);
+
+        assert_eq!(result, input);
+    }
+
     #[test]
     fn test_remove_page_numbers_multiple_patterns() {
         let input = "\n1\n\nContent\n\n2\n\n---\nMore\n\n3\n";

src/markdown/preprocess.rs

@@ -3,7 +3,7 @@
 use std::collections::{HashMap, HashSet};
 
 use crate::structure_tree::StructRole;
-use crate::types::TextLine;
+use crate::types::{TextItem, TextLine};
 
 use super::analysis::detect_header_level;
 
@@ -42,7 +42,12 @@ fn effective_heading_level(
 
     // Fall back to font-size heuristic
     let font = line.items.first().map(|i| i.font_size).unwrap_or(base_size);
-    detect_header_level(font, base_size, heading_tiers)
+    detect_header_level(
+        font,
+        base_size,
+        heading_tiers,
+        crate::markdown::analysis::line_is_mostly_bold(line),
+    )
 }
 
 /// Merge consecutive heading lines at the same level into a single line.
@@ -87,6 +92,41 @@ pub(crate) fn merge_heading_lines(
             false
         };
 
+        // Bold headings at body font size never reach a tier, so wrapped ones
+        // split into two output headings ("…of wood pellets and cost" /
+        // "structure in Japan"). Merge a fully-bold line into the previous
+        // fully-bold line when it reads as a wrap continuation: starts
+        // lowercase, tiny Y gap, and the previous line has no terminal
+        // punctuation. Kept deliberately narrow — bold list labels and bold
+        // sentences start with markers or capitals and are unaffected.
+        let should_merge = should_merge
+            || if let Some(prev) = result.last() {
+                let all_bold = |l: &TextLine| {
+                    !l.items.is_empty() && l.items.iter().all(|i: &TextItem| i.is_bold)
+                };
+                let prev_text = prev.text();
+                let prev_trim = prev_text.trim_end();
+                let curr_text = line.text();
+                let curr_trim = curr_text.trim();
+                let y_gap = prev.y - line.y;
+                // Both lines must be tier-less: a tiered/tagged bold heading
+                // followed by bold body text must not absorb it.
+                line_level.is_none()
+                    && effective_heading_level(prev, base_size, heading_tiers, struct_roles)
+                        .is_none()
+                    && prev.page == line.page
+                    && all_bold(prev)
+                    && all_bold(&line)
+                    && y_gap > 0.0
+                    && y_gap < line_font * 1.6
+                    && curr_trim.chars().next().is_some_and(|c| c.is_lowercase())
+                    && !prev_trim.ends_with(['.', ':', ';', '!', '?'])
+                    && prev_trim.split_whitespace().count() + curr_trim.split_whitespace().count()
+                        <= 20
+            } else {
+                false
+            };
+
         if should_merge {
             // Append this line's items to the previous line
             let prev = result.last_mut().unwrap();
@@ -542,6 +582,8 @@ mod tests {
             page: 1,
             is_bold: false,
             is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid,
         }
@@ -683,4 +725,68 @@ mod tests {
             .unwrap();
         assert_eq!(first_header.page, 1, "first occurrence should be on page 1");
     }
+
+    fn make_bold_line(text: &str, page: u32, y: f32) -> TextLine {
+        let mut item = make_item(text, 12.0, None);
+        item.is_bold = true;
+        TextLine {
+            items: vec![item],
+            y,
+            page,
+            adaptive_threshold: 0.10,
+        }
+    }
+
+    #[test]
+    fn merge_wrapped_bold_heading_lowercase_continuation() {
+        // Bold-at-body-size heading wrapped across two lines: the second line
+        // starts lowercase and must merge into the first.
+        let lines = vec![
+            make_bold_line(
+                "3. Perspective of supply and demand balance and cost",
+                1,
+                700.0,
+            ),
+            make_bold_line("structure in Japan", 1, 686.0),
+            make_line("Body text paragraph follows here.", 12.0, 1, 660.0, None),
+        ];
+        let result = merge_heading_lines(lines, 12.0, &[], None);
+        assert_eq!(result.len(), 2, "wrapped bold heading should merge");
+        assert!(result[0].text().contains("cost structure in Japan"));
+    }
+
+    #[test]
+    fn no_merge_for_bold_sentences_or_new_headings() {
+        // Second bold line starts with a capital — a new heading or label,
+        // not a wrap continuation.
+        let lines = vec![
+            make_bold_line("Replace", 1, 700.0),
+            make_bold_line("Trash", 1, 686.0),
+        ];
+        let result = merge_heading_lines(lines, 12.0, &[], None);
+        assert_eq!(result.len(), 2, "distinct bold lines must not merge");
+
+        // Previous line ends a sentence — continuation must not merge.
+        let lines = vec![
+            make_bold_line("This is a bold sentence.", 1, 700.0),
+            make_bold_line("another bold line", 1, 686.0),
+        ];
+        let result = merge_heading_lines(lines, 12.0, &[], None);
+        assert_eq!(result.len(), 2, "sentence-final bold line must not merge");
+    }
+
+    #[test]
+    fn tiered_bold_heading_does_not_absorb_bold_body() {
+        // Previous line is a tier-level bold heading (16pt vs 12pt body);
+        // a following lowercase bold body line must NOT merge into it.
+        let mut heading = make_bold_line("Section Title", 1, 700.0);
+        heading.items[0].font_size = 16.0;
+        heading.items[0].height = 16.0;
+        let lines = vec![
+            heading,
+            make_bold_line("emphasized body text continues here", 1, 686.0),
+        ];
+        let result = merge_heading_lines(lines, 12.0, &[16.0], None);
+        assert_eq!(result.len(), 2, "tiered heading must not absorb bold body");
+    }
 }

src/python.rs

@@ -30,6 +30,9 @@ pub struct PyPdfResult {
     /// 1-indexed page numbers that need OCR.
     #[pyo3(get)]
     pub pages_needing_ocr: Vec<u32>,
+    /// Machine-readable OCR reasons by 1-indexed page.
+    #[pyo3(get)]
+    pub ocr_reasons_by_page: Vec<PyPageOcrReasons>,
     /// Title from PDF metadata.
     #[pyo3(get)]
     pub title: Option<String>,
@@ -60,6 +63,28 @@ impl PyPdfResult {
     }
 }
 
+/// OCR reasons for a single 1-indexed page.
+#[pyclass(name = "PageOcrReasons")]
+#[derive(Clone)]
+pub struct PyPageOcrReasons {
+    /// 1-indexed page number.
+    #[pyo3(get)]
+    pub page: u32,
+    /// Machine-readable OCR reason identifiers.
+    #[pyo3(get)]
+    pub reasons: Vec<String>,
+}
+
+#[pymethods]
+impl PyPageOcrReasons {
+    fn __repr__(&self) -> String {
+        format!(
+            "PageOcrReasons(page={}, reasons={:?})",
+            self.page, self.reasons
+        )
+    }
+}
+
 // ---------------------------------------------------------------------------
 // Classification wrapper (lightweight)
 // ---------------------------------------------------------------------------
@@ -106,6 +131,9 @@ pub struct PyRegionText {
     /// True when the text should not be trusted (empty, GID fonts, garbage, encoding issues).
     #[pyo3(get)]
     pub needs_ocr: bool,
+    /// Machine-readable OCR reason when the cause is known.
+    #[pyo3(get)]
+    pub ocr_reason: Option<String>,
 }
 
 #[pymethods]
@@ -160,6 +188,9 @@ pub struct PyPageMarkdown {
     /// encoding issues, garbage text, or empty extraction).
     #[pyo3(get)]
     pub needs_ocr: bool,
+    /// Machine-readable OCR reason when the cause is known.
+    #[pyo3(get)]
+    pub ocr_reason: Option<String>,
 }
 
 #[pymethods]
@@ -190,6 +221,9 @@ pub struct PyPagesExtractionResult {
     /// 1-indexed pages that need OCR (scanned/image-based or unreliable text).
     #[pyo3(get)]
     pub pages_needing_ocr: Vec<u32>,
+    /// Machine-readable OCR reasons by 1-indexed page.
+    #[pyo3(get)]
+    pub ocr_reasons_by_page: Vec<PyPageOcrReasons>,
     /// True if any page has tables or columns.
     #[pyo3(get)]
     pub is_complex: bool,
@@ -232,7 +266,17 @@ pub struct PyTextItem {
     #[pyo3(get)]
     pub is_italic: bool,
     #[pyo3(get)]
+    pub is_underline: bool,
+    #[pyo3(get)]
+    pub is_strikeout: bool,
+    #[pyo3(get)]
     pub item_type: String,
+    /// Marked Content ID from the content stream's BDC/BMC operator, None
+    /// when the text is not part of marked content. Join with the
+    /// (page, mcid) pairs from extract_structure_elements to attach
+    /// structure-tree roles (headings, paragraphs, ...) in tagged PDFs.
+    #[pyo3(get)]
+    pub mcid: Option<i64>,
 }
 
 #[pymethods]
@@ -248,6 +292,32 @@ impl PyTextItem {
     }
 }
 
+/// One structure-tree element reference from a tagged PDF.
+#[pyclass(name = "StructureElement")]
+#[derive(Clone)]
+pub struct PyStructureElement {
+    /// 1-indexed page number (matches TextItem.page).
+    #[pyo3(get)]
+    pub page: u32,
+    /// Marked Content ID from the page's content stream (matches
+    /// TextItem.mcid).
+    #[pyo3(get)]
+    pub mcid: i64,
+    /// Standard structure type name ("H1".."H6", "P", "Table", "TD", ...).
+    #[pyo3(get)]
+    pub role: String,
+}
+
+#[pymethods]
+impl PyStructureElement {
+    fn __repr__(&self) -> String {
+        format!(
+            "StructureElement(page={}, mcid={}, role='{}')",
+            self.page, self.mcid, self.role
+        )
+    }
+}
+
 // ---------------------------------------------------------------------------
 // Helpers
 // ---------------------------------------------------------------------------
@@ -268,6 +338,7 @@ fn to_py_result(r: crate::PdfProcessResult) -> PyPdfResult {
         page_count: r.page_count,
         processing_time_ms: r.processing_time_ms,
         pages_needing_ocr: r.pages_needing_ocr,
+        ocr_reasons_by_page: to_py_page_ocr_reasons(r.ocr_reasons_by_page),
         title: r.title,
         confidence: r.confidence,
         is_complex_layout: r.layout.is_complex,
@@ -277,6 +348,16 @@ fn to_py_result(r: crate::PdfProcessResult) -> PyPdfResult {
     }
 }
 
+fn to_py_page_ocr_reasons(reasons: Vec<crate::PageOcrReasons>) -> Vec<PyPageOcrReasons> {
+    reasons
+        .into_iter()
+        .map(|reason| PyPageOcrReasons {
+            page: reason.page,
+            reasons: reason.reasons,
+        })
+        .collect()
+}
+
 fn to_py_err(e: crate::PdfError) -> PyErr {
     PyValueError::new_err(e.to_string())
 }
@@ -304,7 +385,21 @@ fn convert_text_items(items: Vec<crate::TextItem>) -> Vec<PyTextItem> {
             page: item.page,
             is_bold: item.is_bold,
             is_italic: item.is_italic,
+            is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item_type_str(&item.item_type),
+            mcid: item.mcid,
+        })
+        .collect()
+}
+
+fn convert_structure_elements(elements: Vec<crate::StructureElement>) -> Vec<PyStructureElement> {
+    elements
+        .into_iter()
+        .map(|e| PyStructureElement {
+            page: e.page,
+            mcid: e.mcid,
+            role: e.role,
         })
         .collect()
 }
@@ -350,11 +445,13 @@ fn to_py_pages_result(r: crate::PagesExtractionResult) -> PyPagesExtractionResul
                 page: p.page,
                 markdown: p.markdown,
                 needs_ocr: p.needs_ocr,
+                ocr_reason: p.ocr_reason,
             })
             .collect(),
         pages_with_tables: r.pages_with_tables,
         pages_with_columns: r.pages_with_columns,
         pages_needing_ocr: r.pages_needing_ocr,
+        ocr_reasons_by_page: to_py_page_ocr_reasons(r.ocr_reasons_by_page),
         is_complex: r.is_complex,
     }
 }
@@ -370,6 +467,7 @@ fn convert_region_results(results: Vec<crate::PageRegionResult>) -> Vec<PyPageRe
                 .map(|r| PyRegionText {
                     text: r.text,
                     needs_ocr: r.needs_ocr,
+                    ocr_reason: r.ocr_reason,
                 })
                 .collect(),
         })
@@ -559,12 +657,56 @@ fn extract_pages_markdown_bytes(
     Ok(to_py_pages_result(result))
 }
 
+/// Extract structure-tree element references from a tagged PDF file.
+///
+/// Parses the document's structure tree (when present) and returns one
+/// entry per marked-content reference, resolved to its 1-indexed page,
+/// MCID, and structure type name ("H1".."H6", "P", "Table", ...). Returns
+/// an empty list when the PDF is not tagged.
+///
+/// Join (page, mcid) against the page/mcid attributes from
+/// [`extract_text_with_positions`] to attach heading levels and other
+/// semantic roles to extracted text.
+///
+/// Args:
+///     path: Path to the PDF file.
+///     pages: Optional list of 1-indexed pages (matching TextItem.page).
+///         When None (default), the whole document is returned.
+///
+/// Returns:
+///     List of StructureElement sorted by (page, mcid).
+#[pyfunction]
+#[pyo3(signature = (path, pages=None))]
+fn extract_structure_elements(
+    path: &str,
+    pages: Option<Vec<u32>>,
+) -> PyResult<Vec<PyStructureElement>> {
+    let elements = crate::extract_structure_elements(path, pages.as_deref()).map_err(to_py_err)?;
+    Ok(convert_structure_elements(elements))
+}
+
+/// Extract structure-tree element references from tagged PDF bytes.
+///
+/// See [`extract_structure_elements`] for details.
+#[pyfunction]
+#[pyo3(signature = (data, pages=None))]
+fn extract_structure_elements_bytes(
+    data: &[u8],
+    pages: Option<Vec<u32>>,
+) -> PyResult<Vec<PyStructureElement>> {
+    let elements =
+        crate::extract_structure_elements_mem(data, pages.as_deref()).map_err(to_py_err)?;
+    Ok(convert_structure_elements(elements))
+}
+
 /// Python module definition.
 #[pymodule]
 fn pdf_inspector(m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_class::<PyPdfResult>()?;
+    m.add_class::<PyPageOcrReasons>()?;
     m.add_class::<PyPdfClassification>()?;
     m.add_class::<PyTextItem>()?;
+    m.add_class::<PyStructureElement>()?;
     m.add_class::<PyRegionText>()?;
     m.add_class::<PyPageRegionTexts>()?;
     m.add_class::<PyPageMarkdown>()?;
@@ -579,6 +721,8 @@ fn pdf_inspector(m: &Bound<'_, PyModule>) -> PyResult<()> {
     m.add_function(wrap_pyfunction!(extract_text_bytes, m)?)?;
     m.add_function(wrap_pyfunction!(extract_text_with_positions, m)?)?;
     m.add_function(wrap_pyfunction!(extract_text_with_positions_bytes, m)?)?;
+    m.add_function(wrap_pyfunction!(extract_structure_elements, m)?)?;
+    m.add_function(wrap_pyfunction!(extract_structure_elements_bytes, m)?)?;
     m.add_function(wrap_pyfunction!(extract_text_in_regions, m)?)?;
     m.add_function(wrap_pyfunction!(extract_text_in_regions_bytes, m)?)?;
     m.add_function(wrap_pyfunction!(extract_pages_markdown, m)?)?;

src/tables/detect_struct.rs

@@ -586,6 +586,8 @@ mod tests {
             page,
             is_bold: false,
             is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid,
         }

src/tables/financial.rs

@@ -108,6 +108,8 @@ pub(crate) fn try_split_financial_item(item: &TextItem) -> Option<Vec<TextItem>>
             page: item.page,
             is_bold: item.is_bold,
             is_italic: item.is_italic,
+            is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item.item_type.clone(),
             mcid: item.mcid,
         });

src/tables/format.rs

@@ -123,6 +123,7 @@ fn format_toc_as_list(cells: &[Vec<String>], footnotes: &[String]) -> String {
 
 /// True when the cell looks like a page number.  Accepts:
 ///   - plain digit tokens: "42", "86 86"
+///   - canonical roman numerals (front-matter pages): "vii", "ix", "xii"
 ///   - dashed section-page IDs: "5-21", "A-1", "B--3", "TC-2" (common in
 ///     technical manuals)
 fn is_page_number_cell(cell: &str) -> bool {
@@ -138,6 +139,9 @@ fn is_page_number_cell(cell: &str) -> bool {
         if all_digits {
             return t.len() <= 4;
         }
+        if super::canonical_roman_value(t).is_some() {
+            return true;
+        }
         // Section-page form: uppercase letters, digits, dashes; at least
         // one digit present.
         t.chars()
@@ -184,6 +188,21 @@ fn starts_with_numbered_label(cell: &str) -> bool {
             .is_some_and(|c| matches!(c, '.' | ')' | '-' | ':'))
 }
 
+fn starts_with_hierarchical_numbered_label(cell: &str) -> bool {
+    let token = cell
+        .split_whitespace()
+        .next()
+        .unwrap_or("")
+        .trim_end_matches(['.', ')', ':', '-']);
+    let levels: Vec<&str> = token.split('.').collect();
+    (2..=4).contains(&levels.len())
+        && levels.iter().all(|level| {
+            !level.is_empty()
+                && level.len() <= 3
+                && level.chars().all(|character| character.is_ascii_digit())
+        })
+}
+
 fn alpha_word_count(cell: &str) -> usize {
     cell.split_whitespace()
         .filter(|word| word.chars().any(|c| c.is_alphabetic()))
@@ -337,11 +356,12 @@ fn clean_table_cells(cells: &[Vec<String>]) -> (Vec<Vec<String>>, Vec<String>) {
         // mid-sentence/lowercase ("continued text here", "with 3.5%...") or
         // carry lowercase fragments in the later cells, so keep those mergeable.
         let looks_like_hierarchical_subrow = first_cell.is_empty()
-            && row.len() >= 3
             && first_non_empty_col == Some(1)
             && looks_like_compact_entry_label(first_non_empty_cell)
-            && ((non_first_cells.len() >= 2 && title_like_later_cells > 0)
+            && ((row.len() == 2 && starts_with_hierarchical_numbered_label(first_non_empty_cell))
+                || (row.len() >= 3 && non_first_cells.len() >= 2 && title_like_later_cells > 0)
                 || (non_first_cells.len() == 1
+                    && row.len() >= 3
                     && prev_first_cell_empty
                     && alpha_word_count(first_non_empty_cell) >= 2));
         let looks_like_new_first_column_entry = !first_cell.is_empty()
@@ -684,6 +704,46 @@ mod tests {
         assert_eq!(cleaned[4][1], "Model training");
     }
 
+    #[test]
+    fn test_clean_table_cells_two_column_numbered_subrows_not_merged() {
+        let cells = vec![
+            vec!["Area".into(), "Competence".into()],
+            vec![
+                "1. Embodying sustainability values".into(),
+                "1.1 Valuing sustainability".into(),
+            ],
+            vec!["".into(), "1.2 Supporting fairness".into()],
+            vec!["".into(), "1.3 Promoting nature".into()],
+            vec![
+                "2. Embracing complexity".into(),
+                "2.1 Systems thinking".into(),
+            ],
+            vec!["".into(), "2.2 Critical thinking".into()],
+        ];
+        let (cleaned, _) = clean_table_cells(&cells);
+
+        assert_eq!(cleaned.len(), 6);
+        assert_eq!(cleaned[2], vec!["", "1.2 Supporting fairness"]);
+        assert_eq!(cleaned[3], vec!["", "1.3 Promoting nature"]);
+        assert_eq!(cleaned[5], vec!["", "2.2 Critical thinking"]);
+    }
+
+    #[test]
+    fn test_clean_table_cells_two_column_numbered_continuation_merges() {
+        let cells = vec![
+            vec!["Area".into(), "Requirement".into()],
+            vec!["Safety".into(), "The program includes".into()],
+            vec!["".into(), "1. First requirement for every operator".into()],
+        ];
+        let (cleaned, _) = clean_table_cells(&cells);
+
+        assert_eq!(cleaned.len(), 2);
+        assert_eq!(
+            cleaned[1][1],
+            "The program includes 1. First requirement for every operator"
+        );
+    }
+
     #[test]
     fn test_clean_table_cells_partial_hierarchical_subrow_not_merged() {
         let cells = vec![

src/tables/grid.rs

@@ -369,6 +369,10 @@ pub(crate) fn join_cell_items(items: &[&TextItem]) -> String {
             let prev_ends_with_hyphen = result.ends_with('-');
             let curr_is_hyphen = text == "-";
             let curr_starts_with_hyphen = text.starts_with('-');
+            let prev_ends_with_open_delimiter =
+                result.ends_with('(') || result.ends_with('[') || result.ends_with('{');
+            let curr_starts_with_close_delimiter =
+                text.starts_with(')') || text.starts_with(']') || text.starts_with('}');
 
             // Detect subscript/superscript: smaller font size and/or Y offset
             let font_ratio = item.font_size / prev_item.font_size;
@@ -385,6 +389,8 @@ pub(crate) fn join_cell_items(items: &[&TextItem]) -> String {
                 || curr_starts_with_hyphen
                 || is_sub_super
                 || was_sub_super
+                || prev_ends_with_open_delimiter
+                || curr_starts_with_close_delimiter
             {
                 result.push_str(text);
             } else {
@@ -430,7 +436,8 @@ pub(crate) fn recover_header_row(
         .iter()
         .enumerate()
         .filter(|(_, item)| {
-            item.font_size > small_font_threshold
+            !item.is_strikeout
+                && item.font_size > small_font_threshold
                 && item.y > first_row_y
                 && item.y <= first_row_y + row_gap_limit
         })
@@ -514,6 +521,8 @@ mod tests {
             page: 1,
             is_bold: false,
             is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -727,6 +736,18 @@ mod tests {
         assert_eq!(join_cell_items(&[&a, &b, &c]), "pre-fix");
     }
 
+    #[test]
+    fn test_join_cell_items_parenthetical_no_inner_spaces() {
+        let a = make_item("The first sentence", 100.0, 500.0, 10.0);
+        let b = make_item("(", 190.0, 500.0, 10.0);
+        let c = make_item("twice", 195.0, 500.0, 10.0);
+        let d = make_item(")", 220.0, 500.0, 10.0);
+        assert_eq!(
+            join_cell_items(&[&a, &b, &c, &d]),
+            "The first sentence (twice)"
+        );
+    }
+
     #[test]
     fn test_join_cell_items_subscript_no_space() {
         let a = make_item("H", 100.0, 500.0, 12.0);
@@ -784,6 +805,31 @@ mod tests {
         assert_eq!(table.rows.len(), rows_before);
     }
 
+    #[test]
+    fn test_recover_header_row_skips_strikeout_candidates() {
+        let mut old_col1 = make_item("Old Col1", 100.0, 520.0, 12.0);
+        old_col1.is_strikeout = true;
+        let mut old_col2 = make_item("Old Col2", 200.0, 520.0, 12.0);
+        old_col2.is_strikeout = true;
+        let all_items = vec![
+            old_col1,
+            old_col2,
+            make_item("A", 100.0, 500.0, 8.0),
+            make_item("B", 200.0, 500.0, 8.0),
+        ];
+        let mut table = Table {
+            columns: vec![100.0, 200.0],
+            rows: vec![500.0, 480.0],
+            cells: vec![vec!["A".into(), "B".into()], vec!["C".into(), "D".into()]],
+            item_indices: vec![2, 3],
+            kind: TableKind::Data,
+        };
+
+        recover_header_row(&mut table, &all_items, 9.0);
+        assert_eq!(table.rows.len(), 2);
+        assert_eq!(table.cells[0], vec!["A", "B"]);
+    }
+
     #[test]
     fn test_recover_header_row_too_far_above() {
         let all_items = vec![
@@ -866,6 +912,8 @@ mod tests {
                         font: String::new(),
                         is_bold: false,
                         is_italic: false,
+                        is_underline: false,
+                        is_strikeout: false,
                         item_type: ItemType::Text,
                         mcid: None,
                         page: 1,
@@ -902,6 +950,8 @@ mod tests {
                         font: String::new(),
                         is_bold: false,
                         is_italic: false,
+                        is_underline: false,
+                        is_strikeout: false,
                         item_type: ItemType::Text,
                         mcid: None,
                         page: 1,

src/text_quality.rs

@@ -0,0 +1,520 @@
+//! Text-quality detection: deciding when an extracted text layer is too broken
+//! to serve and a page should fall back to OCR.
+//!
+//! Extraction can produce plausible-looking bytes that are actually garbage —
+//! failed CID→Unicode mappings, broken ToUnicode CMaps, mojibake. These
+//! detectors catch that and let callers set `needs_ocr`. They come in two
+//! layers, sharing the same primitives:
+//!
+//! - **Markdown-level** ([`detect_encoding_issues`], [`is_garbage_text`],
+//!   [`is_cid_garbage`]) run on a page's final markdown string. Used as a
+//!   backstop on the region-extraction and whole-document paths.
+//! - **Item/span-level** ([`analyze_text_quality`],
+//!   [`region_items_have_decoding_issue`]) run on individual `TextItem`s and
+//!   accumulate per-page evidence, so localized garbled spans on an otherwise
+//!   clean page are caught without a single span having to condemn the page.
+//!
+//! Detection classes, roughly by signal:
+//! - **Replacement runs**: U+FFFD clusters ([`has_replacement_text_run`]).
+//! - **Private-use / C1-control runs**: CID passthrough landing in PUA or the
+//!   C1 block ([`has_private_use_text_run`], [`has_cid_control_token`]).
+//! - **Dollar-as-space**: `Word$Word$Word` from broken CMaps
+//!   ([`has_dollar_as_space_pattern`]).
+//! - **Non-alphanumeric dominance**: symbol soup ([`is_garbage_text`]).
+//! - **Substitution-cipher letter statistics**: pure-ASCII output whose letter
+//!   distribution is a permutation of natural language ([`CipherGarbleStats`]).
+
+use crate::types::TextItem;
+use crate::{add_ocr_reason, OCR_REASON_SUSPECTED_GARBLED_TEXT};
+use std::collections::BTreeMap;
+
+/// Detect broken font encodings in extracted markdown text.
+///
+/// Two heuristics:
+/// 1. **U+FFFD**: Any replacement character indicates decode failures.
+/// 2. **Dollar-as-space**: Pattern like `Word$Word$Word` where `$` is used as a
+///    word separator due to broken ToUnicode CMaps. Triggers when either:
+///    - More than 50% of `$` are between letters (clear substitution pattern), OR
+///    - More than 20 letter-dollar-letter occurrences (even if some `$` are also
+///      used as trailing/leading separators, 20+ is far beyond normal financial text).
+pub(crate) fn detect_encoding_issues(markdown: &str) -> bool {
+    // Heuristic 1: U+FFFD replacement characters
+    if markdown.contains('\u{FFFD}') {
+        return true;
+    }
+
+    // Heuristic 2: dollar-as-space pattern
+    if has_dollar_as_space_pattern(markdown) {
+        return true;
+    }
+
+    // Heuristic 3: substitution-cipher letter statistics (broken ToUnicode)
+    let mut stats = CipherGarbleStats::default();
+    stats.add_text(markdown);
+    stats.looks_garbled()
+}
+
+fn has_dollar_as_space_pattern(markdown: &str) -> bool {
+    let total_dollars = markdown.matches('$').count();
+    if total_dollars > 10 {
+        let bytes = markdown.as_bytes();
+        let mut letter_dollar_letter = 0usize;
+        for i in 1..bytes.len().saturating_sub(1) {
+            if bytes[i] == b'$'
+                && bytes[i - 1].is_ascii_alphabetic()
+                && bytes[i + 1].is_ascii_alphabetic()
+            {
+                letter_dollar_letter += 1;
+            }
+        }
+        if letter_dollar_letter > 20 || letter_dollar_letter * 2 > total_dollars {
+            return true;
+        }
+    }
+
+    false
+}
+
+/// English letter frequencies (percent, a–z). Used as a natural-language
+/// reference: every Latin-script language in the eval corpus (Swedish,
+/// Finnish, Turkish, German, romaji) scores ≥ 0.80 cosine similarity against
+/// it, while substitution-cipher text scores ~0.53.
+const ENGLISH_LETTER_FREQ: [f64; 26] = [
+    8.2, 1.5, 2.8, 4.3, 12.7, 2.2, 2.0, 6.1, 7.0, 0.15, 0.8, 4.0, 2.4, 6.7, 7.5, 1.9, 0.1, 6.0,
+    6.3, 9.1, 2.8, 1.0, 2.4, 0.15, 2.0, 0.07,
+];
+
+/// Letter statistics for detecting substitution-cipher garbling: broken
+/// ToUnicode CMaps that shift every character by a per-range constant (e.g.
+/// `Certificate` extracted as `8VceZWZTReV`). Such text is 100% printable
+/// ASCII with word-like token lengths, so it defeats `is_garbage_text` and
+/// produces no replacement characters — it needs its own discriminator.
+#[derive(Debug, Default)]
+struct CipherGarbleStats {
+    /// Case-folded ASCII letter histogram.
+    letter_counts: [u32; 26],
+    ascii_letters: usize,
+    ascii_vowels: usize,
+    /// Accented Latin letters (Latin-1 Supplement through Latin Extended-B,
+    /// plus Latin Extended Additional). Count toward Latin dominance only.
+    latin_ext_letters: usize,
+    non_latin_letters: usize,
+    /// Adjacent ASCII-letter pairs, and how many of them switch from
+    /// lowercase straight to uppercase mid-word.
+    letter_bigrams: usize,
+    case_shift_bigrams: usize,
+}
+
+impl CipherGarbleStats {
+    fn add_text(&mut self, text: &str) {
+        let mut prev: Option<char> = None;
+        for ch in text.chars() {
+            if ch.is_ascii_alphabetic() {
+                let idx = (ch.to_ascii_lowercase() as u8 - b'a') as usize;
+                self.letter_counts[idx] += 1;
+                self.ascii_letters += 1;
+                if matches!(ch.to_ascii_lowercase(), 'a' | 'e' | 'i' | 'o' | 'u') {
+                    self.ascii_vowels += 1;
+                }
+                if let Some(p) = prev {
+                    self.letter_bigrams += 1;
+                    if p.is_ascii_lowercase() && ch.is_ascii_uppercase() {
+                        self.case_shift_bigrams += 1;
+                    }
+                }
+                prev = Some(ch);
+            } else {
+                if ch.is_alphabetic() {
+                    if matches!(ch as u32, 0xC0..=0x24F | 0x1E00..=0x1EFF) {
+                        self.latin_ext_letters += 1;
+                    } else {
+                        self.non_latin_letters += 1;
+                    }
+                }
+                prev = None;
+            }
+        }
+    }
+
+    /// Cosine similarity between the observed letter histogram and English
+    /// letter frequencies. A shifted alphabet permutes the histogram, which
+    /// destroys the similarity regardless of the shift amount.
+    fn english_cosine(&self) -> f64 {
+        if self.ascii_letters == 0 {
+            return 1.0;
+        }
+        let n = self.ascii_letters as f64;
+        let mut dot = 0.0;
+        let mut norm_obs = 0.0;
+        for (count, freq) in self.letter_counts.iter().zip(ENGLISH_LETTER_FREQ) {
+            let p = *count as f64 / n;
+            dot += p * freq;
+            norm_obs += p * p;
+        }
+        let norm_en = ENGLISH_LETTER_FREQ
+            .iter()
+            .map(|f| f * f)
+            .sum::<f64>()
+            .sqrt();
+        dot / (norm_obs.sqrt() * norm_en)
+    }
+
+    /// Cosine similarity between the observed histogram and English
+    /// frequencies after sorting BOTH descending — i.e. comparing the *shape*
+    /// of the frequency profile, ignoring which letter sits where. A
+    /// substitution cipher is a bijection, so it preserves this shape exactly
+    /// (att10k 0.97, arbitrary shifts 0.99) regardless of case or offset.
+    /// Non-linguistic ASCII has a different profile: a small alphabet is far
+    /// steeper (random DNA 0.74, hex dumps 0.81), so the shape diverges.
+    fn english_shape_cosine(&self) -> f64 {
+        if self.ascii_letters == 0 {
+            return 1.0;
+        }
+        let n = self.ascii_letters as f64;
+        let mut obs: [f64; 26] = std::array::from_fn(|i| self.letter_counts[i] as f64 / n);
+        obs.sort_unstable_by(|a, b| b.total_cmp(a));
+        let mut en = ENGLISH_LETTER_FREQ;
+        en.sort_unstable_by(|a, b| b.total_cmp(a));
+
+        let dot: f64 = obs.iter().zip(en).map(|(o, e)| o * e).sum();
+        let norm_obs = obs.iter().map(|o| o * o).sum::<f64>().sqrt();
+        let norm_en = en.iter().map(|e| e * e).sum::<f64>().sqrt();
+        dot / (norm_obs * norm_en)
+    }
+
+    /// Thresholds validated against the 380-document pdf-evals snapshot
+    /// corpus (0 false positives) and the garbled ParseBench `att10k` page
+    /// (vowel ratio 0.245, case-shift rate 0.225, cosine 0.532). Closest
+    /// legitimate document on each axis: vowel ratio 0.264 (circuit
+    /// schematic), case-shift rate 0.021, cosine 0.801.
+    fn looks_garbled(&self) -> bool {
+        // Need a statistically meaningful, Latin-dominant sample.
+        if self.ascii_letters < 200
+            || self.non_latin_letters > self.ascii_letters + self.latin_ext_letters
+        {
+            return false;
+        }
+
+        // Real Latin-script text keeps vowels above ~30% of letters even in
+        // acronym- and part-number-heavy documents; shifted text starves them.
+        let vowel_ratio = self.ascii_vowels as f64 / self.ascii_letters as f64;
+        if vowel_ratio > 0.30 {
+            return false;
+        }
+
+        // Signal 1: lowercase→uppercase transitions inside words. A shifted
+        // lowercase alphabet straddles the ASCII uppercase block ('i'→'Z',
+        // 't'→'e'), so garbled words flip case constantly. Real documents
+        // stay ≤ 0.02 even with camelCase identifiers.
+        let case_shifts = self.letter_bigrams >= 100
+            && self.case_shift_bigrams as f64 >= self.letter_bigrams as f64 * 0.10;
+
+        // Signal 2: the histogram is a permutation of natural language — an
+        // English-like frequency SHAPE (sorted cosine high) but with letters
+        // in the wrong POSITIONS (unsorted cosine low). This is the signature
+        // of a substitution cipher and is case-independent, so it catches
+        // all-lowercase and all-uppercase shifts as well as case-straddling
+        // ones. Genuinely non-linguistic ASCII that is merely "unlike English"
+        // fails one of the two halves: DNA/hex dumps have too steep a profile
+        // (shape cosine < 0.90), while protein sequences, ticker symbols and
+        // base64 are not sufficiently unlike English in position (unsorted
+        // cosine ≥ 0.60) — so none of them are routed to OCR.
+        let permuted_language = self.english_cosine() < 0.60 && self.english_shape_cosine() >= 0.90;
+
+        case_shifts || permuted_language
+    }
+}
+
+#[derive(Debug, Default)]
+pub(crate) struct TextQualityReport {
+    pub(crate) pages_needing_ocr: Vec<u32>,
+    pub(crate) has_encoding_issues: bool,
+    pub(crate) reasons_by_page: BTreeMap<u32, Vec<String>>,
+}
+
+#[derive(Debug, Default)]
+struct PageTextQualityEvidence {
+    chars: usize,
+    replacement_chars: usize,
+    replacement_spans: usize,
+    longest_replacement_run: usize,
+    cipher_garble: CipherGarbleStats,
+}
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq)]
+enum TextSpanIssueKind {
+    Replacement,
+    Strong,
+}
+
+pub(crate) fn analyze_text_quality(items: &[TextItem]) -> TextQualityReport {
+    let mut reasons_by_page = BTreeMap::new();
+    let mut evidence_by_page = BTreeMap::<u32, PageTextQualityEvidence>::new();
+
+    for item in items {
+        if !matches!(item.item_type, crate::types::ItemType::Text) {
+            continue;
+        }
+
+        let evidence = evidence_by_page.entry(item.page).or_default();
+        evidence.chars += item.text.chars().filter(|ch| !ch.is_whitespace()).count();
+        evidence.cipher_garble.add_text(&item.text);
+
+        match text_span_decoding_issue_kind(&item.text) {
+            Some(TextSpanIssueKind::Strong) => {
+                add_ocr_reason(
+                    &mut reasons_by_page,
+                    item.page,
+                    OCR_REASON_SUSPECTED_GARBLED_TEXT,
+                );
+            }
+            Some(TextSpanIssueKind::Replacement) => {
+                let stats = replacement_text_stats(&item.text);
+                evidence.replacement_chars += stats.0;
+                evidence.replacement_spans += 1;
+                evidence.longest_replacement_run = evidence.longest_replacement_run.max(stats.1);
+            }
+            None => {}
+        }
+    }
+
+    for (page, evidence) in evidence_by_page {
+        if reasons_by_page.contains_key(&page) {
+            continue;
+        }
+        if page_replacement_evidence_needs_ocr(&evidence) || evidence.cipher_garble.looks_garbled()
+        {
+            add_ocr_reason(
+                &mut reasons_by_page,
+                page,
+                OCR_REASON_SUSPECTED_GARBLED_TEXT,
+            );
+        }
+    }
+
+    let pages_needing_ocr: Vec<u32> = reasons_by_page.keys().copied().collect();
+    TextQualityReport {
+        has_encoding_issues: !pages_needing_ocr.is_empty(),
+        pages_needing_ocr,
+        reasons_by_page,
+    }
+}
+
+pub(crate) fn region_items_have_decoding_issue(items: &[TextItem]) -> bool {
+    items.iter().any(|item| {
+        matches!(item.item_type, crate::types::ItemType::Text)
+            && text_span_has_decoding_issue(&item.text)
+    })
+}
+
+fn text_span_has_decoding_issue(text: &str) -> bool {
+    text_span_decoding_issue_kind(text).is_some()
+}
+
+fn text_span_decoding_issue_kind(text: &str) -> Option<TextSpanIssueKind> {
+    let text = text.trim();
+    if text.is_empty() {
+        return None;
+    }
+
+    if has_dollar_as_space_pattern(text)
+        || has_private_use_text_run(text)
+        || is_cid_garbage(text)
+        || has_cid_control_token(text)
+    {
+        return Some(TextSpanIssueKind::Strong);
+    }
+
+    if has_replacement_text_run(text) {
+        return Some(TextSpanIssueKind::Replacement);
+    }
+
+    None
+}
+
+fn replacement_text_stats(text: &str) -> (usize, usize) {
+    let mut replacement = 0usize;
+    let mut current_run = 0usize;
+    let mut longest_run = 0usize;
+
+    for ch in text.chars() {
+        if ch == '\u{FFFD}' {
+            replacement += 1;
+            current_run += 1;
+            longest_run = longest_run.max(current_run);
+        } else {
+            current_run = 0;
+        }
+    }
+
+    (replacement, longest_run)
+}
+
+fn page_replacement_evidence_needs_ocr(evidence: &PageTextQualityEvidence) -> bool {
+    if evidence.replacement_chars == 0 || evidence.chars == 0 {
+        return false;
+    }
+
+    // If the entire page is only a short broken text layer, even a short
+    // replacement run is enough evidence. On otherwise text-heavy pages,
+    // require density so math formulas do not force full-page OCR.
+    if evidence.chars <= 80 && evidence.longest_replacement_run >= 2 {
+        return true;
+    }
+
+    let replacement_density_bps = evidence.replacement_chars * 10_000 / evidence.chars;
+    let enough_bad_text = evidence.replacement_chars >= 12 && replacement_density_bps >= 500;
+    let repeated_bad_spans = evidence.replacement_spans >= 3 && replacement_density_bps >= 250;
+    let long_bad_run = evidence.longest_replacement_run >= 8 && replacement_density_bps >= 250;
+
+    enough_bad_text || repeated_bad_spans || long_bad_run
+}
+
+fn has_replacement_text_run(text: &str) -> bool {
+    let (replacement, longest_run) = replacement_text_stats(text);
+    longest_run >= 2 || replacement >= 3
+}
+
+fn has_private_use_text_run(text: &str) -> bool {
+    let mut total = 0usize;
+    let mut private_use = 0usize;
+    let mut current_run = 0usize;
+    let mut longest_run = 0usize;
+
+    for ch in text.chars() {
+        if ch.is_whitespace() {
+            current_run = 0;
+            continue;
+        }
+        total += 1;
+        if is_private_use_char(ch) {
+            private_use += 1;
+            current_run += 1;
+            longest_run = longest_run.max(current_run);
+        } else {
+            current_run = 0;
+        }
+    }
+
+    if private_use == 0 {
+        return false;
+    }
+
+    longest_run >= 3 || (total >= 5 && private_use >= 2 && private_use * 2 >= total)
+}
+
+fn has_cid_control_token(text: &str) -> bool {
+    text.split_whitespace().any(token_has_cid_control)
+}
+
+fn token_has_cid_control(token: &str) -> bool {
+    let mut total = 0usize;
+    let mut c1_control = 0usize;
+
+    for ch in token.chars() {
+        total += 1;
+        if ('\u{0080}'..='\u{009F}').contains(&ch) {
+            c1_control += 1;
+        }
+    }
+
+    total >= 5 && c1_control >= 2 && c1_control * 20 >= total
+}
+
+fn is_private_use_char(ch: char) -> bool {
+    matches!(
+        ch as u32,
+        0xE000..=0xF8FF | 0xF0000..=0xFFFFD | 0x100000..=0x10FFFD
+    )
+}
+
+/// Check if extracted text is predominantly garbage (non-alphanumeric).
+///
+/// Broken font encodings produce text like "----1-.-.-.___  --.-. .._ I_---."
+/// where most characters are punctuation/symbols. Real text in any language
+/// has >50% alphanumeric characters.
+pub(crate) fn is_garbage_text(markdown: &str) -> bool {
+    let mut alphanum = 0usize;
+    let mut non_alphanum = 0usize;
+
+    let chars: Vec<char> = markdown.chars().collect();
+    let mut i = 0usize;
+    while i < chars.len() {
+        let ch = chars[i];
+        let mut run_end = i + 1;
+        while run_end < chars.len() && chars[run_end] == ch {
+            run_end += 1;
+        }
+
+        let is_decorative_leader = matches!(ch, '.' | '_' | '·') && run_end - i >= 3;
+        if !is_decorative_leader {
+            for &run_ch in &chars[i..run_end] {
+                if run_ch.is_whitespace() {
+                    continue;
+                }
+                // Skip markdown syntax chars that we add (not from the PDF)
+                if matches!(run_ch, '#' | '*' | '|' | '-' | '\n') {
+                    continue;
+                }
+                if run_ch.is_alphanumeric() {
+                    alphanum += 1;
+                } else {
+                    non_alphanum += 1;
+                }
+            }
+        }
+        i = run_end;
+    }
+
+    let total = alphanum + non_alphanum;
+    total >= 50 && alphanum * 2 < total
+}
+
+/// Detect garbage from failed CID-to-Unicode mapping on Identity-H fonts.
+///
+/// When CID values don't correspond to Unicode codepoints, the raw bytes often
+/// produce characters in the C1 control range (U+0080–U+009F) or Private Use
+/// Area, mixed with random Latin Extended characters.  Valid text in any
+/// language almost never contains C1 controls.  We also fall back to the
+/// general `is_garbage_text` check for non-alphanumeric-heavy patterns.
+pub(crate) fn is_cid_garbage(text: &str) -> bool {
+    if is_garbage_text(text) {
+        return true;
+    }
+    let mut total = 0usize;
+    let mut c1_control = 0usize;
+    let mut high_latin = 0usize;
+    for ch in text.chars() {
+        if ch.is_whitespace() {
+            continue;
+        }
+        total += 1;
+        // C1 control characters (U+0080–U+009F) — almost never in real text
+        if ch == '·' {
+            continue;
+        }
+        if ('\u{0080}'..='\u{009F}').contains(&ch) {
+            c1_control += 1;
+        }
+        // High Latin-1 (U+00A0–U+00FF) — legitimate in Western European text
+        // but when combined with ASCII in CID passthrough, indicates mojibake
+        // from CID values being misinterpreted as Latin-1 characters.
+        if ('\u{00A0}'..='\u{00FF}').contains(&ch) {
+            high_latin += 1;
+        }
+    }
+    if total < 5 {
+        return false;
+    }
+    // If ≥5% of non-whitespace chars are C1 controls, it's garbage
+    if c1_control >= 2 && c1_control * 20 >= total {
+        return true;
+    }
+    // If ≥40% of non-whitespace chars are high Latin-1 AND the text has few
+    // ASCII letters, it's likely CID-as-Latin-1 mojibake (Japanese/CJK PDFs
+    // where CID values 0x80-0xFF become accented Latin characters).  Keep a
+    // minimum length so short math tokens like "2×()×" do not route a clean
+    // page to OCR.
+    let ascii_letters = text.chars().filter(|c| c.is_ascii_alphabetic()).count();
+    total >= 20 && high_latin * 5 >= total * 2 && ascii_letters * 3 < total
+}

src/text_utils.rs

@@ -7,6 +7,83 @@
 use crate::types::TextItem;
 use unicode_normalization::UnicodeNormalization;
 
+/// Return whether text is an explicit page-number expression.
+///
+/// This strict form is suitable before layout, where removing one numeric item
+/// from substantive text such as `Page 42 explains the result` would lose data.
+pub(crate) fn is_explicit_page_number_expression(text: &str) -> bool {
+    let trimmed = text.trim();
+    if trimmed.is_empty() {
+        return false;
+    }
+
+    let is_number = |value: &str| {
+        !value.is_empty() && value.chars().all(|character| character.is_ascii_digit())
+    };
+
+    if trimmed.len() <= 4 && is_number(trimmed) {
+        return true;
+    }
+
+    if trimmed.len() >= 3 && trimmed.starts_with('-') && trimmed.ends_with('-') {
+        let inner = trimmed[1..trimmed.len() - 1].trim();
+        if is_number(inner) {
+            return true;
+        }
+    }
+
+    let lowercase = trimmed.to_ascii_lowercase();
+    if let Some(rest) = lowercase.strip_prefix("page") {
+        let words: Vec<&str> = rest.split_whitespace().collect();
+        if words.len() >= 3 && is_number(words[0]) && words[1] == "of" && is_number(words[2]) {
+            return true;
+        }
+        if words.len() >= 2 && words[0] == "of" && is_number(words[1]) {
+            return true;
+        }
+        return match words.as_slice() {
+            [] | ["of"] => true,
+            [number] => is_number(number),
+            ["of", total] => is_number(total),
+            [number, "of", total] => is_number(number) && is_number(total),
+            _ => false,
+        };
+    }
+
+    let words: Vec<&str> = lowercase.split_whitespace().collect();
+    match words.as_slice() {
+        [number, "of", total] => is_number(number) && is_number(total),
+        _ => false,
+    }
+}
+
+/// Return whether a completed Markdown line looks like a page number or a
+/// labeled running header.
+///
+/// At this stage the complete line and surrounding breaks are available, so a
+/// leading `Page N` remains compatible with the existing header cleanup even
+/// when the PDF appends a chapter or document title.
+pub(crate) fn is_page_number_line(text: &str) -> bool {
+    if is_explicit_page_number_expression(text) {
+        return true;
+    }
+
+    let lowercase = text.trim().to_ascii_lowercase();
+    lowercase.strip_prefix("page").is_some_and(|rest| {
+        let mut characters = rest.trim_start().chars().peekable();
+        let mut has_page_number = false;
+        while characters
+            .peek()
+            .is_some_and(|character| character.is_ascii_digit())
+        {
+            has_page_number = true;
+            characters.next();
+        }
+
+        has_page_number && characters.next().is_none_or(char::is_whitespace)
+    })
+}
+
 /// Check if a character is CJK (Chinese, Japanese, Korean).
 /// CJK languages don't use spaces between words, so word-boundary
 /// heuristics should not apply when CJK characters are involved.
@@ -93,6 +170,9 @@ pub fn is_bold_font(font_name: &str) -> bool {
         || lower.contains("extrabold")
         || lower.contains("ultrabold")
         || lower.contains("medium") && !lower.contains("mediumitalic") // Some fonts use Medium for semi-bold
+        // URW Type 1 fonts abbreviate Medium as "Medi" (e.g. NimbusRomNo9L-Medi,
+        // the Times-Bold substitute in LaTeX documents; -MediItal is bold italic).
+        || lower.contains("-medi") && !lower.contains("mediumital")
 }
 
 /// Detect if a font name indicates italic/oblique style
@@ -762,6 +842,17 @@ mod tests {
     use super::*;
     use crate::types::ItemType;
 
+    #[test]
+    fn bold_font_urw_medi_abbreviation() {
+        // URW Type 1 fonts (LaTeX default Times) abbreviate Medium as "Medi"
+        assert!(is_bold_font("NROFIU+NimbusRomNo9L-Medi"));
+        assert!(is_bold_font("NimbusRomNo9L-MediItal"));
+        assert!(!is_bold_font("DSSZWN+NimbusRomNo9L-Regu"));
+        assert!(!is_bold_font("NimbusRomNo9L-ReguItal"));
+        // Medium-Italic exclusion still holds
+        assert!(!is_bold_font("Foo-MediumItalic"));
+    }
+
     #[test]
     fn strip_soft_hyphen() {
         assert_eq!(expand_ligatures("con\u{00AD}tent"), "content");
@@ -883,6 +974,8 @@ mod tests {
             page: 1,
             is_bold: false,
             is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -1002,6 +1095,8 @@ mod tests {
                 page: 1,
                 is_bold: false,
                 is_italic: false,
+                is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             });
@@ -1078,6 +1173,8 @@ mod tests {
             page: 1,
             is_bold: false,
             is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/tounicode.rs

@@ -4,11 +4,17 @@
 
 use log::{debug, warn};
 use lopdf::{Document, Object, ObjectId};
+use std::borrow::Cow;
 use std::collections::{HashMap, HashSet};
+#[cfg(not(target_arch = "wasm32"))]
 use std::path::{Path, PathBuf};
 
 use crate::glyph_names::glyph_to_char;
 
+#[cfg(target_arch = "wasm32")]
+static BUILTIN_CMAPS: include_dir::Dir<'_> =
+    include_dir::include_dir!("$CARGO_MANIFEST_DIR/external/bcmaps");
+
 /// A parsed ToUnicode CMap mapping CIDs to Unicode strings
 #[derive(Debug, Default, Clone)]
 pub struct ToUnicodeCMap {
@@ -329,7 +335,7 @@ impl ToUnicodeCMap {
                 if let (Some(start), Some(end), Some(base)) = (
                     parse_hex_u16(&start_hex),
                     parse_hex_u16(&end_hex),
-                    parse_hex_u32(&base_hex),
+                    hex_to_unicode_scalar(&base_hex),
                 ) {
                     self.ranges.push((start, end, base));
                 }
@@ -575,32 +581,86 @@ fn parse_hex_u16(hex: &str) -> Option<u16> {
     u16::from_str_radix(hex.trim(), 16).ok()
 }
 
-/// Parse a hex string to u32
-fn parse_hex_u32(hex: &str) -> Option<u32> {
-    u32::from_str_radix(hex.trim(), 16).ok()
-}
-
-/// Convert a hex string to a Unicode string
-/// Handles both 2-byte (BMP) and 4-byte (supplementary) codepoints
+/// Convert a ToUnicode destination hex string to Unicode.
+///
+/// PDF ToUnicode destinations are UTF-16BE strings. Supplementary-plane
+/// characters are encoded as surrogate pairs, so treating each 4-hex chunk as
+/// a scalar drops emoji like D83CDF1F.
 fn hex_to_unicode_string(hex: &str) -> Option<String> {
-    let hex = hex.trim();
-    let mut result = String::new();
-
-    // Process 4 hex digits at a time
-    let mut i = 0;
-    while i + 4 <= hex.len() {
-        if let Ok(cp) = u32::from_str_radix(&hex[i..i + 4], 16) {
-            if let Some(c) = char::from_u32(cp) {
-                result.push(c);
-            }
-        }
-        i += 4;
+    let hex: String = hex.chars().filter(|ch| !ch.is_ascii_whitespace()).collect();
+    if hex.is_empty() || !hex.len().is_multiple_of(2) {
+        return None;
     }
 
-    if result.is_empty() {
-        None
+    let bytes: Option<Vec<u8>> = (0..hex.len())
+        .step_by(2)
+        .map(|i| u8::from_str_radix(hex.get(i..i + 2)?, 16).ok())
+        .collect();
+    let bytes = bytes?;
+
+    if bytes.len().is_multiple_of(2) {
+        let units: Vec<u16> = bytes
+            .chunks_exact(2)
+            .map(|chunk| u16::from_be_bytes([chunk[0], chunk[1]]))
+            .collect();
+        if let Ok(result) = String::from_utf16(&units) {
+            if !result.is_empty() {
+                return Some(normalize_tounicode_destination(result));
+            }
+        }
+    }
+
+    // Be permissive for non-standard one-byte destinations.
+    if bytes.len() == 1 {
+        let ch = bytes[0] as char;
+        if !ch.is_control() || ch == '\t' || ch == '\n' {
+            return Some(ch.to_string());
+        }
+    }
+
+    None
+}
+
+fn normalize_tounicode_destination(text: String) -> String {
+    let is_multi_char = text.chars().nth(1).is_some();
+
+    // Some malformed producer CMaps put a list of alternative whitespace or
+    // hyphen codepoints into one destination. Keep ordinary multi-character
+    // mappings intact unless that malformed signature is present.
+    if is_multi_char
+        && text.chars().all(char::is_whitespace)
+        && text.chars().any(|ch| matches!(ch, '\t' | '\n' | '\r'))
+    {
+        return if text.contains('\t') {
+            "\t".to_string()
+        } else {
+            " ".to_string()
+        };
+    }
+
+    if is_multi_char
+        && text.contains('\u{00ad}')
+        && text.chars().all(|ch| {
+            matches!(
+                ch,
+                '-' | '\u{00ad}' | '\u{2010}' | '\u{2011}' | '\u{2012}' | '\u{2013}' | '\u{2212}'
+            )
+        })
+    {
+        return "-".to_string();
+    }
+
+    text
+}
+
+fn hex_to_unicode_scalar(hex: &str) -> Option<u32> {
+    let text = hex_to_unicode_string(hex)?;
+    let mut chars = text.chars();
+    let ch = chars.next()?;
+    if chars.next().is_none() {
+        Some(ch as u32)
     } else {
-        Some(result)
+        None
     }
 }
 
@@ -820,6 +880,25 @@ fn try_remap_subset_cmap(
         None => return (cmap, None),
     };
 
+    // Both repair paths below assume CIDs are glyph indices that a subsetter can
+    // renumber, which is only true for CIDFontType2 (TrueType). For CIDFontType0
+    // (CFF), CIDs are resolved through the CFF charset, so a valid CMap stays valid
+    // after subsetting and renumbering it corrupts otherwise-correct text.
+    // CIDToGIDMap is likewise CIDFontType2-only (PDF 32000-1:2008, 9.7.4.2), so this
+    // also ignores a CIDToGIDMap that a malformed producer attached to a CFF font.
+    // /Subtype may be an indirect reference, so resolve it through the document.
+    // Only bail out when the descendant is *explicitly* something other than
+    // CIDFontType2: a missing or unresolvable /Subtype keeps the previous
+    // behaviour rather than silently disabling the repair.
+    let subtype = cid_font_dict.get(b"Subtype").ok().and_then(|o| match o {
+        Object::Reference(r) => doc.get_object(*r).ok().and_then(|o| o.as_name().ok()),
+        other => other.as_name().ok(),
+    });
+    if subtype.is_some_and(|name| name != b"CIDFontType2") {
+        debug!("Subset remap skipped for obj={obj_num}: descendant is not CIDFontType2");
+        return (cmap, None);
+    }
+
     // If there's an explicit CIDToGIDMap, build a repaired CMap using it.
     if let Some(cid_to_gid) = get_cid_to_gid_map(cid_font_dict, doc) {
         if let Some(repaired) = build_cmap_with_cid_to_gid_map(&cmap, &cid_to_gid) {
@@ -1095,9 +1174,7 @@ fn build_gid_to_unicode(face: &ttf_parser::Face<'_>) -> Option<HashMap<u16, char
 /// Build a ToUnicodeCMap from pdf.js built-in binary CMaps (bcmaps).
 fn build_cmap_from_builtin_cmap(ordering: &str) -> Option<ToUnicodeCMap> {
     let name = format!("Adobe-{}-UCS2.bcmap", ordering);
-    let dir = find_bcmaps_dir()?;
-    let path = dir.join(name);
-    let data = std::fs::read(&path).ok()?;
+    let data = read_builtin_cmap_file(&name)?;
     let mut cmap = parse_binary_cmap(&data).ok()?;
     if cmap.char_map.is_empty() && cmap.ranges.is_empty() {
         return None;
@@ -1105,13 +1182,14 @@ fn build_cmap_from_builtin_cmap(ordering: &str) -> Option<ToUnicodeCMap> {
     cmap.code_byte_length = 2;
     debug!(
         "Built-in CMap {}: char_map={} ranges={}",
-        path.display(),
+        name,
         cmap.char_map.len(),
         cmap.ranges.len()
     );
     Some(cmap)
 }
 
+#[cfg(not(target_arch = "wasm32"))]
 fn find_bcmaps_dir() -> Option<PathBuf> {
     if let Ok(dir) = std::env::var("PDF_INSPECTOR_BCMAPS_DIR") {
         let p = PathBuf::from(dir);
@@ -1128,6 +1206,18 @@ fn find_bcmaps_dir() -> Option<PathBuf> {
     None
 }
 
+#[cfg(not(target_arch = "wasm32"))]
+fn read_builtin_cmap_file(name: &str) -> Option<Cow<'static, [u8]>> {
+    let path = find_bcmaps_dir()?.join(name);
+    std::fs::read(path).ok().map(Cow::Owned)
+}
+
+#[cfg(target_arch = "wasm32")]
+fn read_builtin_cmap_file(name: &str) -> Option<Cow<'static, [u8]>> {
+    let file = BUILTIN_CMAPS.get_file(name)?;
+    Some(Cow::Borrowed(file.contents()))
+}
+
 fn parse_binary_cmap(data: &[u8]) -> Result<ToUnicodeCMap, String> {
     let mut stream = BinaryCMapStream::new(data);
     let _header = stream.read_byte().ok_or("unexpected EOF in bcmap header")?;
@@ -1438,9 +1528,7 @@ fn parse_encoding_cmap_object(obj: &Object, doc: &Document) -> Option<EncodingCM
 }
 
 fn load_builtin_encoding_cmap(name: &str) -> Option<EncodingCMap> {
-    let dir = find_bcmaps_dir()?;
-    let path = dir.join(format!("{}.bcmap", name));
-    let data = std::fs::read(&path).ok()?;
+    let data = read_builtin_cmap_file(&format!("{}.bcmap", name))?;
     parse_binary_cmap_encoding(&data).ok()
 }
 
@@ -1725,9 +1813,7 @@ fn load_builtin_cmap_by_name(name: &str) -> Option<ToUnicodeCMap> {
     if !name.ends_with("UCS2") {
         return None;
     }
-    let dir = find_bcmaps_dir()?;
-    let path = dir.join(format!("{}.bcmap", name));
-    let data = std::fs::read(&path).ok()?;
+    let data = read_builtin_cmap_file(&format!("{}.bcmap", name))?;
     let mut cmap = parse_binary_cmap(&data).ok()?;
     if cmap.char_map.is_empty() && cmap.ranges.is_empty() {
         return None;
@@ -2520,6 +2606,24 @@ endcmap
         assert_eq!(cmap.lookup(0x0025), Some("B".to_string()));
     }
 
+    #[test]
+    fn test_hex_to_unicode_non_ascii_no_panic() {
+        // A destination containing a multi-byte char makes the byte length even
+        // while a byte offset can land inside a char. Slicing must not panic;
+        // it should be rejected gracefully.
+        assert_eq!(hex_to_unicode_string("XéY"), None);
+        assert_eq!(hex_to_unicode_string("\u{fffd}0"), None);
+    }
+
+    #[test]
+    fn test_parse_bfchar_non_ascii_destination_no_panic() {
+        // Crafted /ToUnicode CMap: a non-hex, non-ASCII destination previously
+        // triggered a char-boundary panic in hex_to_unicode_string.
+        let cmap_content = "beginbfchar <0041> <XéY> endbfchar";
+        // Must not panic; the malformed entry is simply skipped.
+        let _ = ToUnicodeCMap::parse(cmap_content.as_bytes());
+    }
+
     #[test]
     fn test_parse_bfchar_1byte() {
         // This is the pattern that caused the CJK bug: codespace is <0000><FFFF>
@@ -2607,6 +2711,97 @@ endbfrange
         assert_eq!(cmap.lookup(0x0005), Some("C".to_string()));
     }
 
+    #[test]
+    fn test_parse_bfchar_surrogate_pair_emoji() {
+        let cmap_content = r#"
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+2 beginbfchar
+<16> <D83CDF1F>
+<9D> <D83CDFAD>
+endbfchar
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        assert_eq!(cmap.code_byte_length, 1);
+        assert_eq!(cmap.lookup(0x16), Some("🌟".to_string()));
+        assert_eq!(cmap.lookup(0x9D), Some("🎭".to_string()));
+    }
+
+    #[test]
+    fn test_parse_bfrange_surrogate_pair_base() {
+        let cmap_content = r#"
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+1 beginbfrange
+<C8> <C9> <D83CDFD8>
+endbfrange
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        assert_eq!(cmap.code_byte_length, 1);
+        assert_eq!(cmap.lookup(0xC8), Some("🏘".to_string()));
+        assert_eq!(cmap.lookup(0xC9), Some("🏙".to_string()));
+    }
+
+    #[test]
+    fn test_parse_bfrange_preserves_single_hyphen_like_base() {
+        let cmap_content = r#"
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+1 beginbfrange
+<21> <22> <2013>
+endbfrange
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        assert_eq!(cmap.lookup(0x21), Some("–".to_string()));
+        assert_eq!(cmap.lookup(0x22), Some("—".to_string()));
+    }
+
+    #[test]
+    fn test_parse_spaced_destination_hex_without_control_noise() {
+        let cmap_content = r#"
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+3 beginbfchar
+<21> < 0009 000d 0020 00a0 >
+<22> < 002d 00ad 2010 >
+<23> <00a0>
+endbfchar
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        assert_eq!(cmap.lookup(0x21), Some("\t".to_string()));
+        assert_eq!(cmap.lookup(0x22), Some("-".to_string()));
+        assert_eq!(cmap.lookup(0x23), Some("\u{00a0}".to_string()));
+    }
+
+    #[test]
+    fn test_parse_preserves_valid_multi_character_destinations() {
+        let cmap_content = r#"
+1 begincodespacerange
+<00> <FF>
+endcodespacerange
+4 beginbfchar
+<21> <002d002d>
+<22> <20132013>
+<23> <002000a0>
+<24> <00660069>
+endbfchar
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        assert_eq!(cmap.lookup(0x21), Some("--".to_string()));
+        assert_eq!(cmap.lookup(0x22), Some("––".to_string()));
+        assert_eq!(cmap.lookup(0x23), Some(" \u{00a0}".to_string()));
+        assert_eq!(cmap.lookup(0x24), Some("fi".to_string()));
+    }
+
     #[test]
     fn test_remap_to_sequential() {
         // Simulate a broken CMap where GIDs are from pre-subsetting:
@@ -3001,4 +3196,106 @@ endbfrange
             "Remap must fire when CMap's CIDs are outside W array coverage"
         );
     }
+
+    #[test]
+    fn test_try_remap_skipped_for_cid_font_type0() {
+        // Same W/CMap mismatch as the CIDFontType2 case above, but the descendant is
+        // CIDFontType0 (CFF). There CIDs are resolved through the CFF charset, so the
+        // ToUnicode CIDs stay valid after subsetting and must not be renumbered.
+        // Real-world case: Japanese Adobe-Japan1 PDFs (e.g. National Diet Library
+        // minutes) where remapping turned correct text into unrelated glyphs.
+        let cmap_content = r#"
+1 begincodespacerange
+<0000><FFFF>
+endcodespacerange
+1 beginbfrange
+<0200> <0220> <0410>
+endbfrange
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        let mut doc = Document::new();
+
+        // CIDToGIDMap is CIDFontType2-only, but a malformed producer can still emit
+        // one on a CFF font. Use a real stream (not /Identity, which is treated as
+        // "no map") so this also fails if the guard is moved back below the
+        // CIDToGIDMap branch: cid 1 -> gid 0x0200, which the CMap resolves.
+        let mut cid_to_gid = vec![0u8; 68];
+        cid_to_gid[2] = 0x02;
+        cid_to_gid[3] = 0x00;
+        let cid_to_gid_id =
+            doc.add_object(lopdf::Stream::new(lopdf::Dictionary::new(), cid_to_gid));
+
+        let mut cid_font = lopdf::Dictionary::new();
+        cid_font.set("Subtype", lopdf::Object::Name(b"CIDFontType0".to_vec()));
+        cid_font.set("CIDToGIDMap", lopdf::Object::Reference(cid_to_gid_id));
+        cid_font.set(
+            "W",
+            lopdf::Object::Array(vec![
+                lopdf::Object::Integer(1),
+                lopdf::Object::Array(vec![lopdf::Object::Integer(500); 34]),
+            ]),
+        );
+        let cid_font_id = doc.add_object(cid_font);
+
+        let mut font_dict = lopdf::Dictionary::new();
+        font_dict.set("Encoding", lopdf::Object::Name(b"Identity-H".to_vec()));
+        font_dict.set(
+            "DescendantFonts",
+            lopdf::Object::Array(vec![lopdf::Object::Reference(cid_font_id)]),
+        );
+
+        let (primary, remapped) = try_remap_subset_cmap(cmap, &font_dict, &doc, 789);
+        assert!(
+            remapped.is_none(),
+            "Remap must be skipped for CIDFontType0 (CFF) descendants, including a \
+             CIDToGIDMap a malformed producer attached to one"
+        );
+        // The original CMap must still resolve its own CIDs.
+        assert_eq!(primary.lookup(0x0200), Some("\u{0410}".to_string()));
+    }
+
+    #[test]
+    fn test_try_remap_resolves_indirect_subtype() {
+        // /Subtype may be stored as an indirect reference. A genuine CIDFontType2
+        // font must still get the repair, so the guard has to dereference it rather
+        // than treat the unresolved value as "not CIDFontType2".
+        let cmap_content = r#"
+1 begincodespacerange
+<0000><FFFF>
+endcodespacerange
+1 beginbfrange
+<0200> <0220> <0410>
+endbfrange
+"#;
+        let cmap = ToUnicodeCMap::parse(cmap_content.as_bytes()).unwrap();
+
+        let mut doc = Document::new();
+        let subtype_id = doc.add_object(lopdf::Object::Name(b"CIDFontType2".to_vec()));
+
+        let mut cid_font = lopdf::Dictionary::new();
+        cid_font.set("Subtype", lopdf::Object::Reference(subtype_id));
+        cid_font.set("CIDToGIDMap", lopdf::Object::Name(b"Identity".to_vec()));
+        cid_font.set(
+            "W",
+            lopdf::Object::Array(vec![
+                lopdf::Object::Integer(0),
+                lopdf::Object::Array(vec![lopdf::Object::Integer(500); 34]),
+            ]),
+        );
+        let cid_font_id = doc.add_object(cid_font);
+
+        let mut font_dict = lopdf::Dictionary::new();
+        font_dict.set("Encoding", lopdf::Object::Name(b"Identity-H".to_vec()));
+        font_dict.set(
+            "DescendantFonts",
+            lopdf::Object::Array(vec![lopdf::Object::Reference(cid_font_id)]),
+        );
+
+        let (_primary, remapped) = try_remap_subset_cmap(cmap, &font_dict, &doc, 790);
+        assert!(
+            remapped.is_some(),
+            "An indirect /Subtype naming CIDFontType2 must still reach the remap"
+        );
+    }
 }

src/types.rs

@@ -116,6 +116,14 @@ pub struct TextItem {
     pub is_bold: bool,
     /// Whether the font is italic
     pub is_italic: bool,
+    /// Whether the text is underlined (drawn rule/thin rect under the
+    /// baseline — PDFs have no underline font flag, so this is detected
+    /// geometrically after extraction; see `extractor::underline`).
+    pub is_underline: bool,
+    /// Whether the text is struck out (drawn rule/thin rect crossing the
+    /// glyphs at mid x-height). Same geometric detection as underline,
+    /// different vertical window; see `extractor::underline`.
+    pub is_strikeout: bool,
     /// Type of item (text, image, link)
     pub item_type: ItemType,
     /// Marked Content ID from the content stream's BDC/BMC operator.
@@ -137,12 +145,20 @@ pub struct TextLine {
 
 impl TextLine {
     pub fn text(&self) -> String {
-        self.text_with_formatting(false, false)
+        self.text_with_formatting(false, false, false)
     }
 
-    /// Get text with optional bold/italic markdown formatting
-    pub fn text_with_formatting(&self, format_bold: bool, format_italic: bool) -> String {
-        if !format_bold && !format_italic {
+    /// Get text with optional bold/italic/decorative markdown formatting.
+    ///
+    /// `format_decorations` enables both geometrically detected source
+    /// decorations: underline (`<u>`) and strikeout (`<s>`).
+    pub fn text_with_formatting(
+        &self,
+        format_bold: bool,
+        format_italic: bool,
+        format_decorations: bool,
+    ) -> String {
+        if !format_bold && !format_italic && !format_decorations {
             return self.text_plain();
         }
 
@@ -151,6 +167,8 @@ impl TextLine {
         let mut result = String::new();
         let mut current_bold = false;
         let mut current_italic = false;
+        let mut current_underline = false;
+        let mut current_strikeout = false;
 
         for (i, item) in self.items.iter().enumerate() {
             let text = item.text.as_str();
@@ -176,9 +194,16 @@ impl TextLine {
             // we push text_trimmed below (which strips it).
             let has_leading_space = text.starts_with(' ');
 
-            // Check for style changes
-            let item_bold = format_bold && item.is_bold;
-            let item_italic = format_italic && item.is_italic;
+            // Check for style changes. Source decorations are exclusive:
+            // `<u>`/`<s>` content stays free of `**`/`*` markers — consumers
+            // (and the eval harnesses this feeds) match tag content literally,
+            // and mixed nesting breaks that. A struck-and-underlined item is
+            // emitted as struck text because deletion is the stronger semantic
+            // distinction in redline documents.
+            let item_strikeout = format_decorations && item.is_strikeout;
+            let item_underline = format_decorations && item.is_underline && !item_strikeout;
+            let item_bold = format_bold && item.is_bold && !item_underline && !item_strikeout;
+            let item_italic = format_italic && item.is_italic && !item_underline && !item_strikeout;
 
             // Close previous styles if they change
             if current_italic && !item_italic {
@@ -189,6 +214,14 @@ impl TextLine {
                 result.push_str("**");
                 current_bold = false;
             }
+            if current_underline && !item_underline {
+                result.push_str("</u>");
+                current_underline = false;
+            }
+            if current_strikeout && !item_strikeout {
+                result.push_str("</s>");
+                current_strikeout = false;
+            }
 
             // Add space: either from spacing logic or preserved from item text
             if needs_space || (has_leading_space && !result.is_empty() && !result.ends_with(' ')) {
@@ -196,6 +229,14 @@ impl TextLine {
             }
 
             // Open new styles
+            if item_underline && !current_underline {
+                result.push_str("<u>");
+                current_underline = true;
+            }
+            if item_strikeout && !current_strikeout {
+                result.push_str("<s>");
+                current_strikeout = true;
+            }
             if item_bold && !current_bold {
                 result.push_str("**");
                 current_bold = true;
@@ -215,6 +256,12 @@ impl TextLine {
         if current_bold {
             result.push_str("**");
         }
+        if current_underline {
+            result.push_str("</u>");
+        }
+        if current_strikeout {
+            result.push_str("</s>");
+        }
 
         result
     }
@@ -280,3 +327,88 @@ impl TextLine {
             || space_already_exists)
     }
 }
+
+#[cfg(test)]
+mod formatting_tests {
+    use super::{ItemType, TextItem, TextLine};
+
+    fn item(text: &str, x: f32, width: f32, strikeout: bool) -> TextItem {
+        TextItem {
+            text: text.to_string(),
+            x,
+            y: 100.0,
+            width,
+            height: 12.0,
+            font: "F1".to_string(),
+            font_size: 12.0,
+            page: 1,
+            is_bold: false,
+            is_italic: false,
+            is_underline: false,
+            is_strikeout: strikeout,
+            item_type: ItemType::Text,
+            mcid: None,
+        }
+    }
+
+    fn line(items: Vec<TextItem>) -> TextLine {
+        TextLine {
+            items,
+            y: 100.0,
+            page: 1,
+            adaptive_threshold: 0.1,
+        }
+    }
+
+    #[test]
+    fn formatting_emits_semantic_strikeout() {
+        let line = line(vec![item("deleted", 10.0, 42.0, true)]);
+
+        assert_eq!(
+            line.text_with_formatting(true, true, true),
+            "<s>deleted</s>"
+        );
+    }
+
+    #[test]
+    fn formatting_closes_strikeout_before_live_text() {
+        let line = line(vec![
+            item("keep", 10.0, 24.0, false),
+            item("remove", 40.0, 42.0, true),
+            item("keep", 88.0, 24.0, false),
+        ]);
+
+        assert_eq!(
+            line.text_with_formatting(true, true, true),
+            "keep <s>remove</s> keep"
+        );
+    }
+
+    #[test]
+    fn formatting_coalesces_adjacent_struck_items() {
+        let line = line(vec![
+            item("deleted", 10.0, 42.0, true),
+            item("words", 58.0, 30.0, true),
+        ]);
+
+        assert_eq!(
+            line.text_with_formatting(true, true, true),
+            "<s>deleted words</s>"
+        );
+    }
+
+    #[test]
+    fn strikeout_takes_precedence_over_other_styles() {
+        let mut decorated = item("deleted", 10.0, 42.0, true);
+        decorated.is_bold = true;
+        decorated.is_italic = true;
+        decorated.is_underline = true;
+        let line = line(vec![decorated]);
+
+        assert_eq!(
+            line.text_with_formatting(true, true, true),
+            "<s>deleted</s>"
+        );
+        assert_eq!(line.text(), "deleted");
+    }
+}

tests/fixtures/broken_startxref_pointer.pdf

@@ -0,0 +1,68 @@
+%PDF-1.3
+%“Œ‹ž ReportLab Generated PDF document (opensource)
+1 0 obj
+<<
+/F1 2 0 R
+>>
+endobj
+2 0 obj
+<<
+/BaseFont /Helvetica /Encoding /WinAnsiEncoding /Name /F1 /Subtype /Type1 /Type /Font
+>>
+endobj
+3 0 obj
+<<
+/Contents 7 0 R /MediaBox [ 0 0 612 792 ] /Parent 6 0 R /Resources <<
+/Font 1 0 R /ProcSet [ /PDF /Text /ImageB /ImageC /ImageI ]
+>> /Rotate 0 /Trans <<
+
+>> 
+  /Type /Page
+>>
+endobj
+4 0 obj
+<<
+/PageMode /UseNone /Pages 6 0 R /Type /Catalog
+>>
+endobj
+5 0 obj
+<<
+/Author (anonymous) /CreationDate (D:20260803112923+00'00') /Creator (anonymous) /Keywords () /ModDate (D:20260803112923+00'00') /Producer (ReportLab PDF Library - \(opensource\)) 
+  /Subject (unspecified) /Title (untitled) /Trapped /False
+>>
+endobj
+6 0 obj
+<<
+/Count 1 /Kids [ 3 0 R ] /Type /Pages
+>>
+endobj
+7 0 obj
+<<
+/Filter [ /ASCII85Decode /FlateDecode ] /Length 202
+>>
+stream
+GarW05mr9@&;9NOME,dW.,B;'jAjYq0S4Z`*D9aMA;]$5J)A/$3lESen1?F)ZJsa4$4&N%-%cs)#qW5EVhhbPiRDrAV>MC%.spto@CU"ZdipR'TtFiMR_%m*Hm$N%qL7a"ckkp9T/s[N2"Og377mP*M^akb2XQZ@'l*qT(9bVtDb5+)S&Q.#%E)<]Ao`TSk2AE'/E\fn~>endstream
+endobj
+xref
+0 8
+0000000000 65535 f 
+0000000061 00000 n 
+0000000092 00000 n 
+0000000199 00000 n 
+0000000392 00000 n 
+0000000460 00000 n 
+0000000721 00000 n 
+0000000780 00000 n 
+trailer
+<<
+/ID 
+[<6d7ea1213c5974c78613d5d2a08423b5><6d7ea1213c5974c78613d5d2a08423b5>]
+% ReportLab generated PDF document -- digest (opensource)
+
+/Info 5 0 R
+/Root 4 0 R
+/Size 8
+>>
+startxref
+9072
+%%EOF

tests/integration_tests.rs

@@ -8,12 +8,13 @@ use pdf_inspector::{
     detect_pdf_type, detect_vector_grid_in_region_mem, extract_pages_markdown,
     extract_pages_markdown_mem, extract_tables_in_regions_mem, extract_text,
     extract_text_in_regions_mem, extract_text_with_positions, extract_text_with_positions_mem,
-    process_pdf_mem, process_pdf_with_options, to_markdown, MarkdownOptions, PdfError, PdfOptions,
+    process_pdf_mem, process_pdf_mem_with_options, process_pdf_with_options, to_markdown,
+    to_markdown_from_items_with_rects_and_page_count, MarkdownOptions, PdfError, PdfOptions,
     PdfType, TextItem,
 };
 use std::collections::HashSet;
 
-fn make_minimal_text_pdf() -> Vec<u8> {
+fn make_text_pdf(content: &str, media_box: &str) -> Vec<u8> {
     let mut pdf = b"%PDF-1.4\n".to_vec();
     let mut offsets = vec![0usize];
 
@@ -40,10 +41,11 @@ fn make_minimal_text_pdf() -> Vec<u8> {
         &mut pdf,
         &mut offsets,
         3,
-        "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << /Font << /F1 5 0 R >> >> /Contents 4 0 R >>",
+        &format!(
+            "<< /Type /Page /Parent 2 0 R /MediaBox [{media_box}] /Resources << /Font << /F1 5 0 R >> >> /Contents 4 0 R >>"
+        ),
     );
 
-    let content = "BT /F1 12 Tf 100 700 Td (Hello World) Tj 0 -14 Td (Second Line) Tj 0 -14 Td (Third Line) Tj ET";
     add_object(
         &mut pdf,
         &mut offsets,
@@ -79,6 +81,186 @@ fn make_minimal_text_pdf() -> Vec<u8> {
     pdf
 }
 
+fn make_recurring_contextual_folio_pdf() -> Vec<u8> {
+    let mut pdf = b"%PDF-1.4\n".to_vec();
+    let mut offsets = vec![0usize];
+
+    fn add_object(pdf: &mut Vec<u8>, offsets: &mut Vec<usize>, id: usize, body: &str) {
+        offsets.push(pdf.len());
+        pdf.extend_from_slice(format!("{id} 0 obj\n").as_bytes());
+        pdf.extend_from_slice(body.as_bytes());
+        pdf.extend_from_slice(b"\nendobj\n");
+    }
+
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        1,
+        "<< /Type /Catalog /Pages 2 0 R >>",
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        2,
+        "<< /Type /Pages /Kids [3 0 R 5 0 R 7 0 R 9 0 R] /Count 4 >>",
+    );
+    for page_index in 0..4 {
+        let page_id = 3 + page_index * 2;
+        let content_id = page_id + 1;
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            page_id,
+            &format!(
+                "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << /Font << /F1 11 0 R >> >> /Contents {content_id} 0 R >>"
+            ),
+        );
+        let page_number = page_index + 1;
+        let content = format!(
+            "BT /F1 12 Tf 1 0 0 1 25 30 Tm ({page_number}) Tj 1 0 0 1 41 30 Tm (Company report footer) Tj 1 0 0 1 72 700 Tm (Body page {page_number}) Tj ET"
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            content_id,
+            &format!(
+                "<< /Length {} >>\nstream\n{}\nendstream",
+                content.len(),
+                content
+            ),
+        );
+    }
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        11,
+        "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>",
+    );
+
+    let xref_start = pdf.len();
+    pdf.extend_from_slice(format!("xref\n0 {}\n", offsets.len()).as_bytes());
+    pdf.extend_from_slice(b"0000000000 65535 f \n");
+    for offset in offsets.iter().skip(1) {
+        pdf.extend_from_slice(format!("{offset:010} 00000 n \n").as_bytes());
+    }
+    pdf.extend_from_slice(
+        format!(
+            "trailer\n<< /Size {} /Root 1 0 R >>\nstartxref\n{}\n%%EOF",
+            offsets.len(),
+            xref_start
+        )
+        .as_bytes(),
+    );
+
+    pdf
+}
+
+fn make_pdf_with_malformed_unselected_page() -> Vec<u8> {
+    let mut pdf = b"%PDF-1.4\n".to_vec();
+    let mut offsets = vec![0usize];
+
+    fn add_object(pdf: &mut Vec<u8>, offsets: &mut Vec<usize>, id: usize, body: &str) {
+        offsets.push(pdf.len());
+        pdf.extend_from_slice(format!("{id} 0 obj\n").as_bytes());
+        pdf.extend_from_slice(body.as_bytes());
+        pdf.extend_from_slice(b"\nendobj\n");
+    }
+
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        1,
+        "<< /Type /Catalog /Pages 2 0 R >>",
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        2,
+        "<< /Type /Pages /Kids [3 0 R 5 0 R] /Count 2 >>",
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        3,
+        "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << /Font << /F1 7 0 R >> >> /Contents 4 0 R >>",
+    );
+    let content = "BT /F1 12 Tf 1 0 0 1 25 30 Tm (1) Tj 1 0 0 1 41 30 Tm (Company report footer) Tj 1 0 0 1 72 700 Tm (Selected page text) Tj 0 -16 Td (More selected text) Tj 0 -16 Td (Still selected text) Tj ET";
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        4,
+        &format!(
+            "<< /Length {} >>\nstream\n{}\nendstream",
+            content.len(),
+            content
+        ),
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        5,
+        "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 612 792] /Resources << /Font << /F1 7 0 R >> >> /Contents 6 0 R >>",
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        6,
+        "<< /Length 3 >>\nstream\nBI \nendstream",
+    );
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        7,
+        "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>",
+    );
+
+    let xref_start = pdf.len();
+    pdf.extend_from_slice(format!("xref\n0 {}\n", offsets.len()).as_bytes());
+    pdf.extend_from_slice(b"0000000000 65535 f \n");
+    for offset in offsets.iter().skip(1) {
+        pdf.extend_from_slice(format!("{offset:010} 00000 n \n").as_bytes());
+    }
+    pdf.extend_from_slice(
+        format!(
+            "trailer\n<< /Size {} /Root 1 0 R >>\nstartxref\n{}\n%%EOF",
+            offsets.len(),
+            xref_start
+        )
+        .as_bytes(),
+    );
+
+    pdf
+}
+
+fn make_minimal_text_pdf() -> Vec<u8> {
+    make_text_pdf(
+        "BT /F1 12 Tf 100 700 Td (Hello World) Tj 0 -14 Td (Second Line) Tj 0 -14 Td (Third Line) Tj ET",
+        "0 0 612 792",
+    )
+}
+
+fn make_digit_run_repro_pdf() -> Vec<u8> {
+    let content = r#"BT
+/F1 12 Tf
+1 0 0 1 72 780 Tm (A\)) Tj
+1 0 0 1 96 780 Tm (The) Tj
+1 0 0 1 126 780 Tm (total) Tj
+1 0 0 1 166 780 Tm (of) Tj
+1 0 0 1 186 780 Tm (730) Tj
+1 0 0 1 220 780 Tm (seats) Tj
+1 0 0 1 262 780 Tm (was) Tj
+1 0 0 1 296 780 Tm (approved.) Tj
+1 0 0 1 72 755 Tm (B\)) Tj
+1 0 0 1 96 755 Tm (let) Tj
+1 0 0 1 120 755 Tm (log) Tj
+1 0 0 1 150 755 Tm (2) Tj
+1 0 0 1 164 755 Tm (=) Tj
+1 0 0 1 180 755 Tm (a) Tj
+1 0 0 1 72 720 Tm (C\) Control: The total of 730 seats was approved. let log 2 = a) Tj
+ET"#;
+    make_text_pdf(content, "0 0 595 842")
+}
+
 fn truncate_eof_marker(mut pdf: Vec<u8>) -> Vec<u8> {
     assert!(pdf.ends_with(b"%%EOF"));
     pdf.pop();
@@ -104,6 +286,8 @@ fn make_text_item(text: &str, x: f32, y: f32, font_size: f32, page: u32) -> Text
         page,
         is_bold: false,
         is_italic: false,
+        is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::Text,
         mcid: None,
     }
@@ -129,6 +313,8 @@ fn make_text_item_with_font(
         page,
         is_bold: is_bold_font(font),
         is_italic: is_italic_font(font),
+        is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::Text,
         mcid: None,
     }
@@ -326,6 +512,21 @@ fn test_group_into_lines_sorting_by_x() {
     assert_eq!(lines[0].text(), "First Second Third");
 }
 
+#[test]
+fn test_digit_only_text_runs_are_preserved_in_markdown() {
+    let pdf = make_digit_run_repro_pdf();
+
+    let items = extract_text_with_positions_mem(&pdf).expect("extract positioned text");
+    assert!(items.iter().any(|item| item.text == "730"));
+    assert!(items.iter().any(|item| item.text == "2"));
+
+    let result = process_pdf_mem(&pdf).expect("convert PDF to markdown");
+    assert_eq!(
+        result.markdown.expect("markdown output").trim(),
+        "A) The total of 730 seats was approved.\nB) let log 2 = a\nC) Control: The total of 730 seats was approved. let log 2 = a"
+    );
+}
+
 // ============================================================================
 // MarkdownOptions Tests
 // ============================================================================
@@ -333,6 +534,7 @@ fn test_group_into_lines_sorting_by_x() {
 #[test]
 fn test_markdown_options_default() {
     let opts = MarkdownOptions::default();
+    assert_eq!(opts.profile, pdf_inspector::MarkdownProfile::Fidelity);
     assert!(opts.detect_headers);
     assert!(opts.detect_lists);
     assert!(opts.detect_code);
@@ -342,6 +544,7 @@ fn test_markdown_options_default() {
 #[test]
 fn test_markdown_options_custom() {
     let opts = MarkdownOptions {
+        profile: pdf_inspector::MarkdownProfile::Compact,
         detect_headers: false,
         detect_lists: true,
         detect_code: false,
@@ -357,6 +560,7 @@ fn test_markdown_options_custom() {
         ..Default::default()
     };
     assert!(!opts.detect_headers);
+    assert_eq!(opts.profile, pdf_inspector::MarkdownProfile::Compact);
     assert!(opts.detect_lists);
     assert!(!opts.detect_code);
     assert_eq!(opts.base_font_size, Some(14.0));
@@ -540,6 +744,30 @@ fn test_markdown_from_items_page_breaks() {
     assert!(md.contains("Content on second page"));
 }
 
+#[test]
+fn test_markdown_page_count_overload_includes_trailing_blank_pages_in_folio_coverage() {
+    let mut items = Vec::new();
+    for (page, value) in [(1, "1"), (2, "2"), (3, "3"), (4, "4")] {
+        items.push(make_text_item(value, 25.0, 30.0, 12.0, page));
+        items.push(make_text_item(
+            "Company report footer",
+            41.0,
+            30.0,
+            12.0,
+            page,
+        ));
+    }
+    let options = MarkdownOptions {
+        strip_headers_footers: false,
+        ..MarkdownOptions::default()
+    };
+
+    let md = to_markdown_from_items_with_rects_and_page_count(items, options, &[], 20);
+
+    assert!(md.contains("1 Company report footer"));
+    assert!(md.contains("4 Company report footer"));
+}
+
 // ============================================================================
 // Markdown From Lines Tests
 // ============================================================================
@@ -1082,6 +1310,18 @@ fn test_snapshot_2013_app2() {
     assert_snapshot("2013-app2");
 }
 
+/// First two pages of Shannon's "A Mathematical Theory of Communication"
+/// (1998 dvips 5.58 → Distiller 3 retypesetting). Canonical legacy-TeX PDF:
+/// non-embedded base-14 fonts with no /Widths (exercises the built-in AFM
+/// metrics fallback), Type3 PK bitmap math fonts with FontMatrix
+/// [1 0 0 -1 0 0] (exercises visual-size scaling), a two-line embedded drop
+/// cap, indent-only paragraph breaks, and display math that must not be
+/// detected as tables or headings.
+#[test]
+fn test_snapshot_shannon_entropy() {
+    assert_snapshot("shannon-entropy-p1-2");
+}
+
 // ============================================================================
 // Pages Needing OCR Tests
 // ============================================================================
@@ -1098,6 +1338,7 @@ fn test_pages_needing_ocr_field_accessible() {
         title: None,
         ocr_recommended: false,
         pages_needing_ocr: Vec::new(),
+        ocr_reasons_by_page: std::collections::BTreeMap::new(),
     };
     assert!(detection_result.pages_needing_ocr.is_empty());
 
@@ -1107,6 +1348,7 @@ fn test_pages_needing_ocr_field_accessible() {
         page_count: 1,
         processing_time_ms: 0,
         pages_needing_ocr: vec![1, 3],
+        ocr_reasons_by_page: Vec::new(),
         title: None,
         confidence: 1.0,
         layout: pdf_inspector::LayoutComplexity::default(),
@@ -1187,6 +1429,77 @@ fn test_firecrawl_tagged_pdf_struct_tree() {
     assert_eq!(fence_count % 2, 0, "Code fences should be balanced");
 }
 
+#[test]
+fn test_tagged_pdf_text_items_carry_mcid() {
+    let buf = std::fs::read("tests/fixtures/firecrawl_docs_tagged.pdf").unwrap();
+    let items = pdf_inspector::extractor::extract_text_with_positions_mem(&buf).unwrap();
+    assert!(
+        items.iter().any(|i| i.mcid.is_some()),
+        "Tagged PDF text items should carry Marked Content IDs"
+    );
+}
+
+#[test]
+fn test_extract_structure_elements_tagged_pdf() {
+    let buf = std::fs::read("tests/fixtures/firecrawl_docs_tagged.pdf").unwrap();
+    let elements = pdf_inspector::extract_structure_elements_mem(&buf, None).unwrap();
+    assert!(!elements.is_empty(), "Tagged PDF should yield elements");
+    assert!(
+        elements.iter().any(|e| e.role == "H1"),
+        "Should surface H1 heading roles"
+    );
+    assert!(
+        elements.iter().all(|e| !e.role.is_empty()),
+        "Every element should carry a role name"
+    );
+
+    // Sorted by (page, mcid) for deterministic output
+    assert!(
+        elements
+            .windows(2)
+            .all(|w| (w[0].page, w[0].mcid) <= (w[1].page, w[1].mcid)),
+        "Elements should be sorted by (page, mcid)"
+    );
+
+    // The advertised join: (page, mcid) pairs must line up with the
+    // mcid-carrying TextItems from positioned extraction, and joining the
+    // H1 entries must recover non-empty heading text.
+    let items = pdf_inspector::extractor::extract_text_with_positions_mem(&buf).unwrap();
+    let h1_refs: std::collections::HashSet<(u32, i64)> = elements
+        .iter()
+        .filter(|e| e.role == "H1")
+        .map(|e| (e.page, e.mcid))
+        .collect();
+    let h1_text: String = items
+        .iter()
+        .filter(|i| i.mcid.is_some_and(|mcid| h1_refs.contains(&(i.page, mcid))))
+        .map(|i| i.text.as_str())
+        .collect();
+    assert!(
+        !h1_text.trim().is_empty(),
+        "Joining H1 structure elements to text items should recover heading text"
+    );
+
+    // Page filter is 1-indexed (matching TextItem.page) and equals the
+    // corresponding subset of the full document result.
+    let page1 = pdf_inspector::extract_structure_elements_mem(&buf, Some(&[1])).unwrap();
+    assert!(!page1.is_empty(), "Page 1 should have elements");
+    assert!(page1.iter().all(|e| e.page == 1));
+    let full_page1_count = elements.iter().filter(|e| e.page == 1).count();
+    assert_eq!(page1.len(), full_page1_count);
+}
+
+#[test]
+fn test_extract_structure_elements_untagged_pdf_empty() {
+    let buf = std::fs::read("tests/fixtures/thermo-freon12.pdf").unwrap();
+    let elements = pdf_inspector::extract_structure_elements_mem(&buf, None).unwrap();
+    assert!(
+        elements.is_empty(),
+        "Untagged PDF should yield no structure elements, got {:?}",
+        elements
+    );
+}
+
 #[test]
 fn test_identity_h_no_tounicode_suppresses_garbage() {
     // shinagawa_identity_h.pdf uses YuGothic with Identity-H encoding and no
@@ -1353,6 +1666,32 @@ fn test_extract_regions_mem_identity_h_needs_ocr() {
     );
 }
 
+/// ParseBench `text_simple__att10k.pdf` (issue #118): the producer authored a
+/// broken ToUnicode CMap that shifts every character by a per-range constant,
+/// and the embedded subset font has no `cmap` table to recover from. The
+/// resulting ciphertext is 100% printable ASCII, so it must be caught by the
+/// substitution-cipher statistics and routed to OCR instead of served silently.
+#[test]
+fn test_extract_pages_mem_shifted_cipher_tounicode_needs_ocr() {
+    let buf = std::fs::read("tests/fixtures/shifted_cipher_tounicode.pdf").unwrap();
+    let result = extract_pages_markdown_mem(&buf, None).unwrap();
+
+    assert_eq!(result.pages.len(), 1);
+    assert!(
+        result.pages[0].needs_ocr,
+        "shifted-cipher garbled page should be flagged needs_ocr"
+    );
+    assert!(
+        result.pages[0].markdown.is_empty(),
+        "garbled markdown should be suppressed"
+    );
+    assert_eq!(result.pages_needing_ocr, vec![1]);
+    assert_eq!(
+        result.pages[0].ocr_reason.as_deref(),
+        Some("suspected_garbled_text")
+    );
+}
+
 #[test]
 fn test_extract_regions_mem_multiple_regions_per_page() {
     let buf = std::fs::read("tests/fixtures/nexo-price-en.pdf").unwrap();
@@ -2880,6 +3219,57 @@ fn test_extract_pages_markdown_basic() {
     assert!(!result.pages[0].needs_ocr);
 }
 
+#[test]
+fn test_extract_pages_markdown_uses_document_wide_folio_context() {
+    let pdf = make_recurring_contextual_folio_pdf();
+    let result = extract_pages_markdown_mem(&pdf, None).unwrap();
+
+    assert_eq!(result.pages.len(), 4);
+    for (index, page) in result.pages.iter().enumerate() {
+        assert!(page.markdown.contains("Company report footer"));
+        assert!(
+            !page
+                .markdown
+                .contains(&format!("{} Company report footer", index + 1)),
+            "recurring contextual folio survived on page {}: {}",
+            index + 1,
+            page.markdown
+        );
+    }
+}
+
+#[test]
+fn test_process_pdf_page_filter_uses_document_wide_folio_context() {
+    let pdf = make_recurring_contextual_folio_pdf();
+    let result = process_pdf_mem_with_options(&pdf, PdfOptions::new().pages([1])).unwrap();
+    let markdown = result.markdown.unwrap();
+
+    assert!(markdown.contains("Company report footer"));
+    assert!(!markdown.contains("1 Company report footer"), "{markdown}");
+    assert!(markdown.contains("Body page 1"));
+    assert!(!markdown.contains("Body page 2"));
+}
+
+#[test]
+fn test_selected_page_ignores_context_only_extraction_failure() {
+    let pdf = make_pdf_with_malformed_unselected_page();
+
+    let pages = extract_pages_markdown_mem(&pdf, Some(&[0])).unwrap();
+    assert_eq!(pages.pages.len(), 1);
+    assert!(pages.pages[0].markdown.contains("Selected page text"));
+
+    let result = process_pdf_mem_with_options(&pdf, PdfOptions::new().pages([1])).unwrap();
+    let markdown = result.markdown.unwrap();
+    assert!(markdown.contains("Selected page text"));
+}
+
+#[test]
+fn test_requested_page_extraction_failure_remains_fatal() {
+    let pdf = make_pdf_with_malformed_unselected_page();
+
+    assert!(extract_pages_markdown_mem(&pdf, Some(&[1])).is_err());
+}
+
 #[test]
 fn test_extract_pages_markdown_page_ordering() {
     let buf = std::fs::read("tests/fixtures/nexo-price-en.pdf").unwrap();
@@ -3575,3 +3965,161 @@ fn test_markdown_options_default_has_include_images_false() {
     let opts = MarkdownOptions::default();
     assert!(!opts.include_images);
 }
+
+#[test]
+fn encrypted_pdf_decrypts_with_correct_password() {
+    let path = "tests/fixtures/encrypted-secret123.pdf";
+
+    // No password: the file is encrypted and can't be read.
+    let no_pw = process_pdf_with_options(path, PdfOptions::new());
+    assert!(
+        matches!(no_pw, Err(PdfError::Encrypted)),
+        "expected Encrypted without a password, got {no_pw:?}"
+    );
+
+    // Wrong password: still rejected.
+    let wrong = process_pdf_with_options(path, PdfOptions::new().password("wrong"));
+    assert!(
+        matches!(wrong, Err(PdfError::Encrypted)),
+        "expected Encrypted with a wrong password, got {wrong:?}"
+    );
+
+    // Correct password: decrypts and extracts real content.
+    let ok = process_pdf_with_options(path, PdfOptions::new().password("secret123"))
+        .expect("correct password should decrypt");
+    let md = ok.markdown.unwrap_or_default();
+    // Assert a stable fixture token so a garbled-but-long extraction (the
+    // encrypted-stream regression this guards) still fails the test.
+    assert!(
+        md.contains("Procurement"),
+        "decrypted markdown should contain the fixture's real text, got {} chars",
+        md.len()
+    );
+}
+
+/// Regression for the #231 review finding: `extract_pages_markdown`'s
+/// `has_template_image` check must be gated the same way
+/// `classify_pdf`/`detect_pdf_type` gates it (image_count <= 1, few text
+/// ops, low alphanumeric diversity) — not treated as sufficient on its
+/// own. The fixture is a real text page with substantial, richly varied
+/// body text (>=50 Tj ops) drawn over a full-bleed background image
+/// (e.g. letterhead/watermark). Before the fix, has_template_image alone
+/// forced needs_ocr=true and discarded the page's clean markdown; now the
+/// page must extract normally.
+#[test]
+fn test_extract_pages_markdown_does_not_ocr_text_page_with_watermark_image() {
+    let buf = std::fs::read("tests/fixtures/text_page_with_watermark_image.pdf").unwrap();
+
+    let ext = extract_pages_markdown_mem(&buf, None).expect("fixture should extract");
+    let page = &ext.pages[0];
+    assert!(
+        !page.needs_ocr,
+        "a text page with substantial real text should not be routed to OCR \
+         just because it has a background image"
+    );
+    assert!(
+        page.markdown.contains("watermark"),
+        "expected the page's real body text to be preserved, got: {:?}",
+        page.markdown
+    );
+}
+
+/// Regression for the #231 review finding: `extract_pages_markdown` never
+/// checked `has_vector_text` at all, even though `detect_from_document`'s
+/// Mixed-type per-page routing always sends vector-outlined-text pages to
+/// OCR (outlined glyphs can't be extracted as text). A page with massive
+/// path ops (outlined decorative text) plus a short genuine caption would
+/// extract that caption cleanly — non-empty, non-garbled — so the
+/// existing empty/garbage-text checks alone couldn't catch it.
+#[test]
+fn test_extract_pages_markdown_ocrs_page_with_vector_outlined_text() {
+    let buf = std::fs::read("tests/fixtures/vector_outlined_text_with_caption.pdf").unwrap();
+
+    let cls = pdf_inspector::detector::detect_pdf_type_mem(&buf).expect("fixture should classify");
+    assert!(
+        cls.pages_needing_ocr.contains(&1),
+        "classify_pdf should flag page 1 as needing OCR (vector-outlined text), got: {:?}",
+        cls.pages_needing_ocr
+    );
+
+    let ext = extract_pages_markdown_mem(&buf, None).expect("fixture should extract");
+    let page = &ext.pages[0];
+    assert!(
+        page.needs_ocr,
+        "extract_pages_markdown must agree with classify_pdf that this page needs OCR"
+    );
+    assert!(
+        page.markdown.is_empty(),
+        "a page flagged needs_ocr must not return markdown as if extraction were \
+         trustworthy, got: {:?}",
+        page.markdown
+    );
+}
+
+#[test]
+fn pdf_options_debug_redacts_password() {
+    let opts = PdfOptions::new().password("secret123");
+    let dbg = format!("{opts:?}");
+    assert!(
+        !dbg.contains("secret123"),
+        "password leaked in Debug: {dbg}"
+    );
+    assert!(dbg.contains("REDACTED"), "expected redaction marker: {dbg}");
+}
+
+/// Regression for #228: a `startxref` pointer corrupted to point at the
+/// wrong byte offset (a single flipped digit — a real, common writer bug)
+/// must not make the whole file unprocessable. The real classic xref table
+/// is still present and findable by scanning for the `xref` keyword; both
+/// pypdf and pdfium recover the same way. Before this fix, every entry
+/// point raised "Invalid PDF structure" on a file whose object data was
+/// otherwise completely intact.
+#[test]
+fn test_process_pdf_recovers_corrupted_startxref_pointer() {
+    let result = process_pdf_with_options(
+        "tests/fixtures/broken_startxref_pointer.pdf",
+        PdfOptions::new(),
+    )
+    .expect("a corrupted startxref pointer should be recoverable, like pypdf/pdfium");
+
+    assert_eq!(result.page_count, 1);
+    let md = result.markdown.unwrap_or_default();
+    assert!(
+        md.contains("Order Detail Report by Account") && md.contains("WIDGET ASSEMBLY"),
+        "recovered document should extract its real text, got: {md:?}"
+    );
+}
+
+/// Regression for #227: `extract_pages_markdown`'s per-page `needs_ocr`
+/// must agree with `classify_pdf`/`detect_pdf_type` on the same page. The
+/// fixture is a full-page raster "scan" with a single line of genuine
+/// native text drawn over it (a header) — the native text extracts
+/// perfectly cleanly (no decoding issues, non-empty), so a needs_ocr
+/// computation based on text-quality signals alone says `false`, while
+/// detection correctly sees a dominant background image and says the page
+/// needs OCR. Both must now agree it needs OCR, and the markdown must not
+/// be returned as if the extraction were trustworthy.
+#[test]
+fn test_extract_pages_markdown_agrees_with_classify_on_scan_with_native_header() {
+    let buf = std::fs::read("tests/fixtures/scan_with_native_header_text.pdf").unwrap();
+
+    let cls = pdf_inspector::detector::detect_pdf_type_mem(&buf).expect("fixture should classify");
+    assert!(
+        cls.pages_needing_ocr.contains(&1),
+        "classify_pdf should flag page 1 as needing OCR (image-dominated), got: {:?}",
+        cls.pages_needing_ocr
+    );
+
+    let ext = extract_pages_markdown_mem(&buf, None).expect("fixture should extract");
+    let page = &ext.pages[0];
+    assert!(
+        page.needs_ocr,
+        "extract_pages_markdown must agree with classify_pdf that this page needs OCR"
+    );
+    assert!(
+        page.markdown.is_empty(),
+        "a page flagged needs_ocr must not return markdown as if extraction were \
+         trustworthy, got: {:?}",
+        page.markdown
+    );
+}

tests/snapshots/2013-app2.md

@@ -188,7 +188,7 @@
 |156|23/7|General renovation works to Block B at Belonie Secondary School|MOE|Belvedere Builders|SR869,505.75|
 |157|30/7|Procurement of Engine Block and Crankshaft for Engine A11|PUC|Ras Tek Pvt Ltd|Euro798,650.00|
 |158|30/7|procurement of Wartsila Engine spares|PUC|Wartsila Eastern Africa ltd|Euro158,424.00|
-|159|30/7|Proposed walkway, Drain, rock armoring , road and Bridge widening at Anse Talbot( Ex-Golden Egg)|SLTA|G&S Enterpise|SR1,113,010.00|
+|159|30/7|Proposed walkway, Drain, rock armoring, road and Bridge widening at Anse Talbot( Ex-Golden Egg)|SLTA|G&S Enterpise|SR1,113,010.00|
 |160|30/7|Procurement of transfer pump control panel|PUC|CA Engineering Consultancy Pte Ltd|SGD14,600.00|
 |161|30/7|Consultancy service for North to South Victoria Bye- Pass road and utilities organisation|MLUH|Sonnel Seychelles LTD|SR1,332,000.00|
 |162 AUG|30/7|Procurement of the supply of sodium cardonate|PUC|HPL Chemical LTD|USD42,600.00|
@@ -237,7 +237,7 @@
 |201|24/9|Procurement of vehicle x 2|SLTA|Abhaye Valabhji Pty Ltd|SR1000.000.00|
 ||OCT|||||
 |202|1/10|Proposed new traffic lane to 5th June Avenue|SLTA|Divy Constrution|SR2,864,589.00|
-|203|1/10||Proposed Walkway, Drain, rock armoring , road and Bridge widening at Anse Talbot( Ex-Golden Egg) - Variations SLTA|G & S Enterprise|SR200,448.00|
+|203|1/10||Proposed Walkway, Drain, rock armoring, road and Bridge widening at Anse Talbot(Ex-Golden Egg) - Variations SLTA|G & S Enterprise|SR200,448.00|
 |204|1/10|Proposed Reconstrcution of Burnt House-Au Cap|MLUH|Furui Construction|SR946,130.00|
 |205|1/10|Variation on the project associated with the procurement of seven 100m3/day containerised plant|PUC|Tornado Group (UAE)|USD172,500.00|
 |206|1/10|Works on the breaker system at Bel Omber desalination plant|PUC|United Concrete Products (Sey)Ltd|SR1,998,993.11|

tests/snapshots/p1244-1996.md

@@ -10,7 +10,7 @@ Department of the Treasury **Internal Revenue Service**
 
 ### This publication contains:
 
-**Form 4070A, Employee’s Daily Record of** Tips **Form 4070, Employee’s Report of Tips to** Employer
+**Form 4070A,** Employee’s Daily Record of Tips **Form 4070,** Employee’s Report of Tips to Employer
 
 For the period
 
@@ -22,7 +22,9 @@ Name and address of employee
 
 **Publication 1244 (Rev. 7-96)** Cat. No. 44472W
 
-**Instructions** You must keep sufficient proof to show the amount of your tip income for the year. A daily record of your tip income is considered sufficient proof. Keep a daily record for each workday showing the amount of cash and credit card tips received directly from customers or other employees. Also keep a record of the amount of tips, if any, you paid to other employees through tip sharing, tip pooling or other arrangements, and the names of employees to whom you paid tips. Show the date that each entry is made. This date should be on or near the date you received the tip income. You may use Form 4070A, Employee’s Daily Record of Tips, or any other daily record to record your tips. **Reporting Tips to Your Employer.—If you** receive tips that total $20 or more for any month while working for one employer, you must report the tips to your employer. Tips include cash left by customers, tips customers add to credit card charges, and tips you receive from other employees. You must report your tips for any one month by the 10th day of the next month. If the 10th day falls on a Saturday, Sunday, or legal holiday, you may give the report to your employer on the next business day that is not a Saturday, Sunday, or legal holiday. You must report tips that total $20 or more every month regardless of your total wages and tips for the year. You may use Form 4070, Employee’s Report of Tips to Employer, to report your tips to your employer. See the instructions on the back of Form 4070. You must include all tips, including tips not reported to your employer, as wages on your income tax return. You may use the last page of this publication to total your tips for the year. Your employer must withhold income, social security, and Medicare (or railroad retirement) taxes on tips you report. Your employer usually deducts the withholding due on tips from your regular wages.
+### Instructions
+
+You must keep sufficient proof to show the amount of your tip income for the year. A daily record of your tip income is considered sufficient proof. Keep a daily record for each workday showing the amount of cash and credit card tips received directly from customers or other employees. Also keep a record of the amount of tips, if any, you paid to other employees through tip sharing, tip pooling or other arrangements, and the names of employees to whom you paid tips. Show the date that each entry is made. This date should be on or near the date you received the tip income. You may use **Form 4070A**, Employee’s Daily Record of Tips, or any other daily record to record your tips. **Reporting Tips to Your Employer.—**If you receive tips that total $20 or more for any month while working for one employer, you must report the tips to your employer. Tips include cash left by customers, tips customers add to credit card charges, and tips you receive from other employees. You must report your tips for any one month by the 10th day of the next month. If the 10th day falls on a Saturday, Sunday, or legal holiday, you may give the report to your employer on the next business day that is not a Saturday, Sunday, or legal holiday. You must report tips that total $20 or more every month regardless of your total wages and tips for the year. You may use **Form 4070**, Employee’s Report of Tips to Employer, to report your tips to your employer. See the instructions on the back of Form 4070. You must include all tips, including tips not reported to your employer, as wages on your income tax return. You may use the last page of this publication to total your tips for the year. Your employer must withhold income, social security, and Medicare (or railroad retirement) taxes on tips you report. Your employer usually deducts the withholding due on tips from your regular wages.
 
 *(continued on inside of back cover)*
 
@@ -30,14 +32,14 @@ Form **4070A** Employee’s Daily Record of Tips (Rev. July 1996) **This is a vo
 
 Establishment name (if different)
 
-Date Date **a. Tips received**
+Date Date **a.** Tips received
 
-**b. Credit card tips c. Tips paid out to d. Names of employees to whom you**
+**b.** Credit card tips **c.** Tips paid out to **d.** Names of employees to whom you
 tips of directly from customers received other employees paid tips rec’d. entry and other employees 1 2 3 4 5 **Subtotals** **For Paperwork Reduction Act Notice, see Instructions on the back of Form 4070. Page 1**
 
-Date Date **a. Tips received**
+Date Date **a.** Tips received
 
-**b. Credit card tips c. Tips paid out to d. Names of employees to whom you**
+**b.** Credit card tips **c.** Tips paid out to **d.** Names of employees to whom you
 tips of directly from customers received other employees paid tips rec’d. entry and other employees
 
 7 8 9 10 11 12 13 14 15 **Subtotals**
@@ -48,13 +50,13 @@ tips of directly from customers received other employees paid tips rec’d. entr
 
 **Page 3**
 
-27 28 29 30 31 **Subtotals** **from pages** **1, 2, and 3** **Totals**
+27 28 29 30 31 **Subtotals from pages** **1, 2, and 3** **Totals**
 
-**1.** Report total cash tips (col. a) on Form 4070, line 1.
-**2.** Report total credit card tips (col. b) on Form 4070, line 2.
-**3.** Report total tips paid out (col. c) on Form 4070, line 3. **Page 4**
+**1.** Report total cash tips (col. **a**) on Form 4070, line **1.**
+**2.** Report total credit card tips (col. **b**) on Form 4070, line **2.**
+**3.** Report total tips paid out (col. **c**) on Form 4070, line **3.** **Page 4**
 
-Form Employee’s Report (Rev. July 1996)
+Form **4070** Employee’s Report (Rev. July 1996)
 
 ## of Tips to EmployerOMB No. 1545-0065
 
@@ -66,17 +68,16 @@ Employer’s name and address (include establishment name, if different) **1** C
 
 **3** Tips paid out
 
-Month or shorter period in which tips were received **4** Net tips (lines 1 + 2 - 3) from, 19, to, 19 Signature Date
+Month or shorter period in which tips were received **4** Net tips (lines **1 + 2 - 3**) from, 19, to, 19 Signature Date
 
-**Paperwork Reduction Act Notice.—We ask for the** information on these forms to carry out the Internal Revenue laws of the United States. You are required to give us the information. We need it to ensure that you are complying with these laws and to allow us to figure and collect the right amount of tax. You are not required to provide the information requested on a form that is subject to the Paperwork Reduction Act unless the form displays a valid OMB control number. Books or records relating to a form or its instructions must be retained as long as their contents may become material in the administration of any Internal Revenue law. Generally, tax returns and return information are confidential, as required by Code section 6103. The time needed to complete Forms 4070 and 4070A will vary depending on individual circumstances. The estimated average times are: Recordkeeping—Form 4070, 7 min.; Form 4070A, 3 hr. and 23 min.; Learning **about the law—each form, 2 min.; Preparing Form 4070,** 13 min.; Form 4070A, 55 min.; and Copying and **providing Form 4070, 10 min.; Form 4070A, 14 min.** If you have comments concerning the accuracy of these time estimates or suggestions for making these
+**Paperwork Reduction Act Notice.—**We ask for the information on these forms to carry out the Internal Revenue laws of the United States. You are required to give us the information. We need it to ensure that you are complying with these laws and to allow us to figure and collect the right amount of tax. You are not required to provide the information requested on a form that is subject to the Paperwork Reduction Act unless the form displays a valid OMB control number. Books or records relating to a form or its instructions must be retained as long as their contents may become material in the administration of any Internal Revenue law. Generally, tax returns and return information are confidential, as required by Code section 6103. The time needed to complete Forms 4070 and 4070A will vary depending on individual circumstances. The estimated average times are: **Recordkeeping**—Form 4070, 7 min.; Form 4070A, 3 hr. and 23 min.; **Learning** **about the law**—each form, 2 min.; **Preparing** Form 4070, 13 min.; Form 4070A, 55 min.; and **Copying and** **providing** Form 4070, 10 min.; Form 4070A, 14 min. If you have comments concerning the accuracy of these time estimates or suggestions for making these
 
-forms simpler, we would be happy to hear from you. You can write to the Tax Forms Committee, Western Area Distribution Center, Rancho Cordova, CA 95743-0001. **Purpose.—Use this form to report tips you receive to** your employer. This includes cash tips, tips you receive from other employees, and credit card tips. You must report tips every month regardless of your total wages and tips for the year. However, you do not have to report tips to your employer for any month you received less than $20 in tips while working for that employer. Report tips by the 10th day of the month following the month that you receive them. If the 10th day is a Saturday, Sunday, or legal holiday, report tips by the next day that is not a Saturday, Sunday, or legal holiday. See Pub. 531, Reporting Tip Income, for more information. You can get additional copies of Pub. 1244, Employee’s Daily Record of Tips and Report to Employer, which contains both Forms 4070A and 4070, by calling 1-800-TAX-FORM (1-800-829-3676).
+forms simpler, we would be happy to hear from you. You can write to the Tax Forms Committee, Western Area Distribution Center, Rancho Cordova, CA 95743-0001. **Purpose.—**Use this form to report tips you receive to your employer. This includes cash tips, tips you receive from other employees, and credit card tips. You must report tips every month regardless of your total wages and tips for the year. However, you do not have to report tips to your employer for any month you received less than $20 in tips while working for that employer. Report tips by the 10th day of the month following the month that you receive them. If the 10th day is a Saturday, Sunday, or legal holiday, report tips by the next day that is not a Saturday, Sunday, or legal holiday. See **Pub. 531**, Reporting Tip Income, for more information. You can get additional copies of **Pub. 1244**, Employee’s Daily Record of Tips and Report to Employer, which contains both Forms 4070A and 4070, by calling 1-800-TAX-FORM (1-800-829-3676).
 
-**Instructions (continued)**
+<u>Instructions (continued)</u>
 
-**Unreported Tips.—If you received tips of $20 or** more for any month while working for one employer but did not report them to your employer, you must figure and pay social security and Medicare taxes on the unreported tips when you file your tax return. If you have unreported tips, you must use Form 1040 and Form 4137, Social Security and Medicare Tax on Unreported Tip Income, to report them. You may not use Form 1040A or 1040EZ. Employees subject to the Railroad Retirement Tax Act cannot use Form 4137 to pay railroad retirement tax on unreported tips. To get railroad retirement credit, you must report tips to your employer. If you do not report tips to your employer as required, you may be charged a penalty of 50% of the social security and Medicare taxes (or railroad retirement tax) due on the unreported tips unless there was reasonable cause for not reporting them. **Additional Information.—Get Pub. 531, Reporting** Tip Income, and Form 4137 for more information on tips. If you are an employee of certain large food or beverage establishments, see Pub. 531 for tip allocation rules. **Recordkeeping.—If you do not keep a daily** record of tips, you must keep other reliable proof of the tip income you received. This proof includes copies of restaurant bills and credit card charges that show amounts customers added as tips. Keep your tip income records for as long as the information on them may be needed in the administration of any Internal Revenue law.
+**Unreported Tips.—**If you received tips of $20 or more for any month while working for one employer but did not report them to your employer, you must figure and pay social security and Medicare taxes on the unreported tips when you file your tax return. If you have unreported tips, you **must** use Form 1040 and **Form 4137,** Social Security and Medicare Tax on Unreported Tip Income, to report them. You may **not** use Form 1040A or 1040EZ. Employees subject to the Railroad Retirement Tax Act **cannot** use Form 4137 to pay railroad retirement tax on unreported tips. To get railroad retirement credit, you must report tips to your employer. If you do not report tips to your employer as required, you may be charged a penalty of 50% of the social security and Medicare taxes (or railroad retirement tax) due on the unreported tips unless there was reasonable cause for not reporting them. **Additional Information.—**Get **Pub. 531,** Reporting Tip Income, and Form 4137 for more information on tips. If you are an employee of certain large food or beverage establishments, see Pub. 531 for tip allocation rules. **Recordkeeping.—**If you do not keep a daily record of tips, you must keep other reliable proof of the tip income you received. This proof includes copies of restaurant bills and credit card charges that show amounts customers added as tips. Keep your tip income records for as long as the information on them may be needed in the administration of any Internal Revenue law.
 
 ### Instructions (continued)
 
 Use this space to total your tips for the year
-

tests/snapshots/real-estate-pricing.md

@@ -6,7 +6,7 @@
 
 8 4 Z E L L / L U R I E R E A L E S T A T E C E N T E R
 
-**Table I: Cap rate correlations** **Cap Rate Correlation With:*** **BBB Corp** **10-Year Bond Yield S&P Dividend** **Treasury (10-15 yr) Yield** Multifamily 0.187 0.771 0.068 Industrial-0.221 0.748-0.307 CBD Office-0.449 0.694-0.458 Retail-0.181 0.649-02.58
+**Table I:** Cap rate correlations **Cap Rate Correlation With:*** **BBB Corp** **10-Year Bond Yield S&P Dividend** **Treasury (10-15 yr) Yield** Multifamily 0.187 0.771 0.068 Industrial-0.221 0.748-0.307 CBD Office-0.449 0.694-0.458 Retail-0.181 0.649-02.58
 
 * Based on 25 years of data for the 10-yrT & S&P DivYld; and 14 years for BBB.
 **Figure 1:** NCREIF cap rates vs. 10-yearTreasury
@@ -20,7 +20,9 @@ R E V I E W 8 5
 
 **Figure 2:** Capratespreadsover10-yearTreasury
 
-**Basis Points -200** -400
+**Basis Points** -200
+
+-400
 
 -600
 
@@ -32,7 +34,7 @@ R E V I E W 8 5
 
 1982 1986 1990 1994 1998 2002 2006
 
-**Table II: Correlationsofspreadsbypropertytype** **Correlation of Cap Rate Spreads Over Treasury** **Multifamily Industrial CBD Office**
+**Table II:** Correlationsofspreadsbypropertytype **Correlation of Cap Rate Spreads Over Treasury** **Multifamily Industrial CBD Office**
 
 ||Multifamily|Industrial|CBD Office|
 |---|---|---|---|

tests/snapshots/shannon-entropy-p1-2.md

@@ -0,0 +1,43 @@
+Reprinted with corrections from *The Bell System Technical Journal,* Vol. 27, pp. 379–423, 623–656, July, October, 1948.
+
+## A Mathematical Theory of Communication
+
+### By C. E. SHANNON
+
+INTRODUCTION
+
+HE recent development of various methods of modulation such as PCM and PPM which exchange
+
+# Tbandwidth for signal-to-noise ratio has intensified the interest in a general theory of communication. A
+
+basis for such a theory is contained in the important papers of Nyquist¹ and Hartley² on this subject. In the present paper we will extend the theory to include a number of new factors, in particular the effect of noise in the channel, and the savings possible due to the statistical structure of the original message and due to the nature of the final destination of the information. The fundamental problem of communication is that of reproducing at one point either exactly or ap- proximately a message selected at another point. Frequently the messages have *meaning*; that is they refer to or are correlated according to some system with certain physical or conceptual entities. These semantic aspects of communication are irrelevant to the engineering problem. The significant aspect is that the actual message is one *selected from a set* of possible messages. The system must be designed to operate for each possible selection, not just the one which will actually be chosen since this is unknown at the time of design. If the number of messages in the set is finite then this number or any monotonic function of this number can be regarded as a measure of the information produced when one message is chosen from the set, all choices being equally likely. As was pointed out by Hartley the most natural choice is the logarithmic function. Although this definition must be generalized considerably when we consider the influence of the statistics of the message and when we have a continuous range of messages, we will in all cases use an essentially logarithmic measure. The logarithmic measure is more convenient for various reasons:
+
+1. It is practically more useful. Parameters of engineering importance such as time, bandwidth, number of relays, etc., tend to vary linearly with the logarithm of the number of possibilities. For example, adding one relay to a group doubles the number of possible states of the relays. It adds 1 to the base 2 logarithm of this number. Doubling the time roughly squares the number of possible messages, or doubles the logarithm, etc.
+2. It is nearer to our intuitive feeling as to the proper measure. This is closely related to (1) since we in- tuitively measures entities by linear comparison with common standards. One feels, for example, that two punched cards should have twice the capacity of one for information storage, and two identical channels twice the capacity of one for transmitting information.
+3. It is mathematically more suitable. Many of the limiting operations are simple in terms of the loga- rithm but would require clumsy restatement in terms of the number of possibilities. The choice of a logarithmic base corresponds to the choice of a unit for measuring information. If the
+base 2 is used the resulting units may be called binary digits, or more briefly *bits,* a word suggested by
+
+J. W. Tukey. A device with two stable positions, such as a relay or a flip-flop circuit, can store one bit of information. *N* such devices can store*N* bits, since the total number of possible states is 2
+*N* and log₂2 *N* = *N*. If the base 10 is used the units may be called decimal digits. Since
+
+log₂*M* = log₁₀*M*= log₁₀2 = 3:32 log₁₀*M*;
+
+1 Nyquist, H., “Certain Factors Affecting Telegraph Speed,” *Bell System Technical Journal,* April 1924, p. 324; “Certain Topics in Telegraph Transmission Theory,” *A.I.E.E. Trans.,* v. 47, April 1928, p. 617. 2 Hartley, R. V. L., “Transmission of Information,” *Bell System Technical Journal,* July 1928, p. 535.
+
+INFORMATION SOURCE TRANSMITTER RECEIVER DESTINATION
+
+SIGNAL RECEIVED SIGNAL MESSAGE MESSAGE
+
+NOISE SOURCE
+
+Fig. 1 — Schematic diagram of a general communication system.
+
+a decimal digit is about 3 13 bits. A digit wheel on a desk computing machine has ten stable positions and therefore has a storage capacity of one decimal digit. In analytical work where integration and differentiation are involved the base *e* is sometimes useful. The resulting units of information will be called natural units. Change from the base *a* to base *b* merely requires multiplication by log*ba*. By a communication system we will mean a system of the type indicated schematically in Fig. 1. It consists of essentially five parts:
+
+1. An *information source* which produces a message or sequence of messages to be communicated to the receiving terminal. The message may be of various types: (a) A sequence of letters as in a telegraph of teletype system; (b) A single function of time *f* (*t*) as in radio or telephony; (c) A function of time and other variables as in black and white television — here the message may be thought of as a function *f* (*x*; *y*;*t*) of two space coordinates and time, the light intensity at point (*x*; *y*) and time *t* on a pickup tube plate; (d) Two or more functions of time, say *f* (*t*), *g*(*t*), *h*(*t*) — this is the case in “three- dimensional” sound transmission or if the system is intended to service several individual channels in multiplex; (e) Several functions of several variables — in color television the message consists of three functions *f* (*x*; *y*;*t*), *g*(*x*; *y*;*t*), *h*(*x*; *y*;*t*) defined in a three-dimensional continuum — we may also think of these three functions as components of a vector field defined in the region — similarly, several black and white television sources would produce “messages” consisting of a number of functions of three variables; (f) Various combinations also occur, for example in television with an associated audio channel.
+2. A *transmitter* which operates on the message in some way to produce a signal suitable for trans- mission over the channel. In telephony this operation consists merely of changing sound pressure into a proportional electrical current. In telegraphy we have an encoding operation which produces a sequence of dots, dashes and spaces on the channel corresponding to the message. In a multiplex PCM system the different speech functions must be sampled, compressed, quantized and encoded, and finally interleaved properly to construct the signal. Vocoder systems, television and frequency modulation are other examples of complex operations applied to the message to obtain the signal.
+3. The *channel* is merely the medium used to transmit the signal from transmitter to receiver. It may be a pair of wires, a coaxial cable, a band of radio frequencies, a beam of light, etc.
+4. The *receiver* ordinarily performs the inverse operation of that done by the transmitter, reconstructing the message from the signal.
+5. The *destination* is the person (or thing) for whom the message is intended. We wish to consider certain general problems involving communication systems. To do this it is first
+necessary to represent the various elements involved as mathematical entities, suitably idealized from their
+

tests/snapshots/td9264.md

@@ -1,17 +1,8 @@
-(e) [Reserved]. For further guidance, see §1.1563-3T(e)(1). Par. 50. Section 1.1563-3T is added to read as follows:
-§1.1563-3T Rules for determining stock ownership (temporary).
-
-(a) through (d)(2)(iii) [Reserved]. For further guidance, see §1.1563-3(a)
-through (d)(2)(iii). (iv) Statement. If the application of paragraph (d)(2)(ii) or (iii) of §1.1563-3 does not result in a corporation being treated as a component member of only one controlled group of corporations on a December 31, then such corporation will be treated as a component member of only one such group on such date. Such corporation may elect the group in which it is to be included by including on or with its income tax return a statement entitled, “STATEMENT TO ELECT CONTROLLED GROUP PURSUANT TO §1.1563-3T(d)(2)(iv).” The statement must include--
-
-(A) A description of each of the controlled groups in which the corporation
-could be included. The description must include the name and employer identification number of each component member of each such group and the stock ownership of the component members of each such group; and
-
-(B) The following representation: [INSERT NAME AND EMPLOYER
-IDENTIFICATION NUMBER OF CORPORATION] ELECTS TO BE TREATED AS A COMPONENT MEMBER OF THE [INSERT DESIGNATION OF GROUP].
-
-(v) Election-- (A) Election filed. An election filed under paragraph (d)(2)(iv) of
-this section is irrevocable and effective until paragraph (d)(2)(ii) or (iii) of §1.1563-3 applies or until a change in the stock ownership of the corporation results in
+||||(e) [Reserved]. For further guidance, see §1.1563-3T(e)(1). Par. 50. Section 1.1563-3T is added to read as follows: §1.1563-3T Rules for determining stock ownership (temporary). (a) through (d)(2)(iii) [Reserved]. For further guidance, see §1.1563-3(a)|
+|---|---|---|---|
+||through (d)(2)(iii).|||
+||(iv)|Statement|. If the application of paragraph (d)(2)(ii) or (iii) of §1.1563-3 does not result in a corporation being treated as a component member of only one controlled group of corporations on a December 31, then such corporation will be treated as a component member of only one such group on such date. Such corporation may elect the group in which it is to be included by including on or with its income tax return a statement entitled, “STATEMENT TO ELECT CONTROLLED GROUP PURSUANT TO §1.1563-3T(d)(2)(iv).” The statement must include-- (A) A description of each of the controlled groups in which the corporation could be included. The description must include the name and employer identification number of each component member of each such group and the stock ownership of the component members of each such group; and (B) The following representation: [INSERT NAME AND EMPLOYER IDENTIFICATION NUMBER OF CORPORATION] ELECTS TO BE TREATED AS A COMPONENT MEMBER OF THE [INSERT DESIGNATION OF GROUP].|
+||(v)|Election|-- (A) Election filed. An election filed under paragraph (d)(2)(iv) of this section is irrevocable and effective until paragraph (d)(2)(ii) or (iii) of §1.1563-3 applies or until a change in the stock ownership of the corporation results in|
 
 |termination of membership in the controlled group in which such corporation has||
 |---|---|
@@ -29,48 +20,48 @@ this section is irrevocable and effective until paragraph (d)(2)(ii) or (iii) of
 Federal income tax return (including any amended return filed on or before the due date (including extensions) of such original return) timely filed on or after May 30,
 
 2006.
-(2) Expiration date. The applicability of this section will expire on May 26,
-2009. Par. 51. Section 1.6012-2 is amended by revising paragraph (c) and adding paragraph (k) to read as follows: §1.6012-2 Corporations required to make returns of income.
+(2) <u>Expiration date</u>. The applicability of this section will expire on May 26,
+2009. Par. 51. Section 1.6012-2 is amended by revising paragraph (c) and adding paragraph (k) to read as follows: <u>§1.6012-2 Corporations required to make returns of income</u>.
 * * * * *
 (c) [Reserved]. For further guidance, see §1.6012-2T(c).
 * * * * *
 (k) [Reserved]. For further guidance, see §1.6012-2T(k)(1).
 
-Par. 52. Section 1.6012-2T is added to read as follows: §1.6012-2T Corporations required to make returns of income (temporary).
+Par. 52. Section 1.6012-2T is added to read as follows: <u>§1.6012-2T Corporations required to make returns of income (temporary)</u>.
 
 (a) through (b) [Reserved]. For further guidance, see §1.6012-2(a) through
 (b).
 (c) Insurance companies-- (1) Domestic life insurance companies-- (i) In
-general. A life insurance company subject to tax under section 801 shall make a return on Form 1120L. Except as provided in paragraph (c)(4) of this section, such company shall file with its return--
+<u>general</u>. A life insurance company subject to tax under section 801 shall make a return on Form 1120L. Except as provided in paragraph (c)(4) of this section, such company shall file with its return--
 
 (A) A copy of its annual statement which shows the reserves used by the
 company in computing the taxable income reported on its return; and
 
 (B) A copy of Schedule A (real estate) and of Schedule D (bonds and stocks),
-or any successor thereto, of such annual statement. (ii) Mutual savings banks. Mutual savings banks conducting life insurance business and meeting the requirements of section 594 are subject to partial tax computed on Form 1120 and partial tax computed on Form 1120L. The Form 1120L is attached as a schedule to Form 1120, together with the annual statement and schedules required to be filed with Form 1120L.
+or any successor thereto, of such annual statement. (ii) <u>Mutual savings banks</u>. Mutual savings banks conducting life insurance business and meeting the requirements of section 594 are subject to partial tax computed on Form 1120 and partial tax computed on Form 1120L. The Form 1120L is attached as a schedule to Form 1120, together with the annual statement and schedules required to be filed with Form 1120L.
 
-(2) Domestic nonlife insurance companies. Every domestic insurance
+(2) <u>Domestic nonlife insurance companies</u>. Every domestic insurance
 company other than a life insurance company shall make a return on Form 1120PC. This includes organizations described in section 501(m)(1) that provide commercial- type insurance and organizations described in section 833. Except as provided in paragraph (c)(4) of this section, such company shall file with its return a copy of its
 
 annual statement (or a pro forma annual statement), including the underwriting and investment exhibit for the year covered by such return.
 
-(3) Foreign insurance companies. The provisions of paragraphs (c)(1) and
+(3) <u>Foreign insurance companies</u>. The provisions of paragraphs (c)(1) and
 (c)(2) of this section concerning the returns and statements of insurance companies subject to tax under section 801 or section 831 also apply to foreign insurance companies subject to tax under those sections, except that the copy of the annual statement required to be submitted with the return shall, in the case of a foreign insurance company that is not required to file an annual statement, be a copy of the pro forma annual statement relating to the United States business of such company.
-(4) Exception for insurance companies filing their Federal income tax returns
-electronically. If an insurance company described in paragraph (c)(1), (c)(2), or
+(4) <u>Exception for insurance companies filing their Federal income tax returns</u>
+<u>electronically</u>. If an insurance company described in paragraph (c)(1), (c)(2), or
 
 (c)(3) of this section files its Federal income tax return electronically, it should not include on or with such return its annual statement (or pro forma annual statement), or any portion thereof. Such statement must be available at all times for inspection by authorized Internal Revenue Service officers or employees and retained for so long as such statements may be material in the administration of any internal revenue law. See §1.6001-1(e).
-(5) Definition. For purposes of this section, the term annual statement means
+(5) <u>Definition</u>. For purposes of this section, the term <u>annual statement</u> means
 the annual statement, the form of which is approved by the National Association of Insurance Commissioners (NAIC), which is filed by an insurance company for the year with the insurance departments of States, Territories, and the District of
 
 Columbia. The term annual statement also includes a pro forma annual statement if the insurance company is not required to file the NAIC annual statement.
 
 (d) through (j) [Reserved]. For further guidance, see §1.6012-2(d) through (j).
-(k) Effective date-- (1) Applicability date. This section applies to any original
+(k) <u>Effective date</u>-- (1) <u>Applicability date</u>. This section applies to any original
 Federal income tax return (including any amended return filed on or before the due date (including extensions) of such original return) timely filed on or after May 30,
 
 2006.
-(2) Expiration date. The applicability of this section will expire on May 26,
+(2) <u>Expiration date</u>. The applicability of this section will expire on May 26,
 2009.
 
 |||Par. 53. For each entry in the “Location” column of the following table,|
@@ -165,7 +156,7 @@ section and paragraph
 PART 602--OMB CONTROL NUMBERS UNDER THE PAPERWORK REDUCTION ACT Par. 54. The authority citation for part 602 continues to read as follows: Authority: 26 U.S.C. 7805. Par. 55. In §602.101, paragraph (b) is amended to read as follows:
 
 1. The following entries to the table are removed:
-§602.101 OMB Control numbers.
+<u>§602.101 OMB Control numbers</u>.
 
 * * * * *
 (b) * * *
@@ -180,7 +171,7 @@ CFR part or section where Current OMB identified or described control No.
 1.1081-11………………………………………………………………. 1545-2019
 * * * * * **______________________________________________________________**
 2. The following entries are added in numerical order to the table:
-§602.101 OMB Control numbers.
+<u>§602.101 OMB Control numbers</u>.
 
 * * * * *
 (b) * * *

tests/snapshots/thermo-freon12.md

@@ -6,27 +6,29 @@
 
 #### Thermodynamic Properties
 
-**of**
-
-®
+**of** ®
 
 # Freon 12
 
-**(R-12)** **Technical Information** **Technical Information**
+##### (R-12)
+
+##### Technical Information Technical Information
 
 **®** **Thermodynamic Properties of Freon 12 Refrigerant** **(R-12)** **SI Units**
 
 Tables of the thermodynamic **Units** properties of R-12 have been developed and are presented here. P = Pressure in kPa. Absolute This information is based on values calculated using the NIST REFPROP T = Temperature in Celcius Database (McLinden, M.O., Klein,
 
 S.A., Lemmon, E.W., and Peskin, Vf = Fluid (liquid) specific volume
-A.P., NIST Standard Reference in cubic meters per kilogram Database 23, NIST thermodynamic and transport properties of Vg = Vapour (gas) specific volume refrigerants and refrigerant in cubic meters per kilogram mixtures – REFPROP version 6.01, Standard Reference Data Program, df and dg = Fluid and Vapour National Institute of Standards and (respectively) densities in Technology, 1998). kilograms per cubic meter
+A.P., NIST Standard Reference in cubic meters per kilogram Database 23, NIST thermodynamic and transport properties of Vg = Vapour (gas) specific volume refrigerants and refrigerant in cubic meters per kilogram mixtures – REFPROP version 6.01, Standard Reference Data Program, df and dg = Fluid and Vapour National Institute of Standards and (respectively) densities in Technology, 1998).
+kilograms per cubic meter
+
 ##### H = Enthalpy (kJ/kg)
 
 ##### S = Entropy (kJ/kg.K)
 
 ##### Physical Properties
 
-|Chemical Formula|CCl2F2|
+|Chemical Formula|CCl₂F₂|
 |---|---|
 |Molecular mass|120.91|
 |Boiling Point At one atmosphere|-29.75°C|
@@ -43,9 +45,9 @@ l
 
 **Freon** **®** **12 Saturation Properties-Temperature Table**
 
-|Temp|Pressure||Volume|||Density||Enthalpy|||Entropy|Temp|
-|---|---|---|---|---|---|---|---|---|---|---|---|---|
-|°C|[kPa]|[m3 Liquid v f|/kg]|Vapour v g|Liquid d f|[kg/m3] Vapour d g|Liquid H f|[kJ/kg] Latent H fg|Vapour H g|Liquid S f|[kJ/K-kg] Vapour S g|°C|
+|Temp|Pressure||Volume||Density||Enthalpy|||Entropy|Temp|
+|---|---|---|---|---|---|---|---|---|---|---|---|
+|°C|[kPa]|[m³ Liquid v f|/kg] Vapour v g|[kg/m³ Liquid d f|] Vapour d g|Liquid H f|[kJ/kg] Latent H fg|Vapour H g|Liquid S f|[kJ/K-kg] Vapour S g|°C|
 
 |-100|1.2|0.0006|10.0000|1679.0|0.100|113.3|192.8|306.1|0.6077|1.7210|-100|
 |---|---|---|---|---|---|---|---|---|---|---|---|

tests/test_python.py

@@ -203,6 +203,79 @@ class TestExtractTextWithPositions:
         assert len(items) > 0
         assert all(item.page == 1 for item in items)
 
+    def test_mcid(self):
+        # Untagged fixture: mcid is None or int, never anything else
+        items = pdf_inspector.extract_text_with_positions(
+            fixture_path("thermo-freon12.pdf")
+        )
+        assert all(item.mcid is None or isinstance(item.mcid, int) for item in items)
+        # Tagged fixture: marked content carries MCIDs
+        tagged = pdf_inspector.extract_text_with_positions(
+            fixture_path("firecrawl_docs_tagged.pdf")
+        )
+        assert any(item.mcid is not None for item in tagged)
+
+
+# ---------------------------------------------------------------------------
+# extract_structure_elements / extract_structure_elements_bytes
+# ---------------------------------------------------------------------------
+
+
+class TestExtractStructureElements:
+    def test_tagged_file(self):
+        elements = pdf_inspector.extract_structure_elements(
+            fixture_path("firecrawl_docs_tagged.pdf")
+        )
+        assert len(elements) > 0
+        assert all(isinstance(e.page, int) for e in elements)
+        assert all(isinstance(e.mcid, int) for e in elements)
+        assert all(isinstance(e.role, str) and len(e.role) > 0 for e in elements)
+        assert any(e.role == "H1" for e in elements)
+
+    def test_join_with_text_items(self):
+        # (page, mcid) joins against extract_text_with_positions to recover
+        # heading text
+        path = fixture_path("firecrawl_docs_tagged.pdf")
+        elements = pdf_inspector.extract_structure_elements(path)
+        items = pdf_inspector.extract_text_with_positions(path)
+        h1_refs = {(e.page, e.mcid) for e in elements if e.role == "H1"}
+        h1_text = "".join(
+            item.text
+            for item in items
+            if item.mcid is not None and (item.page, item.mcid) in h1_refs
+        )
+        assert len(h1_text.strip()) > 0
+
+    def test_with_pages(self):
+        # pages filter is 1-indexed, matching TextItem.page
+        elements = pdf_inspector.extract_structure_elements(
+            fixture_path("firecrawl_docs_tagged.pdf"), pages=[1]
+        )
+        assert len(elements) > 0
+        assert all(e.page == 1 for e in elements)
+
+    def test_bytes(self):
+        data = fixture_bytes("firecrawl_docs_tagged.pdf")
+        elements = pdf_inspector.extract_structure_elements_bytes(data)
+        assert len(elements) > 0
+        assert any(e.role == "H1" for e in elements)
+
+    def test_untagged_returns_empty(self):
+        elements = pdf_inspector.extract_structure_elements(
+            fixture_path("thermo-freon12.pdf")
+        )
+        assert elements == []
+
+    def test_repr(self):
+        elements = pdf_inspector.extract_structure_elements(
+            fixture_path("firecrawl_docs_tagged.pdf")
+        )
+        assert "StructureElement" in repr(elements[0])
+
+    def test_not_a_pdf(self):
+        with pytest.raises(ValueError):
+            pdf_inspector.extract_structure_elements_bytes(b"not a pdf")
+
 
 # ---------------------------------------------------------------------------
 # extract_text_in_regions / extract_text_in_regions_bytes

wasm/Cargo.toml

@@ -0,0 +1,37 @@
+[package]
+name = "pdf-inspector-wasm"
+version = "1.14.0"
+edition = "2021"
+authors = ["Firecrawl Team"]
+description = "Browser WebAssembly bindings for pdf-inspector"
+license = "MIT"
+repository = "https://github.com/firecrawl/pdf-inspector"
+homepage = "https://github.com/firecrawl/pdf-inspector"
+readme = "README.md"
+publish = false
+
+[lib]
+crate-type = ["cdylib", "rlib"]
+
+[dependencies]
+console_error_panic_hook = "0.1"
+js-sys = "0.3"
+pdf-inspector = { path = ".." }
+serde = { version = "1", features = ["derive"] }
+serde-wasm-bindgen = "0.6"
+wasm-bindgen = "0.2"
+
+[dev-dependencies]
+wasm-bindgen-test = "0.3"
+
+[profile.release]
+codegen-units = 1
+lto = true
+opt-level = "s"
+strip = true
+
+[package.metadata.wasm-pack.profile.release]
+# Rust 1.95 emits bulk-memory instructions that the binaryen bundled with
+# wasm-pack 0.15.0 does not yet validate. rustc still performs the release,
+# size, and LTO optimizations above.
+wasm-opt = false

wasm/LICENSE

@@ -0,0 +1,58 @@
+MIT License
+
+Copyright (c) 2026 Firecrawl
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.
+
+Third-party notices
+===================
+
+Adobe CMaps
+-----------
+
+The WebAssembly binary embeds binary CMaps derived from Adobe CMap resources.
+
+Copyright 1990-2009 Adobe Systems Incorporated.
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+
+Redistributions of source code must retain the above copyright notice, this
+list of conditions and the following disclaimer.
+
+Redistributions in binary form must reproduce the above copyright notice,
+this list of conditions and the following disclaimer in the documentation
+and/or other materials provided with the distribution.
+
+Neither the name of Adobe Systems Incorporated nor the names of its
+contributors may be used to endorse or promote products derived from this
+software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+POSSIBILITY OF SUCH DAMAGE.

wasm/README.md

@@ -0,0 +1,60 @@
+# @firecrawl/pdf-inspector-wasm
+
+Browser WebAssembly bindings for [pdf-inspector](https://github.com/firecrawl/pdf-inspector). Classify PDFs and extract structured Markdown locally from a `Uint8Array`, using the same Rust core as the native Node.js, Python, and Rust packages.
+
+## Install
+
+```bash
+npm install @firecrawl/pdf-inspector-wasm
+```
+
+## Usage
+
+```ts
+import init, { processPdf } from "@firecrawl/pdf-inspector-wasm";
+
+await init();
+
+const response = await fetch("/annual-report.pdf");
+const pdf = new Uint8Array(await response.arrayBuffer());
+const result = processPdf(pdf);
+
+console.log(result.pdfType);
+console.log(result.markdown);
+```
+
+Pass options when you need selected pages or compact Markdown:
+
+```ts
+const result = processPdf(pdf, {
+  pages: [1, 3, 5],
+  profile: "compact",
+  includePageMarkers: true,
+});
+```
+
+The package also exports:
+
+- `detectPdf(pdf, options?)` for detection without extraction.
+- `classifyPdf(pdf)` for the lightweight result shape shared with the native Node.js API.
+- `extractText(pdf)` for plain text.
+- `version()` for the WASM package version.
+
+## Browser behavior
+
+- Parsing runs locally. PDF bytes are not uploaded anywhere.
+- The build is single-threaded and does not require cross-origin isolation.
+- CMaps are embedded so CJK font decoding does not depend on a filesystem.
+- Extraction is synchronous after `init()`. For large documents, call it from a Web Worker to keep the UI responsive.
+- Image-only documents still require a separate OCR step.
+
+## Build from source
+
+```bash
+cargo install wasm-pack --version 0.15.0 --locked
+wasm-pack build wasm --target web --scope firecrawl --release
+```
+
+## License
+
+MIT

wasm/src/lib.rs

@@ -0,0 +1,440 @@
+use pdf_inspector::{
+    LayoutComplexity, MarkdownProfile, PageOcrReasons, PdfOptions, PdfProcessResult, PdfType,
+    ProcessMode,
+};
+use serde::{Deserialize, Serialize};
+use wasm_bindgen::prelude::*;
+
+#[wasm_bindgen(typescript_custom_section)]
+const TYPESCRIPT_TYPES: &str = r#"
+export type PdfType = "TextBased" | "Scanned" | "ImageBased" | "Mixed";
+export type MarkdownProfile = "fidelity" | "compact";
+
+export interface ProcessOptions {
+  /** Restrict extraction to these 1-indexed page numbers. */
+  pages?: number[];
+  /** Password for an encrypted PDF. */
+  password?: string;
+  /** Source-faithful output by default, or compact output for fewer tokens. */
+  profile?: MarkdownProfile;
+  /** Insert `<!-- Page N -->` markers between pages. */
+  includePageMarkers?: boolean;
+  /** Include image placeholders in Markdown output. */
+  includeImages?: boolean;
+}
+
+export interface PageOcrReasons {
+  /** 1-indexed page number. */
+  page: number;
+  reasons: string[];
+}
+
+export interface LayoutComplexity {
+  isComplex: boolean;
+  /** 1-indexed page numbers. */
+  pagesWithTables: number[];
+  /** 1-indexed page numbers. */
+  pagesWithColumns: number[];
+}
+
+export interface PdfProcessResult {
+  pdfType: PdfType;
+  markdown?: string;
+  pageCount: number;
+  processingTimeMs: number;
+  /** 1-indexed page numbers. */
+  pagesNeedingOcr: number[];
+  ocrReasonsByPage: PageOcrReasons[];
+  title?: string;
+  confidence: number;
+  layout: LayoutComplexity;
+  hasEncodingIssues: boolean;
+}
+
+export interface PdfClassification {
+  pdfType: PdfType;
+  pageCount: number;
+  /** 0-indexed page numbers, matching the native Node.js API. */
+  pagesNeedingOcr: number[];
+  confidence: number;
+}
+
+export function processPdf(data: Uint8Array, options?: ProcessOptions): PdfProcessResult;
+export function detectPdf(data: Uint8Array, options?: Pick<ProcessOptions, "password">): PdfProcessResult;
+export function classifyPdf(data: Uint8Array): PdfClassification;
+export function extractText(data: Uint8Array): string;
+export function version(): string;
+"#;
+
+#[derive(Debug, Default, Deserialize)]
+#[serde(default, rename_all = "camelCase", deny_unknown_fields)]
+struct WasmProcessOptions {
+    pages: Option<Vec<u32>>,
+    password: Option<String>,
+    profile: Option<WasmMarkdownProfile>,
+    include_page_markers: Option<bool>,
+    include_images: Option<bool>,
+}
+
+#[derive(Debug, Deserialize)]
+#[serde(rename_all = "lowercase")]
+enum WasmMarkdownProfile {
+    Fidelity,
+    Compact,
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct WasmPageOcrReasons {
+    page: u32,
+    reasons: Vec<String>,
+}
+
+impl From<PageOcrReasons> for WasmPageOcrReasons {
+    fn from(value: PageOcrReasons) -> Self {
+        Self {
+            page: value.page,
+            reasons: value.reasons,
+        }
+    }
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct WasmLayoutComplexity {
+    is_complex: bool,
+    pages_with_tables: Vec<u32>,
+    pages_with_columns: Vec<u32>,
+}
+
+impl From<LayoutComplexity> for WasmLayoutComplexity {
+    fn from(value: LayoutComplexity) -> Self {
+        Self {
+            is_complex: value.is_complex,
+            pages_with_tables: value.pages_with_tables,
+            pages_with_columns: value.pages_with_columns,
+        }
+    }
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct WasmPdfProcessResult {
+    pdf_type: &'static str,
+    markdown: Option<String>,
+    page_count: u32,
+    processing_time_ms: f64,
+    pages_needing_ocr: Vec<u32>,
+    ocr_reasons_by_page: Vec<WasmPageOcrReasons>,
+    title: Option<String>,
+    confidence: f64,
+    layout: WasmLayoutComplexity,
+    has_encoding_issues: bool,
+}
+
+impl From<PdfProcessResult> for WasmPdfProcessResult {
+    fn from(value: PdfProcessResult) -> Self {
+        Self {
+            pdf_type: pdf_type_name(value.pdf_type),
+            markdown: value.markdown,
+            page_count: value.page_count,
+            processing_time_ms: value.processing_time_ms as f64,
+            pages_needing_ocr: value.pages_needing_ocr,
+            ocr_reasons_by_page: value
+                .ocr_reasons_by_page
+                .into_iter()
+                .map(Into::into)
+                .collect(),
+            title: value.title,
+            confidence: value.confidence as f64,
+            layout: value.layout.into(),
+            has_encoding_issues: value.has_encoding_issues,
+        }
+    }
+}
+
+#[derive(Serialize)]
+#[serde(rename_all = "camelCase")]
+struct WasmPdfClassification {
+    pdf_type: &'static str,
+    page_count: u32,
+    pages_needing_ocr: Vec<u32>,
+    confidence: f64,
+}
+
+fn pdf_type_name(pdf_type: PdfType) -> &'static str {
+    match pdf_type {
+        PdfType::TextBased => "TextBased",
+        PdfType::Scanned => "Scanned",
+        PdfType::ImageBased => "ImageBased",
+        PdfType::Mixed => "Mixed",
+    }
+}
+
+fn js_error(context: &str, error: impl std::fmt::Display) -> JsValue {
+    js_sys::Error::new(&format!("{context}: {error}")).into()
+}
+
+fn deserialize_options(value: JsValue) -> Result<WasmProcessOptions, JsValue> {
+    if value.is_undefined() || value.is_null() {
+        return Ok(WasmProcessOptions::default());
+    }
+
+    serde_wasm_bindgen::from_value(value).map_err(|error| js_error("invalid options", error))
+}
+
+fn build_options(value: JsValue, mode: ProcessMode) -> Result<PdfOptions, JsValue> {
+    let options = deserialize_options(value)?;
+    if options
+        .pages
+        .as_ref()
+        .is_some_and(|pages| pages.contains(&0))
+    {
+        return Err(js_error(
+            "invalid options",
+            "pages are 1-indexed; page 0 is invalid",
+        ));
+    }
+
+    let mut result = PdfOptions::new().mode(mode);
+    if let Some(pages) = options.pages {
+        result = result.pages(pages);
+    }
+    if let Some(password) = options.password {
+        result = result.password(password);
+    }
+    if let Some(profile) = options.profile {
+        result.markdown.profile = match profile {
+            WasmMarkdownProfile::Fidelity => MarkdownProfile::Fidelity,
+            WasmMarkdownProfile::Compact => MarkdownProfile::Compact,
+        };
+    }
+    if let Some(include_page_markers) = options.include_page_markers {
+        result.markdown.include_page_numbers = include_page_markers;
+    }
+    if let Some(include_images) = options.include_images {
+        result.markdown.include_images = include_images;
+    }
+    Ok(result)
+}
+
+fn serialize<T: Serialize>(value: &T) -> Result<JsValue, JsValue> {
+    serde_wasm_bindgen::to_value(value).map_err(|error| js_error("serialize result", error))
+}
+
+fn initialize() {
+    console_error_panic_hook::set_once();
+}
+
+/// Process PDF bytes entirely inside WebAssembly.
+#[wasm_bindgen(js_name = processPdf, skip_typescript)]
+pub fn process_pdf(data: &[u8], options: JsValue) -> Result<JsValue, JsValue> {
+    initialize();
+    let options = build_options(options, ProcessMode::Full)?;
+    let started = js_sys::Date::now();
+    let mut result = pdf_inspector::process_pdf_mem_with_options(data, options)
+        .map_err(|error| js_error("process PDF", error))?;
+    result.processing_time_ms = (js_sys::Date::now() - started).max(0.0) as u64;
+    serialize(&WasmPdfProcessResult::from(result))
+}
+
+/// Classify PDF bytes without extracting text or producing Markdown.
+#[wasm_bindgen(js_name = detectPdf, skip_typescript)]
+pub fn detect_pdf(data: &[u8], options: JsValue) -> Result<JsValue, JsValue> {
+    initialize();
+    let options = build_options(options, ProcessMode::DetectOnly)?;
+    let started = js_sys::Date::now();
+    let mut result = pdf_inspector::process_pdf_mem_with_options(data, options)
+        .map_err(|error| js_error("detect PDF", error))?;
+    result.processing_time_ms = (js_sys::Date::now() - started).max(0.0) as u64;
+    serialize(&WasmPdfProcessResult::from(result))
+}
+
+/// Return the lightweight classification shape used by the native Node API.
+#[wasm_bindgen(js_name = classifyPdf, skip_typescript)]
+pub fn classify_pdf(data: &[u8]) -> Result<JsValue, JsValue> {
+    initialize();
+    let result =
+        pdf_inspector::classify_pdf_mem(data).map_err(|error| js_error("classify PDF", error))?;
+    serialize(&WasmPdfClassification {
+        pdf_type: pdf_type_name(result.pdf_type),
+        page_count: result.page_count,
+        pages_needing_ocr: result.pages_needing_ocr,
+        confidence: result.confidence as f64,
+    })
+}
+
+/// Extract plain text from PDF bytes without Markdown conversion.
+#[wasm_bindgen(js_name = extractText, skip_typescript)]
+pub fn extract_text(data: &[u8]) -> Result<String, JsValue> {
+    initialize();
+    let items = pdf_inspector::extractor::extract_text_with_positions_mem(data)
+        .map_err(|error| js_error("extract text", error))?;
+    Ok(
+        pdf_inspector::extractor::group_into_lines_preserving_all_text(items)
+            .into_iter()
+            .map(|line| line.text())
+            .filter(|line| !line.trim().is_empty())
+            .collect::<Vec<_>>()
+            .join("\n"),
+    )
+}
+
+/// Return the WebAssembly package version.
+#[wasm_bindgen(skip_typescript)]
+pub fn version() -> String {
+    env!("CARGO_PKG_VERSION").to_string()
+}
+
+#[cfg(all(test, target_arch = "wasm32"))]
+mod tests {
+    use super::*;
+    use js_sys::Reflect;
+    use wasm_bindgen_test::*;
+
+    const TEXT_PDF: &[u8] = include_bytes!("../../tests/fixtures/thermo-freon12.pdf");
+    const ENCRYPTED_PDF: &[u8] = include_bytes!("../../tests/fixtures/encrypted-secret123.pdf");
+
+    fn synthetic_korea1_pdf() -> Vec<u8> {
+        let mut pdf = b"%PDF-1.4\n".to_vec();
+        let mut offsets = vec![0usize];
+
+        fn add_object(pdf: &mut Vec<u8>, offsets: &mut Vec<usize>, id: usize, body: &str) {
+            offsets.push(pdf.len());
+            pdf.extend_from_slice(format!("{id} 0 obj\n").as_bytes());
+            pdf.extend_from_slice(body.as_bytes());
+            pdf.extend_from_slice(b"\nendobj\n");
+        }
+
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            1,
+            "<< /Type /Catalog /Pages 2 0 R >>",
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            2,
+            "<< /Type /Pages /Kids [3 0 R] /Count 1 >>",
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            3,
+            "<< /Type /Page /Parent 2 0 R /MediaBox [0 0 200 200] /Resources << /Font << /F0 5 0 R >> >> /Contents 4 0 R >>",
+        );
+
+        // Adobe-Korea1 CID 1086 (0x043E) maps to U+AC00 (Korean syllable GA).
+        // There is deliberately no ToUnicode stream: decoding must use the
+        // embedded predefined CMap rather than lopdf's plain-text fallback.
+        // Korea1 CIDs 21 and 19 map to ASCII "4" and "2". Place them near
+        // the bottom edge so they look exactly like a numeric page footer.
+        let content = "BT /F0 12 Tf 50 100 Td <043E> Tj 0 -60 Td <00150013> Tj ET";
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            4,
+            &format!(
+                "<< /Length {} >>\nstream\n{}\nendstream",
+                content.len(),
+                content
+            ),
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            5,
+            "<< /Type /Font /Subtype /Type0 /BaseFont /SyntheticKorea1 /Encoding /Identity-H /DescendantFonts [6 0 R] >>",
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            6,
+            "<< /Type /Font /Subtype /CIDFontType2 /BaseFont /SyntheticKorea1 /CIDSystemInfo << /Registry (Adobe) /Ordering (Korea1) /Supplement 2 >> /FontDescriptor 7 0 R /DW 1000 >>",
+        );
+        add_object(
+            &mut pdf,
+            &mut offsets,
+            7,
+            "<< /Type /FontDescriptor /FontName /SyntheticKorea1 /Flags 4 /FontBBox [-100 -200 1000 900] /ItalicAngle 0 /Ascent 800 /Descent -200 /CapHeight 700 /StemV 80 >>",
+        );
+
+        let xref_start = pdf.len();
+        pdf.extend_from_slice(format!("xref\n0 {}\n", offsets.len()).as_bytes());
+        pdf.extend_from_slice(b"0000000000 65535 f \n");
+        for offset in offsets.iter().skip(1) {
+            pdf.extend_from_slice(format!("{offset:010} 00000 n \n").as_bytes());
+        }
+        pdf.extend_from_slice(
+            format!(
+                "trailer\n<< /Size {} /Root 1 0 R >>\nstartxref\n{}\n%%EOF",
+                offsets.len(),
+                xref_start
+            )
+            .as_bytes(),
+        );
+        pdf
+    }
+
+    #[wasm_bindgen_test]
+    fn processes_pdf_to_markdown() {
+        let result = process_pdf(TEXT_PDF, JsValue::UNDEFINED).expect("process PDF");
+        let pdf_type = Reflect::get(&result, &JsValue::from_str("pdfType"))
+            .expect("pdfType")
+            .as_string()
+            .expect("pdfType string");
+        let markdown = Reflect::get(&result, &JsValue::from_str("markdown"))
+            .expect("markdown")
+            .as_string()
+            .expect("markdown string");
+
+        assert_eq!(pdf_type, "TextBased");
+        assert!(!markdown.is_empty());
+    }
+
+    #[wasm_bindgen_test]
+    fn rejects_non_pdf_bytes() {
+        assert!(process_pdf(b"not a PDF", JsValue::UNDEFINED).is_err());
+    }
+
+    #[wasm_bindgen_test]
+    fn classifies_and_extracts_plain_text() {
+        let classification = classify_pdf(TEXT_PDF).expect("classify PDF");
+        let pdf_type = Reflect::get(&classification, &JsValue::from_str("pdfType"))
+            .expect("pdfType")
+            .as_string()
+            .expect("pdfType string");
+        let text = extract_text(TEXT_PDF).expect("extract text");
+
+        assert_eq!(pdf_type, "TextBased");
+        assert!(!text.is_empty());
+    }
+
+    #[wasm_bindgen_test]
+    fn extracts_cjk_and_preserves_numeric_page_footer() {
+        let text = extract_text(&synthetic_korea1_pdf()).expect("extract predefined CMap text");
+
+        assert_eq!(text, "가\n42");
+    }
+
+    #[wasm_bindgen_test]
+    fn opens_encrypted_pdf_with_password() {
+        assert!(process_pdf(ENCRYPTED_PDF, JsValue::UNDEFINED).is_err());
+
+        let options = js_sys::Object::new();
+        Reflect::set(
+            &options,
+            &JsValue::from_str("password"),
+            &JsValue::from_str("secret123"),
+        )
+        .expect("set password");
+        let result = process_pdf(ENCRYPTED_PDF, options.into()).expect("process encrypted PDF");
+        let markdown = Reflect::get(&result, &JsValue::from_str("markdown"))
+            .expect("markdown")
+            .as_string()
+            .expect("markdown string");
+
+        assert!(!markdown.is_empty());
+    }
+}