review: flag the ' show-text suppression path; require nonempty strings

- P1: invisible layers shown via the ' operator never set skipped_invisible,
  so those pages kept the placeholder and fell back to GPU OCR. The '
  suppression path now flags too — pinned by a fixture whose layer is shown
  entirely via ' (nothing through Tj/TJ).
- P3: all three flag sites (Tj, TJ, ') require a nonempty string operand —
  numeric-only TJ kerning arrays and empty shows no longer trigger the
  invisible reparse.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>

.github/workflows/ci.yml

@@ -24,6 +24,9 @@ jobs:
       - name: Cache cargo
         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
 

.github/workflows/publish-crate.yml

@@ -31,6 +31,9 @@ jobs:
         with:
           fetch-depth: 2
 
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
       - name: Check if version changed
         id: check
         run: |

.github/workflows/publish-pypi.yml

@@ -28,6 +28,9 @@ jobs:
         with:
           fetch-depth: 2
 
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
       - name: Check if version changed
         id: check
         run: |

.github/workflows/publish-wasm.yml

@@ -28,6 +28,9 @@ jobs:
         with:
           fetch-depth: 2
 
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
       - name: Check package version
         id: check
         run: |

.github/workflows/publish.yml

@@ -28,6 +28,9 @@ jobs:
         with:
           fetch-depth: 2
 
+      - name: Check package version sync
+        run: python3 scripts/version.py --check
+
       - name: Check if version changed
         id: check
         run: |

.gitignore

@@ -31,12 +31,12 @@ Thumbs.db
 napi/index.js
 napi/index.d.ts
 
-# Local samples and scripts
+# Local samples
 samples/
-scripts/
 
 # Test output
 test_output/
+.firecrawl/
 
 # Python
 __pycache__/

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "pdf-inspector"
-version = "0.1.7"
+version = "1.14.0"
 edition = "2021"
 autobins = false
 authors = ["Firecrawl Team"]

README.md

@@ -110,7 +110,7 @@ Or add it manually:
 
 ```toml
 [dependencies]
-pdf-inspector = "0.1"
+pdf-inspector = "1"
 ```
 
 ```rust

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/publishing.md

@@ -1,38 +1,57 @@
 # Publishing
 
-The Rust crate is published to [crates.io](https://crates.io/crates/pdf-inspector) with trusted publishing from GitHub Actions. The first release was published manually; future releases publish from `.github/workflows/publish-crate.yml` when a `Cargo.toml` version change lands on `main`.
+Every pdf-inspector distribution uses one shared semantic version:
 
-## crates.io Trusted Publisher
+- 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
 
-Configure the trusted publisher for the `pdf-inspector` crate with:
+`Cargo.toml` is the canonical version source. Update every manifest and lockfile
+with:
 
-- Repository: `firecrawl/pdf-inspector`
-- Workflow: `publish-crate.yml`
-- Environment: `crates-io`
+```bash
+python3 scripts/version.py <version>
+```
 
-The workflow uses `rust-lang/crates-io-auth-action@v1` to exchange GitHub's OIDC token for a short-lived crates.io token, then passes it to `cargo publish`.
+Verify that nothing has diverged with:
 
-## Release Steps
+```bash
+python3 scripts/version.py --check
+```
 
-1. Update `version` in `Cargo.toml`.
-2. Merge the version bump to `main`.
-3. The publish workflow compares the new `Cargo.toml` version with `HEAD~1`, runs `cargo publish --dry-run`, then publishes if that version is not already on crates.io.
+CI and every publishing workflow run this check before building or publishing.
 
-If `Cargo.toml` changes without a package version bump, the workflow exits without publishing.
+## Release steps
 
-## Browser WebAssembly package
+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 browser package is published as `@firecrawl/pdf-inspector-wasm`. Its version lives in `wasm/Cargo.toml`, and `.github/workflows/publish-wasm.yml` builds the `web` target with `wasm-pack` before publishing the generated package.
+The independent workflows are intentionally idempotent. A manual dispatch from
+`main` can repair a partial release, and already-published artifacts are skipped.
 
-The npm package must exist before a trusted publisher can be configured. For the first release only:
+## Trusted publishers
 
-1. Build with `wasm-pack build wasm --target web --scope firecrawl --out-dir pkg --release`.
-2. Inspect with `npm pack --dry-run ./wasm/pkg`.
-3. Publish with `npm publish ./wasm/pkg --access public` from an authorized maintainer session.
-4. In the package settings on npm, configure the GitHub Actions trusted publisher:
-   - Organization: `firecrawl`
-   - Repository: `pdf-inspector`
-   - Workflow: `publish-wasm.yml`
-   - Allowed action: `npm publish`
+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:
 
-After that one-time bootstrap, bumping the version in `wasm/Cargo.toml` and merging it to `main` publishes through OIDC. Until the package exists, the workflow exits cleanly without attempting an unauthenticated first publish. See npm's [trusted publishing documentation](https://docs.npmjs.com/trusted-publishers/) for the registry-side setup.
+- 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
+```

napi/Cargo.lock

@@ -851,7 +851,7 @@ checksum = "384b8ab6d37215f3c5301a95a4accb5d64aa607f1fcb26a11b5303878451b4fe"
 
 [[package]]
 name = "pdf-inspector"
-version = "0.1.7"
+version = "1.14.0"
 dependencies = [
  "env_logger",
  "include_dir",
@@ -867,7 +867,7 @@ dependencies = [
 
 [[package]]
 name = "pdf-inspector-napi"
-version = "0.2.2"
+version = "1.14.0"
 dependencies = [
  "napi",
  "napi-build",

napi/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "pdf-inspector-napi"
-version = "0.2.2"
+version = "1.14.0"
 edition = "2021"
 
 [lib]

napi/README.md

@@ -83,6 +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

napi/bun.lock

@@ -8,12 +8,12 @@
         "@napi-rs/cli": "^3.4.1",
       },
       "optionalDependencies": {
-        "@firecrawl/pdf-inspector-darwin-arm64": "1.12.0",
-        "@firecrawl/pdf-inspector-linux-arm64-gnu": "1.12.0",
-        "@firecrawl/pdf-inspector-linux-arm64-musl": "1.12.0",
-        "@firecrawl/pdf-inspector-linux-x64-gnu": "1.12.0",
-        "@firecrawl/pdf-inspector-linux-x64-musl": "1.12.0",
-        "@firecrawl/pdf-inspector-win32-x64-msvc": "1.12.0",
+        "@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",
       },
     },
   },

napi/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@firecrawl/pdf-inspector",
-  "version": "1.12.0",
+  "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",
@@ -52,11 +52,11 @@
     "@napi-rs/cli": "^3.4.1"
   },
   "optionalDependencies": {
-    "@firecrawl/pdf-inspector-linux-x64-gnu": "1.12.0",
-    "@firecrawl/pdf-inspector-linux-x64-musl": "1.12.0",
-    "@firecrawl/pdf-inspector-linux-arm64-gnu": "1.12.0",
-    "@firecrawl/pdf-inspector-linux-arm64-musl": "1.12.0",
-    "@firecrawl/pdf-inspector-darwin-arm64": "1.12.0",
-    "@firecrawl/pdf-inspector-win32-x64-msvc": "1.12.0"
+    "@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

@@ -89,6 +89,11 @@ pub struct TextItem {
     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).
@@ -153,9 +158,7 @@ fn to_napi_result(r: pdf_inspector::PdfProcessResult) -> PdfResult {
     }
 }
 
-fn to_napi_page_ocr_reasons(
-    reasons: Vec<pdf_inspector::PageOcrReasons>,
-) -> Vec<PageOcrReasons> {
+fn to_napi_page_ocr_reasons(reasons: Vec<pdf_inspector::PageOcrReasons>) -> Vec<PageOcrReasons> {
     reasons
         .into_iter()
         .map(|reason| PageOcrReasons {
@@ -202,6 +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
 // ---------------------------------------------------------------------------
@@ -210,15 +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.
@@ -238,16 +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.
@@ -300,12 +311,61 @@ pub fn extract_text_with_positions(
                     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,
@@ -633,25 +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,
-                    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,
-        })
+        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,
     })
 }
 
@@ -692,3 +759,131 @@ fn to_page_region_texts(results: Vec<pdf_inspector::PageRegionResult>) -> Vec<Pa
         })
         .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

@@ -51,6 +51,20 @@ class TextItem:
     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."""
@@ -132,6 +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

@@ -4,9 +4,9 @@ build-backend = "maturin"
 
 [project]
 name = "pdf-inspector"
-# Bump this to publish to PyPI — CI publishes automatically when the version
-# changes on main (same flow as napi/package.json for npm).
-version = "0.2.6"
+# 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" }

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/index.html

@@ -858,22 +858,22 @@
           <p>Evaluated on the <a class="text-link" href="https://github.com/opendataloader-project/opendataloader-bench">opendataloader-bench</a> corpus of 200 PDFs. This comparison covers local engines without model-based PDF parsing, with OCR disabled. Higher scores are better.</p>
         </div>
         <div class="benchmark-card">
-          <div class="benchmark-top"><span><strong>200 PDFs</strong> · OpenDataLoader benchmark</span><span>Apple M4 Pro · median of 3 runs</span></div>
+          <div class="benchmark-top"><span><strong>200 PDFs</strong> · OpenDataLoader benchmark</span><span>Apple M4 Pro · median of 5 runs</span></div>
           <div class="table-scroll">
             <table aria-label="PDF extraction benchmark results">
               <thead>
                 <tr><th>Engine</th><th>Overall</th><th>Reading order</th><th>Tables</th><th>Headings</th><th>Complete run</th></tr>
               </thead>
               <tbody>
-                <tr class="highlight"><td>pdf-inspector</td><td>0.875</td><td>0.915</td><td>0.814</td><td>0.788</td><td>2.8s</td></tr>
-                <tr><td>LiteParse</td><td>0.870</td><td>0.908</td><td>0.693</td><td>0.811</td><td>13.9s</td></tr>
-                <tr><td>OpenDataLoader</td><td>0.843</td><td>0.912</td><td>0.489</td><td>0.760</td><td>9.8s</td></tr>
-                <tr><td>PyMuPDF4LLM</td><td>0.735</td><td>0.886</td><td>0.401</td><td>0.424</td><td>15.5s</td></tr>
-                <tr><td>MarkItDown</td><td>0.583</td><td>0.879</td><td>0.000</td><td>0.000</td><td>6.7s</td></tr>
+                <tr class="highlight"><td>pdf-inspector</td><td>0.875</td><td>0.915</td><td>0.814</td><td>0.788</td><td>0.470s</td></tr>
+                <tr><td>LiteParse</td><td>0.873</td><td>0.913</td><td>0.693</td><td>0.811</td><td>0.750s</td></tr>
+                <tr><td>OpenDataLoader</td><td>0.831</td><td>0.902</td><td>0.489</td><td>0.739</td><td>2.569s</td></tr>
+                <tr><td>PyMuPDF4LLM</td><td>0.735</td><td>0.886</td><td>0.401</td><td>0.424</td><td>17.117s</td></tr>
+                <tr><td>MarkItDown</td><td>0.589</td><td>0.844</td><td>0.273</td><td>0.000</td><td>16.165s</td></tr>
               </tbody>
             </table>
           </div>
-          <div class="benchmark-note">Refreshed July 16, 2026. Scores use the benchmark’s NID, TEDS, and MHS evaluators.</div>
+          <div class="benchmark-note">Refreshed July 31, 2026. Scores use the benchmark’s NID, TEDS, and MHS evaluators; speed is the median of five alternating or rotating complete corpus runs after an excluded warm-up. <a class="text-link" href="https://github.com/firecrawl/opendataloader-bench/tree/abi/pdf-parser-benchmark-results">Versions and raw artifacts</a>.</div>
         </div>
         <div class="best-fit">
           <strong>Best fit</strong>
@@ -975,7 +975,7 @@ result = pdf_inspector.<span class="fn">process_pdf</span>(<span class="str">"do
 <script>
   (() => {
     const MAX_FILE_SIZE = 25 * 1024 * 1024;
-    const WASM_MODULE_URL = "https://cdn.jsdelivr.net/npm/@firecrawl/pdf-inspector-wasm@0.1.1/pdf_inspector_wasm.js";
+    const WASM_MODULE_URL = "https://cdn.jsdelivr.net/npm/@firecrawl/pdf-inspector-wasm@1.14.0/pdf_inspector_wasm.js";
     const input = document.querySelector("#pdf-input");
     const dropZone = document.querySelector("#drop-zone");
     const filePanel = document.querySelector("#demo-file");

src/detector.rs

@@ -1659,7 +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
@@ -1741,6 +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(

src/extractor/content_stream.rs

@@ -137,8 +137,11 @@ fn rise_adjusted(tm: &[f32; 6], rise: f32) -> [f32; 6] {
     ]
 }
 
-/// Returns `(page_extraction, has_gid_fonts)` where `has_gid_fonts` indicates
-/// the page uses fonts with unresolvable gid-encoded glyphs.
+/// Returns `(page_extraction, has_gid_fonts, coords_rotated, skipped_invisible)`
+/// where `has_gid_fonts` indicates the page uses fonts with unresolvable
+/// gid-encoded glyphs and `skipped_invisible` reports that invisible (Tr 3)
+/// text was present but suppressed — callers can use it to decide whether an
+/// `include_invisible` retry could recover anything at all.
 pub(crate) fn extract_page_text_items(
     doc: &Document,
     page_id: ObjectId,
@@ -146,7 +149,7 @@ pub(crate) fn extract_page_text_items(
     font_cmaps: &FontCMaps,
     include_invisible: bool,
     style_cache: &mut FontStyleCache,
-) -> Result<(PageExtraction, bool, bool), PdfError> {
+) -> Result<(PageExtraction, bool, bool, bool), PdfError> {
     use lopdf::content::Content;
 
     let mut items = Vec::new();
@@ -262,12 +265,15 @@ pub(crate) fn extract_page_text_items(
             content.operations.len(),
             MAX_OPERATIONS
         );
-        return Ok(((Vec::new(), Vec::new(), Vec::new()), false, false));
+        return Ok(((Vec::new(), Vec::new(), Vec::new()), false, false, false));
     }
 
     // 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
+                                          // Invisible (Tr 3) text was present but suppressed — reported to callers
+                                          // so an include_invisible retry is attempted only when it can recover.
+    let mut skipped_invisible = false;
     let mut line_width: f32 = 1.0;
     #[derive(Clone)]
     struct SavedGraphicsState {
@@ -494,6 +500,14 @@ pub(crate) fn extract_page_text_items(
                     // For Mixed/template PDFs, include_invisible=true extracts
                     // the OCR text layer that sits behind scanned images.
                     if text_rendering_mode == 3 && !include_invisible {
+                        if op
+                            .operands
+                            .first()
+                            .and_then(get_operand_bytes)
+                            .is_some_and(|raw| !raw.is_empty())
+                        {
+                            skipped_invisible = true;
+                        }
                         if let Some(w_ts) = w_ts_opt {
                             text_matrix[4] += w_ts * text_matrix[0];
                             text_matrix[5] += w_ts * text_matrix[1];
@@ -565,6 +579,16 @@ pub(crate) fn extract_page_text_items(
                 if in_text_block && !op.operands.is_empty() {
                     if let Ok(array) = op.operands[0].as_array() {
                         let font_info = font_widths.get(&current_font);
+                        // Numeric-only TJ arrays (pure kerning) show no
+                        // text — they must not trigger the invisible retry.
+                        if text_rendering_mode == 3
+                            && !include_invisible
+                            && array
+                                .iter()
+                                .any(|el| get_operand_bytes(el).is_some_and(|raw| !raw.is_empty()))
+                        {
+                            skipped_invisible = true;
+                        }
                         let is_invisible = (text_rendering_mode == 3 && !include_invisible)
                             || suppress_glyph_extraction;
                         // Capture first-glyph position for ActualText
@@ -770,6 +794,16 @@ pub(crate) fn extract_page_text_items(
                         )
                     })
                 });
+                if text_rendering_mode == 3
+                    && !include_invisible
+                    && op
+                        .operands
+                        .first()
+                        .and_then(get_operand_bytes)
+                        .is_some_and(|raw| !raw.is_empty())
+                {
+                    skipped_invisible = true;
+                }
                 if !((text_rendering_mode == 3 && !include_invisible)
                     || suppress_glyph_extraction
                     || op.operands.is_empty())
@@ -1300,7 +1334,12 @@ pub(crate) fn extract_page_text_items(
 
     let items = super::merge_text_items(items);
     let items = super::merge_subscript_items(items);
-    Ok(((items, rects, lines), has_gid_fonts, coords_rotated))
+    Ok((
+        (items, rects, lines),
+        has_gid_fonts,
+        coords_rotated,
+        skipped_invisible,
+    ))
 }
 
 /// Counts of text operators with horizontal vs rotated combined matrices.
@@ -1498,7 +1537,7 @@ mod tests {
 
         let (doc, page_id) = simple_doc_with_content(content);
         let font_cmaps = FontCMaps::from_doc(&doc);
-        let ((items, _, _), _, _) = extract_page_text_items(
+        let ((items, _, _), _, _, _) = extract_page_text_items(
             &doc,
             page_id,
             1,
@@ -1736,7 +1775,7 @@ BT /F1 12 Tf 0 1 -1 0 240 100 Tm (WORLD) Tj ET
             &mut FontStyleCache::new(),
         )
         .unwrap();
-        let ((items, rects, lines), _has_gid, _coords_rotated) = result;
+        let ((items, rects, lines), _has_gid, _coords_rotated, _skipped_invisible) = result;
         assert!(items.is_empty());
         assert!(rects.is_empty());
         assert!(lines.is_empty());
@@ -1817,7 +1856,7 @@ BT 30 700 Tm <41> Tj ET";
         doc.trailer.set("Root", Object::Reference(catalog_id));
 
         let font_cmaps = FontCMaps::from_doc(&doc);
-        let ((items, _, _), _, _) = extract_page_text_items(
+        let ((items, _, _), _, _, _) = extract_page_text_items(
             &doc,
             page_id,
             1,

src/extractor/layout.rs

@@ -41,15 +41,110 @@ pub(crate) fn detect_columns(
     }
     debug!("page {}: detect_columns: {} items", page, page_items.len());
 
-    // Find page bounds
-    let x_min = page_items.iter().map(|i| i.x).fold(f32::INFINITY, f32::min);
-    let x_max = page_items
-        .iter()
-        .map(|i| i.x + effective_width(i))
-        .fold(f32::NEG_INFINITY, f32::max);
+    // The width of one ordinary page, used three ways below: as the largest
+    // credible width for a single text run, as the size of empty gap that marks
+    // content as detached, and as the span past which those checks run at all.
+    // This is a heuristic, not a format rule: PDF 2.0 sets no page-size limit,
+    // and since PDF 1.6 `UserUnit` scales a page's physical size independently
+    // of its coordinates. 14_400 units (200in at the default 1/72in unit) is
+    // the traditional Acrobat architectural limit, which makes it a reasonable
+    // "wider than any ordinary page" mark in coordinate space.
+    const MAX_PAGE_EXTENT: f32 = 14_400.0;
+    // A detached cluster is only dropped if it also holds a small minority of
+    // the items, so a genuine two-part layout keeps its full bounds even when
+    // the halves are far apart.
+    const MAX_TRIM_FRACTION: f32 = 0.10;
+
+    // Position and width of each item, skipping only non-finite geometry.
+    let finite_span = |i: &&TextItem| -> Option<(f32, f32)> {
+        let (left, width) = (i.x, effective_width(i));
+        (left.is_finite() && (left + width).is_finite()).then_some((left, width))
+    };
+
+    let (min_left, max_right, total) = page_items.iter().filter_map(finite_span).fold(
+        (f32::INFINITY, f32::NEG_INFINITY, 0usize),
+        |(lo, hi, n), (left, width)| (lo.min(left), hi.max(left + width), n + 1),
+    );
+
+    // No item had usable geometry, so there is no layout to report.
+    if total == 0 {
+        return vec![];
+    }
+
+    // Every threshold below (gutter margins, spanning-item width, the XY-cut
+    // margin) is a fraction of the page width, so a far item can set the scale
+    // for the whole page and shrink the effective detection window to a
+    // rounding error — real gutters then fall inside the margin band and a
+    // genuine multi-column page collapses to one region.
+    //
+    // Anything inside one page extent is ordinary, so the common case keeps the
+    // plain bounds and skips the work below entirely.
+    let (x_min, x_max) = if max_right - min_left <= MAX_PAGE_EXTENT {
+        (min_left, max_right)
+    } else {
+        // Discarding content needs positive evidence that it is not part of the
+        // layout, because a count-based rule alone cannot tell a stray from a
+        // sparse far sidebar. The evidence is geometric: positions are grouped
+        // into clusters separated by more than a whole page of continuous
+        // emptiness. Real content, however sparse, does not leave a void that
+        // large; a malformed coordinate sits alone beyond one.
+        let mut spans: Vec<(f32, f32)> = page_items.iter().filter_map(finite_span).collect();
+        spans.sort_by(|a, b| a.0.total_cmp(&b.0));
+
+        let mut core: Option<std::ops::Range<usize>> = None;
+        let mut start = 0usize;
+        for i in 1..=spans.len() {
+            if i < spans.len() && spans[i].0 - spans[i - 1].0 <= MAX_PAGE_EXTENT {
+                continue;
+            }
+            if core.as_ref().is_none_or(|best| i - start > best.len()) {
+                core = Some(start..i);
+            }
+            start = i;
+        }
+        let mut core = core.unwrap_or(0..spans.len());
+
+        // Only drop the detached clusters when they are a small minority, so a
+        // genuine two-part layout keeps its full bounds.
+        let dropped = spans.len() - core.len();
+        if dropped as f32 > spans.len() as f32 * MAX_TRIM_FRACTION {
+            core = 0..spans.len();
+        }
+        let core = &spans[core];
+
+        // Positions cannot be inflated by a bogus width, so the spread of the
+        // content is a sound scale for judging one. A run much wider than the
+        // page's own content is a malformed width — the test is relative, so a
+        // genuinely large page keeps its genuinely long runs.
+        let (lo, widest_left) = (core[0].0, core[core.len() - 1].0);
+        let max_run_width = (widest_left - lo) + MAX_PAGE_EXTENT;
+        let hi = core
+            .iter()
+            .filter(|&&(_, width)| width <= max_run_width)
+            .map(|&(left, width)| left + width)
+            .fold(widest_left, f32::max);
+
+        if lo != min_left || hi != max_right {
+            debug!(
+                "page {page}: bounds {min_left}..{max_right} exceed one page; \
+                 dropped {dropped}/{} detached item(s), using {lo}..{hi}",
+                spans.len()
+            );
+        }
+        (lo, hi)
+    };
+
+    // Hard ceiling on the histogram size, independent of the trimming above:
+    // the bounds are attacker-influenced, so an unclamped
+    // `page_width / BIN_WIDTH` lets a crafted PDF force an arbitrarily large
+    // `vec![0u32; num_bins]` allocation. 65_536 bins covers ~128k points at
+    // BIN_WIDTH 2.0 — roughly 9x the largest legal page — so this never binds
+    // on a real layout. Kept as a bound that does not depend on the outlier
+    // heuristic staying correct.
+    const MAX_BINS: usize = 65_536;
 
     let page_width = x_max - x_min;
-    if page_width < 200.0 {
+    if !page_width.is_finite() || page_width < 200.0 {
         return vec![ColumnRegion { x_min, x_max }];
     }
 
@@ -57,13 +152,20 @@ pub(crate) fn detect_columns(
         return vec![ColumnRegion { x_min, x_max }];
     }
 
+    // Widen the bins rather than dropping the tail of the page. Clamping the
+    // count alone would leave anything past MAX_BINS * BIN_WIDTH outside the
+    // histogram, folded into the last bin, which places gutters at the wrong
+    // coordinates. Scaling keeps full coverage under the same allocation
+    // ceiling; only the resolution degrades, and only beyond ~131k points.
+    let bin_width = BIN_WIDTH.max(page_width / MAX_BINS as f32);
+
     // Build occupancy histogram.
     // Exclude items wider than 60% of page width — these are spanning items
     // (titles, full-width paragraphs) that would fill the gutter and prevent
     // detection of partial-page column layouts (e.g. two-column abstracts on
     // a page that also has single-column introduction text).
     let wide_threshold = page_width * 0.6;
-    let num_bins = ((page_width / BIN_WIDTH).ceil() as usize).max(1);
+    let num_bins = ((page_width / bin_width).ceil() as usize).clamp(1, MAX_BINS);
     let mut histogram = vec![0u32; num_bins];
 
     for item in &page_items {
@@ -71,8 +173,8 @@ pub(crate) fn detect_columns(
         if w > wide_threshold {
             continue;
         }
-        let left = ((item.x - x_min) / BIN_WIDTH).floor() as usize;
-        let right = (((item.x + w) - x_min) / BIN_WIDTH).ceil() as usize;
+        let left = ((item.x - x_min) / bin_width).floor() as usize;
+        let right = (((item.x + w) - x_min) / bin_width).ceil() as usize;
         let left = left.min(num_bins);
         let right = right.min(num_bins);
         for count in histogram.iter_mut().take(right).skip(left) {
@@ -109,12 +211,12 @@ pub(crate) fn detect_columns(
     let valleys: Vec<(usize, usize)> = valleys
         .into_iter()
         .filter(|&(start, end)| {
-            let width_pts = (end - start) as f32 * BIN_WIDTH;
+            let width_pts = (end - start) as f32 * bin_width;
             if width_pts < MIN_GUTTER_WIDTH {
                 return false;
             }
             // Valley center must not be within 5% of page edges
-            let center_pts = ((start + end) as f32 / 2.0) * BIN_WIDTH;
+            let center_pts = ((start + end) as f32 / 2.0) * bin_width;
             center_pts > margin_threshold && center_pts < (page_width - margin_threshold)
         })
         .collect();
@@ -132,7 +234,7 @@ pub(crate) fn detect_columns(
             &histogram,
             num_bins,
             x_min,
-            BIN_WIDTH,
+            bin_width,
             page_width,
             margin_threshold,
         );
@@ -141,7 +243,7 @@ pub(crate) fn detect_columns(
                 &rel_valleys,
                 &page_items,
                 x_min,
-                BIN_WIDTH,
+                bin_width,
                 x_max,
                 MIN_ITEMS_PER_COLUMN,
                 MIN_VERTICAL_SPAN_RATIO,
@@ -182,7 +284,7 @@ pub(crate) fn detect_columns(
         &valleys,
         &page_items,
         x_min,
-        BIN_WIDTH,
+        bin_width,
         x_max,
         MIN_ITEMS_PER_COLUMN,
         MIN_VERTICAL_SPAN_RATIO,
@@ -196,7 +298,7 @@ pub(crate) fn detect_columns(
         &valleys,
         &page_items,
         x_min,
-        BIN_WIDTH,
+        bin_width,
         x_max,
         MIN_ITEMS_PER_COLUMN,
         MIN_VERTICAL_SPAN_RATIO,
@@ -1827,7 +1929,7 @@ fn split_column_stragglers(lines: Vec<TextLine>) -> (Vec<TextLine>, Vec<TextLine
         .unwrap();
 
     let (cs, ce) = segments[core_seg];
-    let mut core = Vec::with_capacity(ce - cs);
+    let mut core = Vec::with_capacity(ce.saturating_sub(cs));
     let mut stragglers = Vec::new();
     for (i, line) in lines.into_iter().enumerate() {
         if i >= cs && i < ce {
@@ -2533,6 +2635,214 @@ mod tests {
         );
     }
 
+    #[test]
+    fn extreme_far_coordinate_does_not_allocate_unboundedly() {
+        // A crafted PDF can place a text run at an arbitrary coordinate via the
+        // text matrix. The derived page width must not drive an unbounded
+        // histogram allocation (previously `page_width / BIN_WIDTH` bins with no
+        // upper bound would try to reserve terabytes and abort the process).
+        let mut items = Vec::new();
+        for i in 0..24 {
+            items.push(make_item(1, i as f32 * 10.0, 700.0 - i as f32 * 5.0, "A"));
+        }
+        // Item placed 1e12 points away — 5e11 bins if left unclamped.
+        items.push(make_item(1, 1e12, 700.0, "Z"));
+
+        // Must return without aborting; content is preserved as a single region.
+        let cols = detect_columns(&items, 1, false);
+        assert!(!cols.is_empty());
+    }
+
+    #[test]
+    fn non_finite_coordinates_never_leak_into_region_bounds() {
+        // An inf/NaN coordinate must not escape as a column boundary: callers
+        // treat these as page/column edges.
+        for bad_x in [f32::INFINITY, f32::NEG_INFINITY, f32::NAN] {
+            let mut items = Vec::new();
+            for i in 0..24 {
+                items.push(make_item(1, i as f32 * 10.0, 700.0 - i as f32 * 5.0, "A"));
+            }
+            items.push(make_item(1, bad_x, 700.0, "Z"));
+
+            for col in detect_columns(&items, 1, false) {
+                assert!(
+                    col.x_min.is_finite() && col.x_max.is_finite(),
+                    "bad_x {bad_x} leaked bounds {}..{}",
+                    col.x_min,
+                    col.x_max
+                );
+            }
+        }
+    }
+
+    #[test]
+    fn all_non_finite_coordinates_yield_no_columns() {
+        let items: Vec<TextItem> = (0..24)
+            .map(|i| make_item(1, f32::NAN, 700.0 - i as f32 * 5.0, "A"))
+            .collect();
+
+        assert!(detect_columns(&items, 1, false).is_empty());
+    }
+
+    #[test]
+    fn one_bad_item_does_not_disable_column_detection() {
+        // A single stray item should not collapse a clean two-column page to
+        // one region. Every gutter threshold is a fraction of the page width,
+        // so an untrimmed outlier pushes real gutters inside the rejected
+        // margin band. A malformed *width* at an ordinary position poisons the
+        // bounds just as a malformed position does.
+        for (label, bad_x, bad_width) in [
+            ("nan position", f32::NAN, 0.0),
+            ("inf position", f32::INFINITY, 0.0),
+            ("far position", 50_000.0, 0.0),
+            ("very far position", 1e12, 0.0),
+            ("huge width", 100.0, 1e12),
+            ("inf width", 100.0, f32::INFINITY),
+        ] {
+            let mut items = Vec::new();
+            items.extend(fill_zone(1, 30.0, 280.0, 750.0, 50.0));
+            items.extend(fill_zone(1, 320.0, 570.0, 750.0, 50.0));
+            let mut bad = make_item(1, bad_x, 400.0, "Z");
+            bad.width = bad_width;
+            items.push(bad);
+
+            let cols = detect_columns(&items, 1, false);
+            assert_eq!(
+                cols.len(),
+                2,
+                "{label}: expected 2 columns, got {}",
+                cols.len()
+            );
+            for col in &cols {
+                assert!(
+                    col.x_max - col.x_min <= MAX_PAGE_EXTENT_FOR_TEST,
+                    "{label}: region {}..{} exceeds one page",
+                    col.x_min,
+                    col.x_max
+                );
+            }
+        }
+    }
+
+    /// Mirrors `MAX_PAGE_EXTENT` in `detect_columns`.
+    const MAX_PAGE_EXTENT_FOR_TEST: f32 = 14_400.0;
+
+    #[test]
+    fn very_wide_page_keeps_full_histogram_coverage() {
+        // Beyond MAX_BINS * BIN_WIDTH (~131k points) the bins must widen rather
+        // than stop covering the page. Three zones: the first gutter is inside
+        // the old coverage limit, the second is past it. Because the first
+        // gutter is found, the XY-cut fallback never runs, so a truncated
+        // histogram silently reports two columns instead of three.
+        let mut items = Vec::new();
+        items.extend(fill_zone(1, 0.0, 60_000.0, 750.0, 700.0));
+        items.extend(fill_zone(1, 70_000.0, 140_000.0, 750.0, 700.0));
+        items.extend(fill_zone(1, 160_000.0, 200_000.0, 750.0, 700.0));
+
+        let cols = detect_columns(&items, 1, false);
+        assert_eq!(
+            cols.len(),
+            3,
+            "Expected 3 columns across a 200k-wide page, got {}",
+            cols.len()
+        );
+        assert!(
+            (140_000.0..=160_000.0).contains(&cols[1].x_max),
+            "second gutter at {}, expected inside the real 140k..160k gap",
+            cols[1].x_max
+        );
+    }
+
+    #[test]
+    fn large_page_with_legitimately_long_runs_is_kept() {
+        // On a very large page, individual runs can exceed one ordinary page's
+        // width. They are real content, so they must not be judged malformed:
+        // the page keeps its columns and its full right edge.
+        let mut items = Vec::new();
+        for row in 0..30 {
+            let y = 750.0 - row as f32 * 14.0;
+            let mut left = make_item(1, 0.0, y, "Left run");
+            left.width = 20_000.0;
+            let mut right = make_item(1, 25_000.0, y, "Right run");
+            right.width = 20_000.0;
+            items.extend([left, right]);
+        }
+
+        let cols = detect_columns(&items, 1, false);
+        assert!(
+            !cols.is_empty(),
+            "a page of long-but-valid runs must still report a layout"
+        );
+        let right_edge = cols
+            .iter()
+            .map(|c| c.x_max)
+            .fold(f32::NEG_INFINITY, f32::max);
+        assert!(
+            right_edge > 44_000.0,
+            "long runs were treated as malformed: right edge {right_edge}, expected ~45_000"
+        );
+    }
+
+    #[test]
+    fn sparse_far_sidebar_on_a_large_page_is_kept() {
+        // A large-format page with a thin, sparsely-populated sidebar far from
+        // the main block. The sidebar is a small minority of the items, so an
+        // item-count rule alone would discard it — but nothing about its
+        // geometry says it is invalid, so its bounds must survive.
+        let mut items = Vec::new();
+        items.extend(fill_zone(1, 0.0, 12_000.0, 750.0, 500.0));
+        for i in 0..12 {
+            items.push(make_item(1, 24_000.0, 750.0 - i as f32 * 14.0, "Sidebar"));
+        }
+
+        let cols = detect_columns(&items, 1, false);
+        let right_edge = cols
+            .iter()
+            .map(|c| c.x_max)
+            .fold(f32::NEG_INFINITY, f32::max);
+        assert!(
+            right_edge > 24_000.0,
+            "sidebar was trimmed away: right edge {right_edge}, expected >24_000"
+        );
+    }
+
+    #[test]
+    fn genuinely_wide_layout_keeps_its_true_bounds() {
+        // A large-format page whose content really is spread beyond one
+        // ordinary page must not be trimmed to the median cluster: its far
+        // items are the majority, not strays.
+        let mut items = Vec::new();
+        items.extend(fill_zone(1, 100.0, 20_000.0, 750.0, 600.0));
+        items.extend(fill_zone(1, 22_000.0, 40_000.0, 750.0, 600.0));
+
+        let cols = detect_columns(&items, 1, false);
+        let widest = cols
+            .iter()
+            .map(|c| c.x_max)
+            .fold(f32::NEG_INFINITY, f32::max);
+        assert!(
+            widest > 35_000.0,
+            "wide layout was trimmed: right edge {widest}, expected ~40_000"
+        );
+    }
+
+    #[test]
+    fn oversized_but_legal_page_is_not_trimmed() {
+        // A wide-format page well inside the 14_400pt spec limit must keep its
+        // real bounds — outlier trimming is only for spans beyond a legal page.
+        let mut items = Vec::new();
+        items.extend(fill_zone(1, 100.0, 4_000.0, 750.0, 400.0));
+        items.extend(fill_zone(1, 4_400.0, 8_000.0, 750.0, 400.0));
+
+        let cols = detect_columns(&items, 1, false);
+        assert_eq!(cols.len(), 2, "Expected 2 columns, got {}", cols.len());
+        assert!(
+            cols[1].x_max > 7_000.0,
+            "right column should keep its true extent, got {}",
+            cols[1].x_max
+        );
+    }
+
     #[test]
     fn two_column_regression_guard() {
         // Standard 2-column layout with clear gutter at center

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();
 
@@ -146,9 +186,12 @@ 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,
@@ -158,7 +201,19 @@ pub(crate) fn extract_form_fields(
     }
     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,
@@ -168,6 +223,8 @@ pub(crate) fn extract_form_fields(
                 page_map,
                 &annotation_pages,
                 &mut items,
+                &mut budget,
+                0,
             );
         }
     }
@@ -202,6 +259,7 @@ fn annotation_page_map(
 }
 
 /// 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,
@@ -210,7 +268,22 @@ pub(crate) fn walk_form_fields(
     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,
@@ -241,9 +314,19 @@ 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,
@@ -253,6 +336,8 @@ pub(crate) fn walk_form_fields(
                         page_map,
                         annotation_pages,
                         items,
+                        budget,
+                        depth + 1,
                     );
                 }
             }
@@ -411,4 +496,225 @@ mod tests {
         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/mod.rs

@@ -283,17 +283,20 @@ fn extract_positioned_text_impl(
             include_invisible,
             &mut style_cache,
         );
-        let ((mut items, mut rects, mut lines), has_gid_fonts, coords_rotated) = match page_result {
-            Ok(extraction) => extraction,
-            Err(error) if required_pages.is_some_and(|required| !required.contains(page_num)) => {
-                debug!(
-                    "page {}: skipping context-only extraction error: {}",
-                    page_num, error
-                );
-                continue;
-            }
-            Err(error) => return Err(error),
-        };
+        let ((mut items, mut rects, mut lines), has_gid_fonts, coords_rotated, _skipped_invisible) =
+            match page_result {
+                Ok(extraction) => extraction,
+                Err(error)
+                    if required_pages.is_some_and(|required| !required.contains(page_num)) =>
+                {
+                    debug!(
+                        "page {}: skipping context-only extraction error: {}",
+                        page_num, error
+                    );
+                    continue;
+                }
+                Err(error) => return Err(error),
+            };
         // Clip to the visible page box: single-page extracts and imposed
         // spreads keep neighboring pages' content in the stream, positioned
         // outside the CropBox. Extracting it interleaves invisible text into

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/lib.rs

@@ -491,7 +491,14 @@ pub fn extract_pages_markdown_mem(
 
     // Tables need the original numeric cells; columns use folio-cleaned
     // evidence so removed page numbers cannot create false layout metadata.
-    let complexity = compute_layout_complexity(&all_items, &filtered_items, &all_rects, &all_lines);
+    let chart_regions = markdown::chart_regions_by_page(&all_items, &all_rects, &all_lines);
+    let complexity = compute_layout_complexity_with_chart_regions(
+        &all_items,
+        &filtered_items,
+        &all_rects,
+        &all_lines,
+        &chart_regions,
+    );
 
     // Compute font stats from full document (cross-page consistency).
     let font_stats = markdown::analysis::calculate_font_stats_from_items(&filtered_items);
@@ -509,6 +516,7 @@ pub fn extract_pages_markdown_mem(
     let mut results = Vec::with_capacity(pages_slice.len());
     let mut pages_needing_ocr = Vec::new();
     let mut ocr_reasons_by_page = BTreeMap::new();
+    let lopdf_pages = doc.get_pages();
 
     for &page_0idx in pages_slice {
         // Out-of-range pages → empty + needs_ocr
@@ -542,6 +550,25 @@ pub fn extract_pages_markdown_mem(
         let has_gid = gid_pages.contains(&page_1idx);
         let has_text_quality_issue = text_quality.pages_needing_ocr.contains(&page_1idx);
 
+        // A page can extract cleanly (no decoding issues, non-empty text)
+        // while still being fundamentally a scan: a full-page raster with
+        // a little genuine native text drawn over it (a header, a stamp, a
+        // cover-sheet annotation). Text-quality signals alone can't see
+        // that — consult the same "large background image" signal
+        // classify_pdf/detect_pdf_type already uses, so the two APIs can't
+        // silently disagree on whether a page needs OCR. See #227.
+        // Also covers vector-outlined text (glyphs drawn as paths, not
+        // shown via a text-showing operator): a hybrid page with real
+        // embedded-font body text elsewhere would otherwise still extract
+        // non-empty, non-garbled markdown and miss OCR routing entirely.
+        // detect_from_document's Mixed-type per-page routing always sends
+        // these pages to OCR; mirror that here too. Both signals share one
+        // analyze_page_content pass — see page_ocr_signals's doc comment.
+        let (has_template_image, has_vector_text) = lopdf_pages
+            .get(&page_1idx)
+            .map(|&page_id| detector::page_ocr_signals(&doc, page_id))
+            .unwrap_or((false, false));
+
         // Build markdown with document-wide font stats
         let options = MarkdownOptions {
             base_font_size: Some(font_stats.most_common_size),
@@ -565,6 +592,7 @@ pub fn extract_pages_markdown_mem(
                     page_count,
                     prefiltered_page_number_pages: Some(&removed_page_number_pages),
                     prefiltered_page_number_mask: Some(&page_number_removal_mask),
+                    precomputed_chart_regions: Some(&chart_regions),
                 },
             )
         };
@@ -578,10 +606,20 @@ pub fn extract_pages_markdown_mem(
                 OCR_REASON_SUSPECTED_GARBLED_TEXT,
             );
         }
+        if has_template_image {
+            add_ocr_reason(&mut ocr_reasons_by_page, page_1idx, OCR_REASON_SCANNED);
+        }
+        if has_vector_text {
+            add_ocr_reason(&mut ocr_reasons_by_page, page_1idx, OCR_REASON_VECTOR_TEXT);
+        }
         let ocr_reason = page_ocr_reason(&ocr_reasons_by_page, page_1idx);
 
-        let needs_ocr =
-            ocr_reason.is_some() || md.trim().is_empty() || has_gid || is_garbage_text(&md);
+        let needs_ocr = ocr_reason.is_some()
+            || md.trim().is_empty()
+            || has_gid
+            || is_garbage_text(&md)
+            || has_template_image
+            || has_vector_text;
 
         if needs_ocr {
             pages_needing_ocr.push(page_1idx);
@@ -619,6 +657,80 @@ pub fn extract_pages_markdown<P: AsRef<Path>>(
     extract_pages_markdown_mem(&buffer, pages)
 }
 
+// =========================================================================
+// Structure-tree element extraction (tagged PDFs)
+// =========================================================================
+
+/// One structure-tree element reference from a tagged PDF, resolved to a
+/// page and Marked Content ID.
+///
+/// Join `(page, mcid)` against [`TextItem::page`] / [`TextItem::mcid`] from
+/// [`extract_text_with_positions`] to attach semantic roles (heading levels,
+/// paragraphs, table cells, …) to extracted text.
+#[derive(Debug, Clone)]
+pub struct StructureElement {
+    /// 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 `/RoleMap`; tags
+    /// with no standard mapping are returned verbatim.
+    pub role: String,
+}
+
+/// Extract structure-tree element references from a tagged PDF in memory.
+///
+/// Parses `/StructTreeRoot` (when present) and returns one entry per
+/// marked-content reference, resolved to its 1-indexed page, MCID, and
+/// structure type name. Returns an empty list when the PDF is not tagged.
+///
+/// Pass `Some(&[...])` with 1-indexed page numbers (matching
+/// [`TextItem::page`]) to restrict output to those pages; pass `None` for
+/// the whole document. Entries are sorted by `(page, mcid)`.
+pub fn extract_structure_elements_mem(
+    buffer: &[u8],
+    pages: Option<&[u32]>,
+) -> Result<Vec<StructureElement>, PdfError> {
+    validate_pdf_bytes(buffer)?;
+    let (doc, _page_count) = load_document_from_mem(buffer)?;
+    let Some(tree) = structure_tree::StructTree::from_doc(&doc) else {
+        return Ok(Vec::new());
+    };
+    let page_ids = doc.get_pages();
+    let roles = tree.mcid_to_roles(&page_ids);
+
+    let page_filter: Option<HashSet<u32>> = pages.map(|p| p.iter().copied().collect());
+    let mut elements: Vec<StructureElement> = roles
+        .into_iter()
+        .filter(|(page, _)| page_filter.as_ref().is_none_or(|f| f.contains(page)))
+        .flat_map(|(page, mcids)| {
+            mcids.into_iter().map(move |(mcid, role)| StructureElement {
+                page,
+                mcid,
+                role: role.name().to_string(),
+            })
+        })
+        .collect();
+    elements.sort_unstable_by_key(|e| (e.page, e.mcid));
+    Ok(elements)
+}
+
+/// Path-based wrapper for [`extract_structure_elements_mem`].
+///
+/// Reads the PDF from disk and extracts structure-tree element references.
+/// Pass `None` for `pages` to return the whole document, or `Some(&[...])`
+/// to restrict to specific 1-indexed pages.
+pub fn extract_structure_elements<P: AsRef<Path>>(
+    path: P,
+    pages: Option<&[u32]>,
+) -> Result<Vec<StructureElement>, PdfError> {
+    validate_pdf_file(&path)?;
+    let buffer = std::fs::read(path.as_ref())?;
+    extract_structure_elements_mem(&buffer, pages)
+}
+
 // =========================================================================
 // Region-based text extraction (for hybrid OCR pipelines)
 // =========================================================================
@@ -645,6 +757,23 @@ pub struct PageRegionResult {
     pub regions: Vec<RegionText>,
 }
 
+/// Minimum alphanumeric mass an invisible (Tr 3) text layer must carry for
+/// the OCR-layer fallback in [`extract_text_in_regions_mem`] to adopt it. A
+/// real OCR layer carries far more; a stray watermark or artifact does not.
+const OCR_LAYER_MIN_ALNUM: usize = 40;
+
+/// Alphanumeric mass of extracted items, ignoring raster placeholders.
+/// `[Image: ...]` items (ItemType::Image) are synthesized for image
+/// XObjects — they mark that pixels exist, not that text was read, so they
+/// must not count as coverage.
+fn non_placeholder_alnum(items: &[TextItem]) -> usize {
+    items
+        .iter()
+        .filter(|it| !matches!(it.item_type, types::ItemType::Image))
+        .map(|it| it.text.chars().filter(|c| c.is_alphanumeric()).count())
+        .sum()
+}
+
 /// Extract text within bounding-box regions from a PDF in memory.
 ///
 /// This is designed for hybrid OCR pipelines: a layout model detects regions
@@ -698,7 +827,7 @@ pub fn extract_text_in_regions_mem(
         page_heights.insert(*page_num, height);
 
         // Extract text items for this page
-        let ((mut items, _rects, _lines), has_gid, coords_rotated) =
+        let ((mut items, _rects, _lines), mut has_gid, mut coords_rotated, skipped_invisible) =
             extractor::content_stream::extract_page_text_items(
                 &doc,
                 page_id,
@@ -707,6 +836,51 @@ pub fn extract_text_in_regions_mem(
                 false,
                 &mut style_cache,
             )?;
+        // OCR-layer fallback: scanned pages often carry their text as an
+        // invisible (Tr 3) layer behind the page raster. The visible-only
+        // pass sees nothing there but `[Image: ...]` placeholders, so every
+        // region on the page reports needs_ocr even though the exact text is
+        // embedded in the PDF — and this extractor then disagrees with the
+        // markdown path, which already retries Mixed PDFs with the invisible
+        // layer included. Retry page-scoped, and only when (a) the first
+        // pass actually SKIPPED invisible text — blank pages and image-only
+        // scans without an OCR layer must not pay a second content-stream
+        // parse (review catch) — and (b) the page has NO visible text item
+        // at all (punctuation counts, whitespace-only artifacts don't): an
+        // invisible OCR layer transcribes the raster, so any visible glyph
+        // has an invisible twin there and adoption would duplicate it
+        // (review catches — strict gate, no fuzzy dedupe). Adopt the retry
+        // only when it contributes real, non-garbage text.
+        let has_visible_text = items.iter().any(|it| {
+            !matches!(it.item_type, types::ItemType::Image) && !it.text.trim().is_empty()
+        });
+        if skipped_invisible && !has_visible_text {
+            if let Ok(((inv_items, _inv_rects, _inv_lines), inv_gid, inv_rotated, _)) =
+                extractor::content_stream::extract_page_text_items(
+                    &doc,
+                    page_id,
+                    *page_num,
+                    &font_cmaps,
+                    true,
+                    &mut style_cache,
+                )
+            {
+                let inv_alnum = non_placeholder_alnum(&inv_items);
+                // Judge the WHOLE recovered layer, not a prefix — a broken
+                // OCR layer can hide its garbage past any fixed sample size
+                // (review catch).
+                let sample: String = inv_items
+                    .iter()
+                    .filter(|it| !matches!(it.item_type, types::ItemType::Image))
+                    .map(|it| it.text.as_str())
+                    .collect();
+                if inv_alnum >= OCR_LAYER_MIN_ALNUM && !is_garbage_text(&sample) {
+                    items = inv_items;
+                    has_gid = inv_gid;
+                    coords_rotated = inv_rotated;
+                }
+            }
+        }
         let threshold = text_utils::fix_letterspaced_items(&mut items);
         if threshold > 0.10 {
             page_thresholds.insert(*page_num, threshold);
@@ -863,7 +1037,7 @@ pub fn extract_tables_in_regions_mem(
         let height = get_page_height(&doc, page_id).unwrap_or(792.0);
         page_heights.insert(*page_num, height);
 
-        let ((mut items, rects, lines), has_gid, coords_rotated) =
+        let ((mut items, rects, lines), has_gid, coords_rotated, _skipped_invisible) =
             extractor::content_stream::extract_page_text_items(
                 &doc,
                 page_id,
@@ -1174,7 +1348,7 @@ pub fn detect_vector_grid_in_region_mem(
     let needed_pages = HashSet::from([page_1idx]);
     let font_cmaps = FontCMaps::from_doc_pages_fast(&doc, Some(&needed_pages));
     let page_h = get_page_height(&doc, page_id).unwrap_or(792.0);
-    let ((mut items, rects, lines), _has_gid, coords_rotated) =
+    let ((mut items, rects, lines), _has_gid, coords_rotated, _skipped_invisible) =
         extractor::content_stream::extract_page_text_items(
             &doc,
             page_id,
@@ -1368,15 +1542,16 @@ mod vector_grid_tests {
         let &page_id = pages.get(&1).unwrap();
         let needed: HashSet<u32> = HashSet::from([1]);
         let cmaps = FontCMaps::from_doc_pages_fast(&doc, Some(&needed));
-        let ((items, rects, _lines), _has_gid, _rotated) = extract_page_text_items(
-            &doc,
-            page_id,
-            1,
-            &cmaps,
-            false,
-            &mut crate::extractor::FontStyleCache::new(),
-        )
-        .unwrap();
+        let ((items, rects, _lines), _has_gid, _rotated, _skipped_invisible) =
+            extract_page_text_items(
+                &doc,
+                page_id,
+                1,
+                &cmaps,
+                false,
+                &mut crate::extractor::FontStyleCache::new(),
+            )
+            .unwrap();
 
         let (rect_tables, _) = detect_tables_from_rects(&items, &rects, 1);
         assert_eq!(rect_tables.len(), 1, "expected one rect-detected table");
@@ -1410,15 +1585,16 @@ mod vector_grid_tests {
         let &page_id = pages.get(&page_num).unwrap();
         let needed: HashSet<u32> = HashSet::from([page_num]);
         let cmaps = FontCMaps::from_doc_pages_fast(&doc, Some(&needed));
-        let ((items, rects, _lines), _has_gid, _rotated) = extract_page_text_items(
-            &doc,
-            page_id,
-            page_num,
-            &cmaps,
-            false,
-            &mut crate::extractor::FontStyleCache::new(),
-        )
-        .unwrap();
+        let ((items, rects, _lines), _has_gid, _rotated, _skipped_invisible) =
+            extract_page_text_items(
+                &doc,
+                page_id,
+                page_num,
+                &cmaps,
+                false,
+                &mut crate::extractor::FontStyleCache::new(),
+            )
+            .unwrap();
 
         let (rect_tables, _) = detect_tables_from_rects(&items, &rects, page_num);
         rect_tables
@@ -2146,7 +2322,7 @@ pub fn extract_tables_with_structure_cells_mem(
         let height = get_page_height(&doc, page_id).unwrap_or(792.0);
         page_heights.insert(*page_num, height);
 
-        let ((mut items, _rects, _lines), _has_gid, coords_rotated) =
+        let ((mut items, _rects, _lines), _has_gid, coords_rotated, _skipped_invisible) =
             extractor::content_stream::extract_page_text_items(
                 &doc,
                 page_id,
@@ -2948,7 +3124,7 @@ fn detect_tsr_quality_issue(
     let mut needed: HashSet<u32> = HashSet::new();
     needed.insert(page_1idx);
     let font_cmaps = FontCMaps::from_doc_pages_fast(&doc, Some(&needed));
-    let ((mut items, _rects, _lines), _has_gid, coords_rotated) =
+    let ((mut items, _rects, _lines), _has_gid, coords_rotated, _skipped_invisible) =
         extractor::content_stream::extract_page_text_items(
             &doc,
             page_id,
@@ -3515,6 +3691,7 @@ fn repair_pdf_container_candidates(buf: &[u8]) -> Vec<Vec<u8>> {
     let mut candidates = Vec::new();
 
     add_repair_candidate(&mut candidates, append_missing_eof_marker(buf), buf);
+    add_repair_candidate(&mut candidates, recover_startxref_pointer(buf), buf);
 
     let stripped = strip_leading_pdf_container_bytes(buf);
     if let Some(stripped_buf) = stripped.as_deref() {
@@ -3524,11 +3701,112 @@ fn repair_pdf_container_candidates(buf: &[u8]) -> Vec<Vec<u8>> {
             append_missing_eof_marker(stripped_buf),
             buf,
         );
+        add_repair_candidate(
+            &mut candidates,
+            recover_startxref_pointer(stripped_buf),
+            buf,
+        );
     }
 
     candidates
 }
 
+/// Some PDF writers emit a `startxref` pointer that doesn't actually point
+/// at the cross-reference table — a single corrupted byte in the offset is
+/// enough. lopdf trusts that pointer outright and fails to load rather than
+/// searching for the real table, unlike pypdf/pdfium which both recover by
+/// locating it directly. This finds the real (classic, non-stream) `xref`
+/// table by scanning for the keyword — validating that a plausible
+/// subsection header follows, not just any standalone "xref" token, since
+/// this crate processes untrusted input and a coincidental match inside
+/// unrelated stream/string content must not get "repaired" against a bogus
+/// offset (lopdf would then load successfully against garbage instead of
+/// returning a clean error) — and appends a corrected trailing
+/// `startxref`/`%%EOF` block. lopdf's own `get_xref_start` always uses the
+/// *last* `%%EOF` in the final 512 bytes of the buffer, so ours
+/// transparently supersedes the broken one without needing to touch
+/// anything already in the file.
+///
+/// Doesn't cover cross-reference *streams* (`N 0 obj << /Type /XRef ...`,
+/// used by some PDF 1.5+ writers instead of a classic table) — recovering
+/// those needs the containing object's number, not just a byte offset.
+fn recover_startxref_pointer(buf: &[u8]) -> Option<Vec<u8>> {
+    let xref_pos = find_last_valid_xref_table_start(buf)?;
+
+    let mut repaired = Vec::with_capacity(buf.len() + 32);
+    repaired.extend_from_slice(buf);
+    if !repaired.ends_with(b"\n") {
+        repaired.push(b'\n');
+    }
+    repaired.extend_from_slice(format!("startxref\n{xref_pos}\n%%EOF\n").as_bytes());
+    Some(repaired)
+}
+
+/// Finds the last standalone `xref` token in `buf` that is immediately
+/// followed by a plausible classic cross-reference subsection header
+/// (`<start-id> <count>`, e.g. "0 6") — the shape every real classic xref
+/// table starts with. A single reverse byte scan: O(n) even on a
+/// pathological buffer with many non-matching or non-standalone "xref"
+/// occurrences, unlike repeatedly re-searching a shrinking prefix.
+fn find_last_valid_xref_table_start(buf: &[u8]) -> Option<usize> {
+    const KEYWORD: &[u8] = b"xref";
+    if buf.len() < KEYWORD.len() {
+        return None;
+    }
+    let mut pos = buf.len() - KEYWORD.len();
+    loop {
+        if &buf[pos..pos + KEYWORD.len()] == KEYWORD {
+            let before_ok = pos == 0 || buf[pos - 1].is_ascii_whitespace();
+            let after_ok = buf
+                .get(pos + KEYWORD.len())
+                .is_none_or(|c| c.is_ascii_whitespace());
+            if before_ok && after_ok && looks_like_xref_subsection_header(buf, pos + KEYWORD.len())
+            {
+                return Some(pos);
+            }
+        }
+        if pos == 0 {
+            return None;
+        }
+        pos -= 1;
+    }
+}
+
+/// Checks that `buf[pos..]` starts (after whitespace) with two
+/// whitespace-separated runs of ASCII digits — `<start-id> <count>`, the
+/// first subsection header of a classic PDF cross-reference table.
+fn looks_like_xref_subsection_header(buf: &[u8], pos: usize) -> bool {
+    fn skip_ws(buf: &[u8], mut pos: usize) -> usize {
+        while buf.get(pos).is_some_and(u8::is_ascii_whitespace) {
+            pos += 1;
+        }
+        pos
+    }
+    fn skip_digits(buf: &[u8], mut pos: usize) -> usize {
+        while buf.get(pos).is_some_and(u8::is_ascii_digit) {
+            pos += 1;
+        }
+        pos
+    }
+
+    let pos = skip_ws(buf, pos);
+    let after_first_digits = skip_digits(buf, pos);
+    if after_first_digits == pos {
+        return false; // no start-id
+    }
+    let sep = skip_ws(buf, after_first_digits);
+    if sep == after_first_digits {
+        return false; // start-id and count must be whitespace-separated
+    }
+    let after_count = skip_digits(buf, sep);
+    if after_count == sep {
+        return false; // no count
+    }
+    // The count run must end at whitespace/buffer-end, not run into trailing
+    // garbage (e.g. a coincidental "xref\n0 6garbage" in stream content).
+    buf.get(after_count).is_none_or(u8::is_ascii_whitespace)
+}
+
 fn add_repair_candidate(
     candidates: &mut Vec<Vec<u8>>,
     candidate: Option<Vec<u8>>,
@@ -3816,7 +4094,14 @@ fn process_document(
 
             let text_quality = analyze_text_quality(&items);
             merge_ocr_reasons(&mut ocr_reasons_by_page, text_quality.reasons_by_page);
-            let layout = compute_layout_complexity(&items, &layout_items, &rects, &lines);
+            let chart_regions = markdown::chart_regions_by_page(&items, &rects, &lines);
+            let layout = compute_layout_complexity_with_chart_regions(
+                &items,
+                &layout_items,
+                &rects,
+                &lines,
+                &chart_regions,
+            );
 
             let md = if options.mode == ProcessMode::Analyze {
                 None
@@ -3833,6 +4118,7 @@ fn process_document(
                         page_count,
                         prefiltered_page_number_pages: Some(&removed_pages),
                         prefiltered_page_number_mask: Some(removal_mask.as_slice()),
+                        precomputed_chart_regions: Some(&chart_regions),
                     },
                 ))
             };
@@ -5633,11 +5919,29 @@ fn select_items_with_document_folio_context(
 }
 
 /// Analyse extracted items and rects for layout complexity.
+#[cfg(test)]
 fn compute_layout_complexity(
     items: &[types::TextItem],
     column_items: &[types::TextItem],
     rects: &[types::PdfRect],
     lines: &[types::PdfLine],
+) -> LayoutComplexity {
+    let page_chart_regions = markdown::chart_regions_by_page(items, rects, lines);
+    compute_layout_complexity_with_chart_regions(
+        items,
+        column_items,
+        rects,
+        lines,
+        &page_chart_regions,
+    )
+}
+
+fn compute_layout_complexity_with_chart_regions(
+    items: &[types::TextItem],
+    column_items: &[types::TextItem],
+    rects: &[types::PdfRect],
+    lines: &[types::PdfLine],
+    page_chart_regions: &markdown::PageChartRegions,
 ) -> LayoutComplexity {
     use markdown::analysis::calculate_font_stats_from_items;
 
@@ -5659,6 +5963,10 @@ fn compute_layout_complexity(
         let owned_items: Vec<types::TextItem> = page_items.iter().map(|i| (*i).clone()).collect();
         let page_content_width = tables::content_width(&owned_items);
         let bands = markdown::split_side_by_side(&owned_items);
+        let chart_regions = page_chart_regions
+            .get(&page)
+            .map(Vec::as_slice)
+            .unwrap_or_default();
 
         let band_ranges: Vec<(f32, f32)> = if bands.is_empty() {
             // Single region — use sentinel range that includes everything
@@ -5673,7 +5981,8 @@ fn compute_layout_complexity(
             let band_items: Vec<types::TextItem> = owned_items
                 .iter()
                 .filter(|item| {
-                    x_lo == f32::MIN || (item.x >= x_lo - margin && item.x < x_hi + margin)
+                    (x_lo == f32::MIN || (item.x >= x_lo - margin && item.x < x_hi + margin))
+                        && !markdown::item_is_in_chart_region(item, chart_regions)
                 })
                 .cloned()
                 .collect();
@@ -5725,8 +6034,20 @@ fn compute_layout_complexity(
     }
 
     let mut pages_with_columns: Vec<u32> = Vec::new();
-    for page in seen_pages {
-        let cols = extractor::detect_columns(column_items, page, pages_with_tables.contains(&page));
+    for &page in &seen_pages {
+        let chart_regions = page_chart_regions
+            .get(&page)
+            .map(Vec::as_slice)
+            .unwrap_or_default();
+        let page_column_items: Vec<types::TextItem> = column_items
+            .iter()
+            .filter(|item| {
+                item.page == page && !markdown::item_is_in_chart_region(item, chart_regions)
+            })
+            .cloned()
+            .collect();
+        let cols =
+            extractor::detect_columns(&page_column_items, page, pages_with_tables.contains(&page));
         if cols.len() >= 2 {
             pages_with_columns.push(page);
         }
@@ -5955,6 +6276,66 @@ mod tests {
         assert!(filtered.pages_with_columns.is_empty());
     }
 
+    #[test]
+    fn dense_chart_panel_is_not_reported_as_a_table() {
+        let mut items: Vec<TextItem> = (0..8)
+            .flat_map(|row| {
+                (0..6).map(move |column| {
+                    test_item(
+                        &format!("{}", row * 10 + column),
+                        105.0 + column as f32 * 35.0,
+                        525.0 - row as f32 * 15.0,
+                        24.0,
+                        10.0,
+                    )
+                })
+            })
+            .collect();
+        for row in 0..6 {
+            items.push(test_item(
+                "Left column prose continues here",
+                80.0,
+                320.0 - row as f32 * 15.0,
+                160.0,
+                10.0,
+            ));
+            items.push(test_item(
+                "Right column prose continues here",
+                300.0,
+                320.0 - row as f32 * 15.0,
+                160.0,
+                10.0,
+            ));
+        }
+        let mut lines: Vec<PdfLine> = (0..30)
+            .map(|column| PdfLine {
+                x1: 100.0 + column as f32 * 8.0,
+                y1: 400.0,
+                x2: 100.0 + column as f32 * 8.0,
+                y2: 550.0,
+                page: 1,
+            })
+            .collect();
+        lines.extend((0..6).map(|row| PdfLine {
+            x1: 100.0,
+            y1: 400.0 + row as f32 * 30.0,
+            x2: 332.0,
+            y2: 400.0 + row as f32 * 30.0,
+            page: 1,
+        }));
+        let rects = vec![PdfRect {
+            x: 80.0,
+            y: 350.0,
+            width: 280.0,
+            height: 240.0,
+            page: 1,
+        }];
+
+        let complexity = compute_layout_complexity(&items, &items, &rects, &lines);
+
+        assert!(complexity.pages_with_tables.is_empty());
+    }
+
     #[test]
     fn page_selection_keeps_document_wide_folio_layout_decisions() {
         let mut items = Vec::new();
@@ -6958,4 +7339,65 @@ mod tests {
         // Pre-filled cell was not touched.
         assert_eq!(cells[1].text, "Pre-filled");
     }
+
+    // -- recover_startxref_pointer / find_last_valid_xref_table_start ------
+    //
+    // Direct unit tests on the byte-level scan, addressing review feedback
+    // on #230: a coincidental standalone "xref" token that isn't actually
+    // followed by a subsection header (start-id + count) must not be
+    // treated as a real table — accepting it would let lopdf "succeed"
+    // against a bogus offset and silently return garbled/empty content
+    // instead of a clean error.
+
+    #[test]
+    fn find_xref_rejects_standalone_token_without_subsection_header() {
+        // "xref" appears as a real standalone word, but nothing that looks
+        // like "<start-id> <count>" follows it.
+        let buf = b"Please refer to the xref appendix for details.";
+        assert_eq!(find_last_valid_xref_table_start(buf), None);
+    }
+
+    #[test]
+    fn find_xref_accepts_real_classic_table_header() {
+        let buf = b"garbage\nxref\n0 6\n0000000000 65535 f \n%%EOF";
+        let pos = find_last_valid_xref_table_start(buf).expect("should find the real table");
+        assert_eq!(&buf[pos..pos + 4], b"xref");
+        assert_eq!(&buf[pos..], b"xref\n0 6\n0000000000 65535 f \n%%EOF");
+    }
+
+    #[test]
+    fn find_xref_skips_coincidental_match_and_finds_real_table_before_it() {
+        // A coincidental "xref" (no subsection header) appears *after* the
+        // real table in the buffer — the scan must not stop at the first
+        // (rightmost) standalone token it finds; it must keep looking
+        // backward until one actually validates.
+        let buf = b"xref\n0 3\n0000000000 65535 f \ntrailer\nsee the xref\n";
+        let pos = find_last_valid_xref_table_start(buf).expect("should find the real table");
+        assert_eq!(pos, 0);
+    }
+
+    #[test]
+    fn find_xref_rejects_substring_of_startxref() {
+        // "xref" is a substring of "startxref" but isn't a standalone
+        // token there (not preceded by whitespace) — must not match, even
+        // though a number immediately follows it.
+        let buf = b"startxref\n1234\n%%EOF";
+        assert_eq!(find_last_valid_xref_table_start(buf), None);
+    }
+
+    #[test]
+    fn find_xref_rejects_count_run_with_trailing_garbage() {
+        // "xref\n0 6garbage" has the right shape (digits, whitespace,
+        // digits) but the count run doesn't end at whitespace/EOF — it
+        // runs straight into non-digit garbage, so this must not be
+        // accepted as a real subsection header.
+        let buf = b"xref\n0 6garbage\n%%EOF";
+        assert_eq!(find_last_valid_xref_table_start(buf), None);
+    }
+
+    #[test]
+    fn recover_startxref_pointer_returns_none_without_a_valid_table() {
+        let buf = b"Please refer to the xref appendix for details.";
+        assert!(recover_startxref_pointer(buf).is_none());
+    }
 }

src/markdown/analysis.rs

@@ -171,6 +171,74 @@ pub(crate) fn is_toc_marker_heading(text: &str) -> bool {
 /// 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();
 
@@ -244,9 +312,133 @@ pub(crate) fn is_heading_fragment(text: &str) -> bool {
     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

src/markdown/heading.rs

@@ -127,7 +127,9 @@ fn visual_style(line: &TextLine) -> Option<VisualStyle> {
     })
 }
 
-fn roman_value(token: &str) -> Option<u32> {
+/// Shared with `analysis::starts_with_numbering_prefix` so the veto
+/// exemption and the heading parser agree on what a roman numeral is.
+pub(super) fn roman_value(token: &str) -> Option<u32> {
     if token.is_empty() || token.len() > 8 {
         return None;
     }

src/markdown/mod.rs

@@ -69,7 +69,7 @@ fn is_chart_adjacent_label(item: &TextItem, region: (f32, f32, f32, f32)) -> boo
             || (mostly_inside_chart_width && close_to_chart_edge && category_sized))
 }
 
-fn item_is_in_chart_region(item: &TextItem, regions: &[(f32, f32, f32, f32)]) -> bool {
+pub(crate) fn item_is_in_chart_region(item: &TextItem, regions: &[(f32, f32, f32, f32)]) -> bool {
     regions.iter().any(|&(x0, y0, x1, y1)| {
         let cx = item.x + item.width / 2.0;
         let within_padded_x = cx >= x0 - CHART_REGION_PAD && cx <= x1 + CHART_REGION_PAD;
@@ -92,6 +92,72 @@ fn items_outside_chart_regions(
         .collect()
 }
 
+pub(crate) fn merge_chart_regions(
+    regions: impl IntoIterator<Item = (f32, f32, f32, f32)>,
+) -> Vec<(f32, f32, f32, f32)> {
+    const MERGE_TOLERANCE: f32 = 3.0;
+
+    let mut merged: Vec<(f32, f32, f32, f32)> = Vec::new();
+    for (x0, y0, x1, y1) in regions {
+        let mut current = (x0.min(x1), y0.min(y1), x0.max(x1), y0.max(y1));
+        let mut index = 0;
+        while index < merged.len() {
+            let candidate = merged[index];
+            let overlaps = current.2 + MERGE_TOLERANCE >= candidate.0
+                && candidate.2 + MERGE_TOLERANCE >= current.0
+                && current.3 + MERGE_TOLERANCE >= candidate.1
+                && candidate.3 + MERGE_TOLERANCE >= current.1;
+            if overlaps {
+                current = (
+                    current.0.min(candidate.0),
+                    current.1.min(candidate.1),
+                    current.2.max(candidate.2),
+                    current.3.max(candidate.3),
+                );
+                merged.swap_remove(index);
+            } else {
+                index += 1;
+            }
+        }
+        merged.push(current);
+    }
+    merged
+}
+
+pub(crate) type PageChartRegions = HashMap<u32, Vec<(f32, f32, f32, f32)>>;
+
+/// Compute the chart masks used by both layout analysis and Markdown output.
+///
+/// Keeping the rect-backed and dense-line heuristics behind one entry point
+/// ensures metadata and extraction cannot drift when either detector changes.
+pub(crate) fn chart_regions_by_page(
+    items: &[TextItem],
+    rects: &[PdfRect],
+    lines: &[PdfLine],
+) -> PageChartRegions {
+    let mut page_items: HashMap<u32, Vec<TextItem>> = HashMap::new();
+    for item in items.iter().filter(|item| {
+        matches!(
+            &item.item_type,
+            crate::types::ItemType::Text | crate::types::ItemType::FormField
+        )
+    }) {
+        page_items.entry(item.page).or_default().push(item.clone());
+    }
+
+    page_items
+        .into_iter()
+        .filter_map(|(page, items)| {
+            let rect_regions = crate::tables::detect_chart_regions(&items, rects, page);
+            let line_regions = crate::tables::detect_dense_line_chart_regions(lines, rects, page)
+                .into_iter()
+                .filter(|&region| chart_region_separates_prose_columns(&items, region));
+            let regions = merge_chart_regions(rect_regions.into_iter().chain(line_regions));
+            (!regions.is_empty()).then_some((page, regions))
+        })
+        .collect()
+}
+
 /// Detect side-by-side table layout by finding a significant X-position gap.
 ///
 /// Returns X-band boundaries `[(x_min, split_x), (split_x, x_max)]` when a
@@ -329,6 +395,15 @@ fn chart_spans_prose_split(region: (f32, f32, f32, f32), split_x: f32) -> bool {
     split_x - left >= MIN_CHART_WIDTH_PER_SIDE && right - split_x >= MIN_CHART_WIDTH_PER_SIDE
 }
 
+pub(crate) fn chart_region_separates_prose_columns(
+    items: &[TextItem],
+    region: (f32, f32, f32, f32),
+) -> bool {
+    let outside = items_outside_chart_regions(items, &[region]);
+    chart_page_prose_column_split(&outside)
+        .is_some_and(|split_x| chart_spans_prose_split(region, split_x))
+}
+
 /// True when adjacent physical rows form an unterminated, lowercase prose
 /// continuation in the same projected column.
 fn is_cross_row_prose_continuation(previous: &str, current: &str) -> bool {
@@ -1004,6 +1079,7 @@ pub fn to_markdown_from_items_with_rects_and_page_count(
             page_count: document_page_count,
             prefiltered_page_number_pages: None,
             prefiltered_page_number_mask: None,
+            precomputed_chart_regions: None,
         },
     )
 }
@@ -1021,6 +1097,9 @@ pub(crate) struct MarkdownDocumentContext<'a> {
     /// Table detection consumes the original items; the mask is applied only
     /// after table claims have been established.
     pub(crate) prefiltered_page_number_mask: Option<&'a [bool]>,
+    /// Optional chart masks shared with layout analysis so the geometry is
+    /// detected once and interpreted identically by both pipelines.
+    pub(crate) precomputed_chart_regions: Option<&'a PageChartRegions>,
 }
 
 /// Convert positioned text items to markdown, using rectangles and line segments for table detection.
@@ -1047,6 +1126,7 @@ pub(crate) fn to_markdown_from_items_with_rects_and_lines(
         page_count: document_page_count,
         prefiltered_page_number_pages,
         prefiltered_page_number_mask,
+        precomputed_chart_regions,
     } = context;
 
     if items.is_empty() {
@@ -1119,17 +1199,9 @@ pub(crate) fn to_markdown_from_items_with_rects_and_lines(
 
     // Chart regions per page: their text must not steer column detection
     // during line grouping (it fills the gutter and fuses two-column lines).
-    let mut page_chart_map: HashMap<u32, Vec<(f32, f32, f32, f32)>> = HashMap::new();
-    for &page in page_groups.keys() {
-        let page_items_ref: Vec<TextItem> = page_groups[&page]
-            .iter()
-            .map(|(_, item)| (*item).clone())
-            .collect();
-        let regions = crate::tables::detect_chart_regions(&page_items_ref, rects, page);
-        if !regions.is_empty() {
-            page_chart_map.insert(page, regions);
-        }
-    }
+    let page_chart_map = precomputed_chart_regions
+        .cloned()
+        .unwrap_or_else(|| chart_regions_by_page(&text_items, rects, pdf_lines));
 
     let mut pages: Vec<u32> = page_groups.keys().copied().collect();
     pages.sort();
@@ -2051,6 +2123,7 @@ mod tests {
                 page_count: 1,
                 prefiltered_page_number_pages: Some(&removed_pages),
                 prefiltered_page_number_mask: Some(&removal_mask),
+                precomputed_chart_regions: None,
             },
         );
 

src/markdown/postprocess.rs

@@ -464,6 +464,7 @@ 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]
@@ -507,6 +508,14 @@ mod tests {
         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/python.rs

@@ -271,6 +271,12 @@ pub struct PyTextItem {
     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]
@@ -286,6 +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
 // ---------------------------------------------------------------------------
@@ -356,6 +388,18 @@ fn convert_text_items(items: Vec<crate::TextItem>) -> Vec<PyTextItem> {
             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()
 }
@@ -613,6 +657,48 @@ 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<()> {
@@ -620,6 +706,7 @@ fn pdf_inspector(m: &Bound<'_, PyModule>) -> PyResult<()> {
     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>()?;
@@ -634,6 +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_heuristic.rs

@@ -435,6 +435,72 @@ fn revised_table_cell_indices(
         .collect()
 }
 
+/// Index of candidate "body" items (larger-font attachment targets) sorted by
+/// Y, so script-attachment checks scan a narrow Y window instead of the whole
+/// page per candidate.
+struct ScriptBodyIndex<'a> {
+    /// (y, item), sorted ascending by y
+    by_y: Vec<(f32, &'a TextItem)>,
+    /// widest vertical attachment window any body item can produce
+    max_window: f32,
+}
+
+impl<'a> ScriptBodyIndex<'a> {
+    fn new(items: &'a [TextItem]) -> Self {
+        // Smallest table-candidate font is 6pt, so any possible attachment
+        // target is at least 6 x 1.2 pt.
+        let mut by_y: Vec<(f32, &TextItem)> = items
+            .iter()
+            .filter(|i| i.font_size >= 6.0 * 1.2)
+            .map(|i| (i.y, i))
+            .collect();
+        by_y.sort_by(|a, b| a.0.total_cmp(&b.0));
+        let max_window = by_y
+            .iter()
+            .map(|(_, i)| i.font_size * 0.8)
+            .fold(0.0f32, f32::max);
+        Self { by_y, max_window }
+    }
+
+    /// True when a small-font item is horizontally attached to a larger-font
+    /// item at a script baseline offset — a sub/superscript in running text
+    /// or math (equation subscripts, footnote markers). Script attachments
+    /// are not table cells; without this filter, display equations with
+    /// sub/superscripts form phantom small-font table regions (e.g. TeX
+    /// papers where log subscripts cluster with footnote lines into a fake
+    /// 3-column table). A genuine baseline offset is required so same-line
+    /// table neighbours (a small cell beside a larger label cell) are never
+    /// classified as scripts.
+    ///
+    /// `min_anchor_size` additionally constrains what counts as an
+    /// attachment target: the small-font pass accepts any sufficiently
+    /// larger item (0.0), while the body-font pass requires a heading-sized
+    /// anchor so a body-size table cell beside a slightly larger label with
+    /// baseline jitter is never treated as a script.
+    fn is_script_attachment(&self, small: &TextItem, min_anchor_size: f32) -> bool {
+        let attach_gap = small.font_size.max(4.0) * 0.6;
+        let lo = self
+            .by_y
+            .partition_point(|(y, _)| *y < small.y - self.max_window);
+        self.by_y[lo..]
+            .iter()
+            .take_while(|(y, _)| *y <= small.y + self.max_window)
+            .any(|(_, body)| {
+                let dy = (small.y - body.y).abs();
+                body.font_size >= small.font_size * 1.2
+                    && body.font_size >= min_anchor_size
+                    && dy > body.font_size * 0.05
+                    && dy <= body.font_size * 0.8
+                    && {
+                        let gap_after_body = small.x - (body.x + body.width);
+                        let gap_before_body = body.x - (small.x + small.width);
+                        (-attach_gap..=attach_gap).contains(&gap_after_body)
+                            || (-attach_gap..=attach_gap).contains(&gap_before_body)
+                    }
+            })
+    }
+}
+
 /// Detect tables in a set of text items from a single page
 pub fn detect_tables(items: &[TextItem], base_font_size: f32, skip_body_font: bool) -> Vec<Table> {
     detect_tables_with_page_width(items, base_font_size, skip_body_font, content_width(items))
@@ -483,6 +549,27 @@ pub(crate) fn detect_tables_with_page_width(
     // === Pass 1: Small-font tables (existing behavior) ===
     let table_font_threshold = base_font_size * 0.90;
 
+    // Mark sub/superscript attachments once per pass. They stay candidates —
+    // the masks only remove them from region qualification and column/row
+    // geometry.
+    //
+    // The two passes need different anchor thresholds. In the small-font pass
+    // any sufficiently larger neighbour is a plausible base for a script. In
+    // the body-font pass the candidates are themselves body-sized
+    // (0.85..1.05x), so a merely "slightly larger" neighbour is usually a bold
+    // label or an adjacent column header, not the base of a superscript —
+    // treating it as one would strip real cells out of the geometry and lose
+    // the table. Requiring a heading-sized anchor (>= 1.15x base) keeps the
+    // body pass to genuine scripts hanging off headings.
+    let script_index = ScriptBodyIndex::new(items);
+    let script_flags: Vec<bool> = items
+        .iter()
+        .map(|item| script_index.is_script_attachment(item, 0.0))
+        .collect();
+    let body_script_flags: Vec<bool> = items
+        .iter()
+        .map(|item| script_index.is_script_attachment(item, base_font_size * 1.15))
+        .collect();
     let table_candidates: Vec<(usize, &TextItem)> = items
         .iter()
         .enumerate()
@@ -494,7 +581,14 @@ pub(crate) fn detect_tables_with_page_width(
         .collect();
 
     if table_candidates.len() >= 6 {
-        let regions = find_table_regions(&table_candidates);
+        // Qualify regions from non-script items: a cluster of sub/superscripts
+        // must not, on its own, mark out a table region.
+        let region_evidence: Vec<(usize, &TextItem)> = table_candidates
+            .iter()
+            .filter(|(idx, _)| !script_flags[*idx])
+            .cloned()
+            .collect();
+        let regions = find_table_regions(&region_evidence);
 
         for (y_min, y_max) in regions {
             let region_items: Vec<(usize, &TextItem)> = table_candidates
@@ -508,7 +602,9 @@ pub(crate) fn detect_tables_with_page_width(
             }
 
             if let Some(mut table) =
-                detect_table_in_region(&region_items, TableDetectionMode::SmallFont)
+                detect_table_in_region(&region_items, TableDetectionMode::SmallFont, &|i| {
+                    script_flags[i]
+                })
             {
                 // Try to recover body-font header row above the small-font table
                 recover_header_row(&mut table, items, table_font_threshold);
@@ -553,8 +649,20 @@ pub(crate) fn detect_tables_with_page_width(
             body_font_low,
             body_font_high,
         );
+        // Scripts are NOT filtered out of the candidate set here, mirroring
+        // the small-font pass: they must stay eligible for cell assignment so
+        // a sub/superscript that belongs inside a table cell keeps its text.
+        // The heading-anchored `body_script_flags` mask removes them from
+        // geometry only.
         if body_candidates.len() >= 6 {
-            let regions = find_table_regions_strict(&body_candidates);
+            // Same reasoning as the small-font pass: scripts do not qualify
+            // regions, but remain available for cell assignment within one.
+            let region_evidence: Vec<(usize, &TextItem)> = body_candidates
+                .iter()
+                .filter(|(idx, _)| !body_script_flags[*idx])
+                .cloned()
+                .collect();
+            let regions = find_table_regions_strict(&region_evidence);
             log::debug!("body-font: {} strict regions found", regions.len());
 
             for (y_min, y_max, _x_min, _x_max) in &regions {
@@ -580,7 +688,9 @@ pub(crate) fn detect_tables_with_page_width(
                 }
 
                 if let Some(table) =
-                    detect_table_in_region(&region_items, TableDetectionMode::BodyFont)
+                    detect_table_in_region(&region_items, TableDetectionMode::BodyFont, &|i| {
+                        body_script_flags[i]
+                    })
                 {
                     tables.push(table);
                 }
@@ -808,10 +918,30 @@ fn find_table_regions_strict(items: &[(usize, &TextItem)]) -> Vec<(f32, f32, f32
     regions
 }
 
-/// Detect a table within a specific region
-fn detect_table_in_region(items: &[(usize, &TextItem)], mode: TableDetectionMode) -> Option<Table> {
-    // Find column boundaries
-    let columns = find_column_boundaries(items, mode);
+/// Detect a table within a specific region.
+///
+/// `is_script` marks items that are sub/superscript attachments. Those are
+/// excluded from the *geometry* — they must not be able to create a column,
+/// which is how equation subscript clusters used to fabricate phantom grids —
+/// but they remain eligible for cell assignment, so legitimate cell content
+/// (exponents in an engineering-notation table, footnote markers) stays in
+/// the cell it belongs to instead of leaking out into the reading order.
+fn detect_table_in_region(
+    items: &[(usize, &TextItem)],
+    mode: TableDetectionMode,
+    is_script: &dyn Fn(usize) -> bool,
+) -> Option<Table> {
+    // Column geometry from non-script items only.
+    let geometry_items: Vec<(usize, &TextItem)> = items
+        .iter()
+        .filter(|(idx, _)| !is_script(*idx))
+        .cloned()
+        .collect();
+    // A region that is *entirely* scripts has no table structure at all.
+    if geometry_items.is_empty() {
+        return None;
+    }
+    let columns = find_column_boundaries(&geometry_items, mode);
     let min_cols = 2;
     if columns.len() < min_cols || columns.len() > 25 {
         log::debug!(
@@ -822,8 +952,8 @@ fn detect_table_in_region(items: &[(usize, &TextItem)], mode: TableDetectionMode
         return None;
     }
 
-    // Find row boundaries
-    let rows = find_row_boundaries(items);
+    // Find row boundaries (geometry items only, same reasoning)
+    let rows = find_row_boundaries(&geometry_items);
     let min_rows = 2;
     if rows.len() < min_rows {
         log::debug!(
@@ -842,6 +972,11 @@ fn detect_table_in_region(items: &[(usize, &TextItem)], mode: TableDetectionMode
     );
 
     // Verify this looks like a table: multiple items should align to columns
+    // Validate against ALL items, including scripts. Columns are derived from
+    // non-script geometry so scripts cannot *create* a column, but excluding
+    // them from validation too would let a region manufacture alignment: drop
+    // the awkward items and whatever remains looks like a tidy grid. Block
+    // diagrams did exactly that. Everything in the region must fit.
     let col_alignment = check_column_alignment(items, &columns, mode);
     let min_alignment = match mode {
         TableDetectionMode::SmallFont => 0.5,
@@ -912,6 +1047,29 @@ fn detect_table_in_region(items: &[(usize, &TextItem)], mode: TableDetectionMode
         cells.push(row_cells);
     }
 
+    // Validation 0 (small-font pass only): reject tiny all-numeric
+    // fragments. A <=2-row grid whose every cell is a bare 1-2 digit number
+    // carries no tabular information — in practice these are
+    // exponent/subscript clusters from display math that happen to align.
+    // Body-font tables are not subject to this veto: their cells cannot be
+    // script glyphs.
+    if matches!(mode, TableDetectionMode::SmallFont) {
+        let nonempty_cells: Vec<&String> =
+            cells.iter().flatten().filter(|c| !c.is_empty()).collect();
+        if rows.len() <= 2
+            && !nonempty_cells.is_empty()
+            && nonempty_cells
+                .iter()
+                .all(|c| c.len() <= 2 && c.chars().all(|ch| ch.is_ascii_digit()))
+        {
+            log::debug!(
+                "  validation 0 fail: tiny all-numeric fragment ({} cells)",
+                nonempty_cells.len()
+            );
+            return None;
+        }
+    }
+
     // Validation 1: some rows should have content in first column.
     // Use a lower threshold (25%) for tables with wrapped cells where
     // continuation lines leave the first column empty.
@@ -1977,6 +2135,146 @@ fn try_add_label_column(
 
 #[cfg(test)]
 mod tests {
+
+    fn make_item(text: &str, x: f32, y: f32, font_size: f32, width: f32) -> TextItem {
+        TextItem {
+            text: text.to_string(),
+            x,
+            y,
+            width,
+            height: font_size,
+            font: "TestFont".to_string(),
+            font_size,
+            page: 1,
+            is_bold: false,
+            is_italic: false,
+            is_underline: false,
+            is_strikeout: false,
+            item_type: ItemType::Text,
+            mcid: None,
+        }
+    }
+
+    #[test]
+    fn script_attachment_detects_subscript_after_body_text() {
+        let body = make_item("log", 100.0, 500.0, 10.0, 15.0);
+        let sub = make_item("10", 115.5, 497.0, 7.0, 7.0);
+        let items = vec![body, sub.clone()];
+        assert!(ScriptBodyIndex::new(&items).is_script_attachment(&sub, 0.0));
+    }
+
+    #[test]
+    fn script_attachment_detects_superscript_footnote_marker() {
+        let body = make_item("Hartley", 200.0, 500.0, 10.0, 35.0);
+        let sup = make_item("2", 235.8, 504.0, 6.6, 3.5);
+        let items = vec![body, sup.clone()];
+        assert!(ScriptBodyIndex::new(&items).is_script_attachment(&sup, 0.0));
+    }
+
+    #[test]
+    fn script_attachment_ignores_small_cell_far_from_body_text() {
+        let body = make_item("Revenue", 100.0, 500.0, 10.0, 40.0);
+        let cell = make_item("1,234", 180.0, 500.0, 7.0, 20.0);
+        let items = vec![body, cell.clone()];
+        assert!(!ScriptBodyIndex::new(&items).is_script_attachment(&cell, 0.0));
+    }
+
+    #[test]
+    fn body_pass_anchor_spares_cells_beside_slightly_larger_labels() {
+        // A body-font table cell (10pt) sitting beside a slightly larger,
+        // NON-heading label (12.5pt) with a little baseline jitter. The
+        // small-font pass treats any larger neighbour as a possible script
+        // base, but the body pass must not: at body sizes a slightly larger
+        // neighbour is a bold label or column header, and flagging the cell
+        // would strip it out of the table geometry and lose the table.
+        // Cell at the low end of the body band (0.85x base) beside a 10.5pt
+        // label. 10.5 clears the inherent 1.2x-of-cell rule (10.2) but falls
+        // below the body pass's heading anchor (11.5), which is exactly the
+        // band where the two masks must disagree.
+        let label = make_item("Revenue", 100.0, 500.0, 10.5, 40.0);
+        let cell = make_item("1,234", 141.0, 496.5, 8.5, 22.0);
+        let items = vec![label, cell.clone()];
+        let index = ScriptBodyIndex::new(&items);
+        let base = 10.0;
+        assert!(
+            index.is_script_attachment(&cell, 0.0),
+            "small-font pass anchor should still see this as an attachment"
+        );
+        assert!(
+            !index.is_script_attachment(&cell, base * 1.15),
+            "body pass must not treat a cell beside a slightly larger label \
+             as a script — that removes real cells from the geometry"
+        );
+        // A genuine heading-sized anchor still qualifies in the body pass.
+        let heading = make_item("Section", 100.0, 500.0, 20.0, 60.0);
+        let sup = make_item("3", 161.0, 508.0, 10.0, 5.0);
+        let h_items = vec![heading, sup.clone()];
+        assert!(
+            ScriptBodyIndex::new(&h_items).is_script_attachment(&sup, base * 1.15),
+            "script hanging off a heading must still be excluded in the body pass"
+        );
+    }
+
+    #[test]
+    fn script_attachment_ignores_same_baseline_neighbor_cell() {
+        // A small cell beside a larger label on the SAME baseline is a table
+        // layout, not a subscript — a genuine baseline offset is required.
+        let label = make_item("Total", 100.0, 500.0, 10.0, 25.0);
+        let cell = make_item("42", 127.0, 500.0, 7.5, 9.0);
+        let items = vec![label, cell.clone()];
+        assert!(!ScriptBodyIndex::new(&items).is_script_attachment(&cell, 0.0));
+    }
+
+    #[test]
+    fn script_attachment_ignores_neighbor_on_different_line() {
+        let body = make_item("Header", 100.0, 500.0, 10.0, 30.0);
+        let cell = make_item("42", 131.0, 486.0, 7.0, 10.0);
+        let items = vec![body, cell.clone()];
+        assert!(!ScriptBodyIndex::new(&items).is_script_attachment(&cell, 0.0));
+    }
+
+    /// Equation-subscript + footnote layout from Shannon entropy.pdf page 1,
+    /// with real coordinates. Without the larger-font anchors the small items
+    /// alone DO form a phantom table — proving the layout reaches detection —
+    /// and adding the anchors must suppress it.
+    fn shannon_page1_small_items() -> Vec<TextItem> {
+        vec![
+            make_item("2", 267.4, 133.9, 7.4, 3.7),
+            make_item("10", 306.2, 133.9, 7.4, 7.4),
+            make_item("10", 342.7, 133.9, 7.4, 7.4),
+            make_item("10", 325.0, 118.9, 7.4, 7.4),
+            make_item("Bell System Technical Journal,", 295.7, 101.9, 8.0, 95.0),
+            make_item(
+                "April 1924, p. 324; Certain Topics in",
+                396.7,
+                101.9,
+                8.0,
+                130.0,
+            ),
+            make_item("v. 47, April 1928, p. 617.", 250.9, 92.5, 8.0, 90.0),
+            make_item("Bell System Technical Journal,", 264.2, 82.6, 8.0, 95.0),
+            make_item("July 1928, p. 535.", 364.3, 82.6, 8.0, 65.0),
+        ]
+    }
+
+    #[test]
+    fn equation_scripts_do_not_form_phantom_table() {
+        let bare = shannon_page1_small_items();
+        assert!(
+            !detect_tables(&bare, 10.0, false).is_empty(),
+            "test layout must form a phantom table when the filter cannot fire"
+        );
+        let mut items = shannon_page1_small_items();
+        items.push(make_item("log", 253.0, 137.0, 10.0, 13.5));
+        items.push(make_item("log", 291.5, 137.0, 10.0, 13.5));
+        items.push(make_item("log", 328.0, 137.0, 10.0, 13.5));
+        items.push(make_item("log", 310.3, 122.0, 10.0, 13.5));
+        let tables = detect_tables(&items, 10.0, false);
+        assert!(
+            tables.is_empty(),
+            "equation scripts + footnotes must not become a table: {tables:?}"
+        );
+    }
     use super::*;
     use crate::types::ItemType;
 

src/tables/detect_lines.rs

@@ -4,10 +4,10 @@
 //! gridlines.  Many IRS forms and government PDFs use these instead of
 //! `re` (rectangle) operators.
 
-use std::collections::HashSet;
+use std::collections::{HashMap, HashSet};
 
 use crate::tables::Table;
-use crate::types::{PdfLine, TextItem};
+use crate::types::{PdfLine, PdfRect, TextItem};
 
 use super::detect_rects::{assign_items_to_grid, snap_edges};
 
@@ -15,11 +15,33 @@ const RULE_Y_TOLERANCE: f32 = 2.0;
 const RULE_JOIN_GAP: f32 = 6.0;
 const RULE_SPAN_TOLERANCE: f32 = 8.0;
 const TEXT_ROW_TOLERANCE: f32 = 2.5;
+const DENSE_CHART_MIN_VERTICAL_EDGES: usize = 27;
+const DENSE_CHART_LABEL_PAD: f32 = 20.0;
+const DENSE_CHART_MAX_SHARED_PANEL_GRIDS: usize = 4;
 
 type HorizontalRule = (f32, f32, f32); // (y, x_min, x_max)
 type VerticalRule = (f32, f32, f32); // (x, y_min, y_max)
 type AnchoredRow<'a> = (f32, Vec<(usize, &'a TextItem)>);
 
+fn dense_chart_grids_are_co_located(
+    left: (f32, f32, f32, f32),
+    right: (f32, f32, f32, f32),
+) -> bool {
+    let left_width = left.2 - left.0;
+    let right_width = right.2 - right.0;
+    let left_height = left.3 - left.1;
+    let right_height = right.3 - right.1;
+    let horizontal_overlap = (left.2.min(right.2) - left.0.max(right.0)).max(0.0);
+    let vertical_overlap = (left.3.min(right.3) - left.1.max(right.1)).max(0.0);
+    let horizontal_gap = (left.0.max(right.0) - left.2.min(right.2)).max(0.0);
+    let vertical_gap = (left.1.max(right.1) - left.3.min(right.3)).max(0.0);
+
+    (vertical_overlap >= left_height.min(right_height) * 0.5
+        && horizontal_gap <= left_width.min(right_width) * 0.5)
+        || (horizontal_overlap >= left_width.min(right_width) * 0.5
+            && vertical_gap <= left_height.min(right_height) * 0.5)
+}
+
 #[derive(Debug)]
 struct TextAnchorTable {
     table: Table,
@@ -1187,6 +1209,279 @@ pub fn detect_tables_from_lines(items: &[TextItem], lines: &[PdfLine], page: u32
     detect_tables_from_lines_inner(items, lines, page, true, true)
 }
 
+/// Bounding boxes of chart panels backed by a very dense vector grid.
+///
+/// Tables support at most 25 columns, so a panel with at least 27 distinct,
+/// long vertical coordinates plus repeated horizontal rules is treated as
+/// chart geometry. When the grid is enclosed by a painted panel rectangle,
+/// the region expands to that rectangle so axis labels, legends, and source
+/// notes remain part of the figure instead of forming a heuristic table.
+pub(crate) fn detect_dense_line_chart_regions(
+    lines: &[PdfLine],
+    rects: &[PdfRect],
+    page: u32,
+) -> Vec<(f32, f32, f32, f32)> {
+    const ANGLE_TOLERANCE: f32 = 0.035;
+    const MIN_GRID_LINE_LENGTH: f32 = 40.0;
+    const EXTENT_TOLERANCE: f32 = 6.0;
+
+    let mut verticals = Vec::new();
+    let mut horizontals = Vec::new();
+    for line in lines.iter().filter(|line| line.page == page) {
+        let dx = (line.x2 - line.x1).abs();
+        let dy = (line.y2 - line.y1).abs();
+        let length = dx.hypot(dy);
+        if length < MIN_GRID_LINE_LENGTH {
+            continue;
+        }
+        if dy > 0.01 && dx / dy <= ANGLE_TOLERANCE {
+            verticals.push((
+                (line.x1 + line.x2) / 2.0,
+                line.y1.min(line.y2),
+                line.y1.max(line.y2),
+            ));
+        } else if dx > 0.01 && dy / dx <= ANGLE_TOLERANCE {
+            horizontals.push((
+                (line.y1 + line.y2) / 2.0,
+                line.x1.min(line.x2),
+                line.x1.max(line.x2),
+            ));
+        }
+    }
+    if verticals.len() < DENSE_CHART_MIN_VERTICAL_EDGES || horizontals.len() < 3 {
+        return Vec::new();
+    }
+
+    // Group similar vertical extents once. Neighboring buckets are consulted
+    // below so coordinates that straddle a bucket boundary still form one
+    // family, while each line participates in only a constant number of
+    // candidates instead of being re-scanned for every vertical anchor.
+    let extent_key = |value: f32| (value / EXTENT_TOLERANCE).round() as i32;
+    let mut extent_buckets: HashMap<(i32, i32), Vec<VerticalRule>> = HashMap::new();
+    for vertical in verticals {
+        extent_buckets
+            .entry((extent_key(vertical.1), extent_key(vertical.2)))
+            .or_default()
+            .push(vertical);
+    }
+
+    let mut grid_regions = Vec::new();
+    let extent_keys: Vec<(i32, i32)> = extent_buckets.keys().copied().collect();
+    for key in extent_keys {
+        let anchor_family = &extent_buckets[&key];
+        let anchor_bottom = anchor_family.iter().map(|vertical| vertical.1).sum::<f32>()
+            / anchor_family.len() as f32;
+        let anchor_top = anchor_family.iter().map(|vertical| vertical.2).sum::<f32>()
+            / anchor_family.len() as f32;
+
+        let mut family = Vec::new();
+        for bottom_offset in -1..=1 {
+            for top_offset in -1..=1 {
+                if let Some(bucket) =
+                    extent_buckets.get(&(key.0 + bottom_offset, key.1 + top_offset))
+                {
+                    family.extend(bucket.iter().copied().filter(|vertical| {
+                        (vertical.1 - anchor_bottom).abs() <= EXTENT_TOLERANCE
+                            && (vertical.2 - anchor_top).abs() <= EXTENT_TOLERANCE
+                    }));
+                }
+            }
+        }
+
+        let xs = snap_edges(&family.iter().map(|&(x, _, _)| x).collect::<Vec<_>>(), 3.0);
+        if xs.len() < DENSE_CHART_MIN_VERTICAL_EDGES {
+            continue;
+        }
+
+        let grid_bottom =
+            family.iter().map(|vertical| vertical.1).sum::<f32>() / family.len() as f32;
+        let grid_top = family.iter().map(|vertical| vertical.2).sum::<f32>() / family.len() as f32;
+        if grid_top - grid_bottom < 60.0 {
+            continue;
+        }
+
+        // A horizontal rule must support the same contiguous dense run of
+        // vertical coordinates. Splitting at sparse X gaps prevents a shared
+        // rule from joining a chart to a neighboring ruled table. Keying by
+        // the covered X-index range also lets multiple chart panels sharing
+        // the same Y extents produce independent regions.
+        let mut supported_spans: HashMap<(usize, usize), Vec<f32>> = HashMap::new();
+        for &(y, line_left, line_right) in &horizontals {
+            if y < grid_bottom - EXTENT_TOLERANCE || y > grid_top + EXTENT_TOLERANCE {
+                continue;
+            }
+            let start = xs.partition_point(|&x| x < line_left - EXTENT_TOLERANCE);
+            let end = xs.partition_point(|&x| x <= line_right + EXTENT_TOLERANCE);
+            if end - start < DENSE_CHART_MIN_VERTICAL_EDGES {
+                continue;
+            }
+
+            let mut gaps: Vec<f32> = xs[start..end]
+                .windows(2)
+                .map(|pair| pair[1] - pair[0])
+                .collect();
+            gaps.sort_by(f32::total_cmp);
+            let dense_gap = gaps[gaps.len() / 4];
+            let run_break = (dense_gap * 3.0).max(12.0);
+            let locally_dense_gap_limit = (dense_gap * 1.5).max(6.0);
+            let locally_dense_gaps = gaps
+                .iter()
+                .filter(|&&gap| gap <= locally_dense_gap_limit)
+                .count();
+
+            let mut run_start = start;
+            let mut retained_dense_run = false;
+            for index in start..end - 1 {
+                if xs[index + 1] - xs[index] <= run_break {
+                    continue;
+                }
+                let run_end = index + 1;
+                if run_end - run_start >= DENSE_CHART_MIN_VERTICAL_EDGES
+                    && xs[run_end - 1] - xs[run_start] >= 120.0
+                {
+                    supported_spans
+                        .entry((run_start, run_end))
+                        .or_default()
+                        .push(y);
+                    retained_dense_run = true;
+                }
+                run_start = run_end;
+            }
+            if end - run_start >= DENSE_CHART_MIN_VERTICAL_EDGES
+                && xs[end - 1] - xs[run_start] >= 120.0
+            {
+                supported_spans.entry((run_start, end)).or_default().push(y);
+                retained_dense_run = true;
+            }
+
+            // One or two wider category gaps may split an otherwise dense
+            // chart into sub-threshold runs. Keep the full family only when
+            // its total width remains close to the expected dense spacing;
+            // a neighboring sparse table makes this ratio much larger.
+            let span_width = xs[end - 1] - xs[start];
+            let expected_dense_width = dense_gap * (end - start - 1) as f32;
+            if !retained_dense_run
+                && span_width >= 120.0
+                && span_width <= expected_dense_width * 1.35
+                && gaps.len().saturating_sub(locally_dense_gaps) <= 2
+            {
+                supported_spans.entry((start, end)).or_default().push(y);
+            }
+        }
+
+        for ((start, end), ys) in supported_spans {
+            if snap_edges(&ys, 3.0).len() >= 3 {
+                grid_regions.push((xs[start], grid_bottom, xs[end - 1], grid_top));
+            }
+        }
+    }
+
+    // Prefer the smallest qualifying region when a broad rule happens to
+    // cover a denser nested panel, and retain every non-overlapping panel.
+    grid_regions.sort_by(|left, right| {
+        let left_area = (left.2 - left.0) * (left.3 - left.1);
+        let right_area = (right.2 - right.0) * (right.3 - right.1);
+        left_area.total_cmp(&right_area)
+    });
+    let mut selected_regions: Vec<(f32, f32, f32, f32)> = Vec::new();
+    for region in grid_regions {
+        let area = (region.2 - region.0) * (region.3 - region.1);
+        let duplicates_existing = selected_regions.iter().any(|existing| {
+            let overlap_width = (region.2.min(existing.2) - region.0.max(existing.0)).max(0.0);
+            let overlap_height = (region.3.min(existing.3) - region.1.max(existing.1)).max(0.0);
+            let overlap_area = overlap_width * overlap_height;
+            let existing_area = (existing.2 - existing.0) * (existing.3 - existing.1);
+            overlap_area >= area.min(existing_area) * 0.8
+        });
+        if !duplicates_existing {
+            selected_regions.push(region);
+        }
+    }
+
+    let all_grid_regions = selected_regions.clone();
+    let mut regions: Vec<_> = selected_regions
+        .into_iter()
+        .map(|grid_region| {
+            let (grid_left, grid_bottom, grid_right, grid_top) = grid_region;
+            let enclosing_panel = rects
+                .iter()
+                .filter(|rect| rect.page == page)
+                .filter_map(|rect| {
+                    let (left, width) = if rect.width < 0.0 {
+                        (rect.x + rect.width, -rect.width)
+                    } else {
+                        (rect.x, rect.width)
+                    };
+                    let (bottom, height) = if rect.height < 0.0 {
+                        (rect.y + rect.height, -rect.height)
+                    } else {
+                        (rect.y, rect.height)
+                    };
+                    let right = left + width;
+                    let top = bottom + height;
+                    let enclosed_grids: Vec<_> = all_grid_regions
+                        .iter()
+                        .filter(|&&(other_left, other_bottom, other_right, other_top)| {
+                            left <= other_left + EXTENT_TOLERANCE
+                                && right >= other_right - EXTENT_TOLERANCE
+                                && bottom <= other_bottom + EXTENT_TOLERANCE
+                                && top >= other_top - EXTENT_TOLERANCE
+                        })
+                        .copied()
+                        .collect();
+                    if enclosed_grids.len() > DENSE_CHART_MAX_SHARED_PANEL_GRIDS
+                        || enclosed_grids.iter().any(|&other| {
+                            other != grid_region
+                                && !dense_chart_grids_are_co_located(grid_region, other)
+                        })
+                    {
+                        return None;
+                    }
+                    let enclosed_grid_bounds =
+                        enclosed_grids.into_iter().reduce(|bounds, other| {
+                            (
+                                bounds.0.min(other.0),
+                                bounds.1.min(other.1),
+                                bounds.2.max(other.2),
+                                bounds.3.max(other.3),
+                            )
+                        })?;
+                    let enclosed_width = enclosed_grid_bounds.2 - enclosed_grid_bounds.0;
+                    let enclosed_height = enclosed_grid_bounds.3 - enclosed_grid_bounds.1;
+                    (left <= grid_left + EXTENT_TOLERANCE
+                        && right >= grid_right - EXTENT_TOLERANCE
+                        && bottom <= grid_bottom + EXTENT_TOLERANCE
+                        && top >= grid_top - EXTENT_TOLERANCE
+                        && width <= enclosed_width * 2.0
+                        && height <= enclosed_height * 4.0
+                        && !(left < 5.0 && bottom < 5.0))
+                        .then_some(((left, bottom, right, top), width * height))
+                })
+                .min_by(|left, right| left.1.total_cmp(&right.1))
+                .map(|(region, _)| region);
+
+            enclosing_panel.unwrap_or((
+                grid_left - DENSE_CHART_LABEL_PAD,
+                grid_bottom - DENSE_CHART_LABEL_PAD,
+                grid_right + DENSE_CHART_LABEL_PAD,
+                grid_top + DENSE_CHART_LABEL_PAD,
+            ))
+        })
+        .collect();
+    regions.sort_by(|left, right| {
+        left.0
+            .total_cmp(&right.0)
+            .then_with(|| left.1.total_cmp(&right.1))
+    });
+    regions.dedup_by(|left, right| {
+        (left.0 - right.0).abs() <= EXTENT_TOLERANCE
+            && (left.1 - right.1).abs() <= EXTENT_TOLERANCE
+            && (left.2 - right.2).abs() <= EXTENT_TOLERANCE
+            && (left.3 - right.3).abs() <= EXTENT_TOLERANCE
+    });
+    regions
+}
+
 /// Detect only tables whose cell grid is backed by explicit vector geometry.
 ///
 /// Region-level TSR callers need physical cell boundaries for crop bboxes, so
@@ -1621,6 +1916,159 @@ mod tests {
         }
     }
 
+    #[test]
+    fn dense_vector_grid_expands_to_enclosing_chart_panel() {
+        let mut lines: Vec<PdfLine> = (0..30)
+            .map(|column| make_vline(100.0 + column as f32 * 8.0, 400.0, 550.0, 1))
+            .collect();
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 100.0, 332.0, 1)));
+        let rects = vec![PdfRect {
+            x: 80.0,
+            y: 350.0,
+            width: 280.0,
+            height: 240.0,
+            page: 1,
+        }];
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &rects, 1),
+            vec![(80.0, 350.0, 360.0, 590.0)]
+        );
+    }
+
+    #[test]
+    fn frameless_dense_vector_grid_includes_label_padding() {
+        let mut lines: Vec<PdfLine> = (0..30)
+            .map(|column| make_vline(100.0 + column as f32 * 8.0, 400.0, 550.0, 1))
+            .collect();
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 100.0, 332.0, 1)));
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &[], 1),
+            vec![(80.0, 380.0, 352.0, 570.0)]
+        );
+    }
+
+    #[test]
+    fn multiple_dense_vector_panels_are_retained() {
+        let mut lines = Vec::new();
+        for panel_left in [60.0, 380.0] {
+            lines.extend(
+                (0..30).map(|column| make_vline(panel_left + column as f32 * 8.0, 400.0, 550.0, 1)),
+            );
+            lines.extend((0..6).map(|row| {
+                make_hline(400.0 + row as f32 * 30.0, panel_left, panel_left + 232.0, 1)
+            }));
+        }
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &[], 1),
+            vec![(40.0, 380.0, 312.0, 570.0), (360.0, 380.0, 632.0, 570.0),]
+        );
+    }
+
+    #[test]
+    fn multiple_dense_vector_panels_use_shared_enclosing_panel() {
+        let mut lines = Vec::new();
+        for panel_left in [60.0, 380.0] {
+            lines.extend(
+                (0..30).map(|column| make_vline(panel_left + column as f32 * 8.0, 400.0, 550.0, 1)),
+            );
+            lines.extend((0..6).map(|row| {
+                make_hline(400.0 + row as f32 * 30.0, panel_left, panel_left + 232.0, 1)
+            }));
+        }
+        let rects = vec![PdfRect {
+            x: 40.0,
+            y: 350.0,
+            width: 592.0,
+            height: 240.0,
+            page: 1,
+        }];
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &rects, 1),
+            vec![(40.0, 350.0, 632.0, 590.0)]
+        );
+    }
+
+    #[test]
+    fn shared_rules_do_not_join_dense_chart_to_adjacent_table() {
+        let mut lines: Vec<PdfLine> = (0..30)
+            .map(|column| make_vline(60.0 + column as f32 * 8.0, 400.0, 550.0, 1))
+            .collect();
+
+        lines.extend(
+            [330.0, 390.0, 450.0, 510.0, 570.0, 630.0]
+                .into_iter()
+                .map(|x| make_vline(x, 400.0, 550.0, 1)),
+        );
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 60.0, 630.0, 1)));
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &[], 1),
+            vec![(40.0, 380.0, 312.0, 570.0)]
+        );
+    }
+
+    #[test]
+    fn uneven_dense_spacing_keeps_the_complete_chart_region() {
+        let mut xs: Vec<f32> = (0..15).map(|column| 60.0 + column as f32 * 8.0).collect();
+        xs.extend((0..15).map(|column| 212.0 + column as f32 * 8.0));
+        let mut lines: Vec<PdfLine> = xs.iter().map(|&x| make_vline(x, 400.0, 550.0, 1)).collect();
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 60.0, 324.0, 1)));
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &[], 1),
+            vec![(40.0, 380.0, 344.0, 570.0)]
+        );
+    }
+
+    #[test]
+    fn subthreshold_dense_run_does_not_absorb_adjacent_sparse_grid() {
+        let mut xs: Vec<f32> = (0..21).map(|column| 60.0 + column as f32 * 8.0).collect();
+        xs.extend((0..6).map(|column| 248.0 + column as f32 * 18.0));
+        let mut lines: Vec<PdfLine> = xs.iter().map(|&x| make_vline(x, 400.0, 550.0, 1)).collect();
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 60.0, 338.0, 1)));
+
+        assert!(detect_dense_line_chart_regions(&lines, &[], 1).is_empty());
+    }
+
+    #[test]
+    fn broad_frame_does_not_merge_distant_dense_grids() {
+        let mut lines = Vec::new();
+        for panel_left in [60.0, 700.0] {
+            lines.extend(
+                (0..30).map(|column| make_vline(panel_left + column as f32 * 8.0, 400.0, 550.0, 1)),
+            );
+            lines.extend((0..6).map(|row| {
+                make_hline(400.0 + row as f32 * 30.0, panel_left, panel_left + 232.0, 1)
+            }));
+        }
+        let rects = vec![PdfRect {
+            x: 40.0,
+            y: 350.0,
+            width: 912.0,
+            height: 240.0,
+            page: 1,
+        }];
+
+        assert_eq!(
+            detect_dense_line_chart_regions(&lines, &rects, 1),
+            vec![(40.0, 380.0, 312.0, 570.0), (680.0, 380.0, 952.0, 570.0)]
+        );
+    }
+
+    #[test]
+    fn supported_width_vector_table_is_not_a_dense_chart() {
+        let mut lines: Vec<PdfLine> = (0..26)
+            .map(|column| make_vline(100.0 + column as f32 * 10.0, 400.0, 550.0, 1))
+            .collect();
+        lines.extend((0..6).map(|row| make_hline(400.0 + row as f32 * 30.0, 100.0, 350.0, 1)));
+
+        assert!(detect_dense_line_chart_regions(&lines, &[], 1).is_empty());
+    }
+
     #[test]
     fn test_basic_grid_detection() {
         // 3x2 grid with horizontal lines at y=500, 480, 460 and vertical at x=100, 200, 300

src/tables/detect_rects.rs

@@ -1996,17 +1996,249 @@ fn without_dominant_page_backgrounds(rects: &[(f32, f32, f32, f32)]) -> Vec<(f32
         .collect()
 }
 
-/// Detect a table from cell-background rects that failed grid detection.
+/// Repeated rows of touching cell rectangles are stronger table evidence
+/// than the bar-length variation used by the chart detector.
 ///
-/// Uses rect Y-edges for row boundaries and text X-position clustering for
-/// columns.  Handles tables with cell backgrounds that don't form a clean
-/// X-edge grid (variable column widths, decorative fills).
-/// Chart-bar signature: ≥3 rects sharing an aligned bottom edge (the axis),
-/// with similar widths (bars) but strongly varying heights (data-driven),
-/// holding at most a single numeric data label each. Bar charts drawn as
-/// filled rects otherwise read as cell rects and grid their axis labels
-/// into a phantom table. The mirrored check catches horizontal bar charts.
-fn is_chart_bar_cluster(
+/// Ruled tables with wrapped labels naturally have variable row heights, and
+/// numeric-heavy cells can otherwise resemble horizontal or vertical bars.
+/// Require several rows to repeat a shared edge schema before overriding the
+/// chart hypothesis so sparse plots and independent bars remain unaffected.
+fn is_repeated_cell_grid(group_rects: &[(f32, f32, f32, f32)]) -> bool {
+    type RowGroup = (f32, f32, Vec<(f32, f32)>);
+
+    const ROW_EDGE_TOLERANCE: f32 = 3.0;
+    const MIN_GRID_ROWS: usize = 4;
+    const MIN_CELLS_PER_ROW: usize = 3;
+
+    if group_rects.len() < MIN_GRID_ROWS * MIN_CELLS_PER_ROW {
+        return false;
+    }
+
+    let mut row_groups: Vec<RowGroup> = Vec::new();
+    for &(x, y, width, height) in group_rects {
+        if width < 5.0 || height < 5.0 {
+            continue;
+        }
+        let top = y + height;
+        if let Some((_, _, cells)) = row_groups.iter_mut().find(|(bottom, row_top, _)| {
+            (y - *bottom).abs() <= ROW_EDGE_TOLERANCE
+                && (top - *row_top).abs() <= ROW_EDGE_TOLERANCE
+        }) {
+            cells.push((x, x + width));
+        } else {
+            row_groups.push((y, top, vec![(x, x + width)]));
+        }
+    }
+
+    let mut row_schemas = Vec::new();
+    for (_, _, mut cells) in row_groups {
+        if cells.len() < MIN_CELLS_PER_ROW {
+            continue;
+        }
+        let mut widths: Vec<f32> = cells.iter().map(|&(left, right)| right - left).collect();
+        widths.sort_by(f32::total_cmp);
+        let median_width = widths[widths.len() / 2];
+        cells.retain(|&(left, right)| right - left <= median_width * 2.5);
+        cells.sort_by(|left, right| {
+            left.0
+                .total_cmp(&right.0)
+                .then_with(|| left.1.total_cmp(&right.1))
+        });
+        cells.dedup_by(|left, right| {
+            (left.0 - right.0).abs() <= ROW_EDGE_TOLERANCE
+                && (left.1 - right.1).abs() <= ROW_EDGE_TOLERANCE
+        });
+        if cells.len() < MIN_CELLS_PER_ROW
+            || cells
+                .windows(2)
+                .any(|pair| pair[1].0 > pair[0].1 + ROW_EDGE_TOLERANCE)
+        {
+            continue;
+        }
+        let edges: Vec<f32> = cells
+            .iter()
+            .flat_map(|&(left, right)| [left, right])
+            .collect();
+        let schema = snap_edges(&edges, ROW_EDGE_TOLERANCE);
+        if schema.len() > MIN_CELLS_PER_ROW {
+            row_schemas.push(schema);
+        }
+    }
+    if row_schemas.len() < MIN_GRID_ROWS {
+        return false;
+    }
+
+    let reference = row_schemas
+        .iter()
+        .max_by_key(|schema| schema.len())
+        .expect("grid rows are non-empty");
+    row_schemas
+        .iter()
+        .filter(|schema| {
+            let comparable_edges = reference.len().min(schema.len());
+            let matched_edges = schema
+                .iter()
+                .filter(|edge| {
+                    reference
+                        .iter()
+                        .any(|reference_edge| (*edge - *reference_edge).abs() <= ROW_EDGE_TOLERANCE)
+                })
+                .count();
+            matched_edges > MIN_CELLS_PER_ROW && matched_edges * 4 >= comparable_edges * 3
+        })
+        .count()
+        >= MIN_GRID_ROWS
+}
+
+fn repeated_cell_grid_overrides_bar_hypothesis(group_rects: &[(f32, f32, f32, f32)]) -> bool {
+    is_repeated_cell_grid(group_rects)
+        && without_dominant_page_backgrounds(group_rects).len() == group_rects.len()
+}
+
+/// Detect horizontal segmented stacks from aligned rows of touching rects.
+///
+/// Category rows must have visible gutters and data-varying internal segment
+/// boundaries, unlike the stable boundaries of a ruled table.
+struct SegmentedBarGeometry {
+    bounds: (f32, f32, f32, f32),
+    row_bands: Vec<(f32, f32)>,
+}
+
+fn segmented_stacked_bar_geometry(
+    group_rects: &[(f32, f32, f32, f32)],
+) -> Option<SegmentedBarGeometry> {
+    type BarRow = (f32, f32, Vec<(f32, f32)>);
+
+    const EDGE_TOLERANCE: f32 = 3.0;
+    const MIN_ROWS: usize = 4;
+    const MIN_SEGMENTS: usize = 3;
+
+    let mut rows: Vec<BarRow> = Vec::new();
+    for &(x, y, width, height) in group_rects {
+        if width < 5.0 || height < 5.0 {
+            continue;
+        }
+        let top = y + height;
+        if let Some((_, _, segments)) = rows.iter_mut().find(|(bottom, row_top, _)| {
+            (y - *bottom).abs() <= EDGE_TOLERANCE && (top - *row_top).abs() <= EDGE_TOLERANCE
+        }) {
+            segments.push((x, x + width));
+        } else {
+            rows.push((y, top, vec![(x, x + width)]));
+        }
+    }
+
+    rows.retain_mut(|(_, _, segments)| {
+        segments.sort_by(|left, right| left.0.total_cmp(&right.0));
+        segments.len() >= MIN_SEGMENTS
+            && segments
+                .windows(2)
+                .all(|pair| (pair[1].0 - pair[0].1).abs() <= EDGE_TOLERANCE)
+    });
+    if rows.len() < MIN_ROWS {
+        return None;
+    }
+    rows.sort_by(|left, right| left.0.total_cmp(&right.0));
+
+    // Table rows normally share borders. Horizontal stacked bars instead
+    // leave a visible gutter between category rows.
+    if rows.windows(2).any(|pair| {
+        let shorter_height = (pair[0].1 - pair[0].0).min(pair[1].1 - pair[1].0);
+        pair[1].0 - pair[0].1 < (shorter_height * 0.25).max(2.0)
+    }) {
+        return None;
+    }
+
+    // At least two rows must move an internal segment boundary. Stable
+    // boundaries across every row are stronger evidence for a ruled table.
+    let reference_edges: Vec<f32> = rows[0]
+        .2
+        .iter()
+        .take(rows[0].2.len() - 1)
+        .map(|segment| segment.1)
+        .collect();
+    let drifting_rows = rows
+        .iter()
+        .skip(1)
+        .filter(|(_, _, segments)| {
+            let edges: Vec<f32> = segments
+                .iter()
+                .take(segments.len() - 1)
+                .map(|segment| segment.1)
+                .collect();
+            edges.len() == reference_edges.len()
+                && edges
+                    .iter()
+                    .zip(&reference_edges)
+                    .any(|(edge, reference)| (edge - reference).abs() > EDGE_TOLERANCE)
+        })
+        .count();
+    if drifting_rows < 2 {
+        return None;
+    }
+
+    let left = rows
+        .iter()
+        .flat_map(|row| &row.2)
+        .map(|segment| segment.0)
+        .reduce(f32::min)?;
+    let right = rows
+        .iter()
+        .flat_map(|row| &row.2)
+        .map(|segment| segment.1)
+        .reduce(f32::max)?;
+    let bottom = rows.iter().map(|row| row.0).reduce(f32::min)?;
+    let top = rows.iter().map(|row| row.1).reduce(f32::max)?;
+    let row_bands = rows.iter().map(|row| (row.0, row.1)).collect();
+    Some(SegmentedBarGeometry {
+        bounds: (left, bottom, right, top),
+        row_bands,
+    })
+}
+
+/// Category labels beside multiple bar rows are independent chart evidence:
+/// numeric table text stays inside its cells, regardless of whether the table
+/// has an outer border or extra padding.
+fn has_external_segmented_bar_labels(
+    items: &[TextItem],
+    page: u32,
+    geometry: &SegmentedBarGeometry,
+) -> bool {
+    const LABEL_EDGE_TOLERANCE: f32 = 3.0;
+    const LABEL_CLAIM_PAD: f32 = 20.0;
+
+    let (content_left, _, content_right, _) = geometry.bounds;
+    let labeled_rows = geometry
+        .row_bands
+        .iter()
+        .filter(|&&(row_bottom, row_top)| {
+            items.iter().any(|item| {
+                if item.page != page || item.text.trim().is_empty() {
+                    return false;
+                }
+                let item_left = item.x.min(item.x + item.width);
+                let item_right = item.x.max(item.x + item.width);
+                let item_center_x = (item_left + item_right) / 2.0;
+                let item_center_y = item.y + item.height / 2.0;
+                let beside_stack = (item_center_x <= content_left + LABEL_EDGE_TOLERANCE
+                    && item_center_x >= content_left - LABEL_CLAIM_PAD
+                    && item_left < content_left)
+                    || (item_center_x >= content_right - LABEL_EDGE_TOLERANCE
+                        && item_center_x <= content_right + LABEL_CLAIM_PAD
+                        && item_right > content_right);
+                beside_stack
+                    && item_center_y >= row_bottom - LABEL_EDGE_TOLERANCE
+                    && item_center_y <= row_top + LABEL_EDGE_TOLERANCE
+            })
+        })
+        .count();
+
+    labeled_rows >= 2 && labeled_rows * 2 >= geometry.row_bands.len()
+}
+
+/// Recognize filled vertical or horizontal bars whose geometry and labels are
+/// data-driven rather than uniform table cells.
+fn has_chart_bar_signature(
     items: &[TextItem],
     group_rects: &[(f32, f32, f32, f32)],
     page: u32,
@@ -2113,6 +2345,30 @@ fn is_chart_bar_cluster(
         || bar_family(|r| r.1, |r| r.3, |r| r.2, |r| r.0)
 }
 
+fn is_chart_bar_cluster(
+    items: &[TextItem],
+    group_rects: &[(f32, f32, f32, f32)],
+    page: u32,
+) -> bool {
+    let has_bar_signature = has_chart_bar_signature(items, group_rects, page);
+
+    // A segmented horizontal chart can share most of its edges across rows.
+    // Row-aligned category labels outside the stack distinguish it from a
+    // numeric table without depending on whether either shape has a frame.
+    if has_bar_signature {
+        if let Some(geometry) = segmented_stacked_bar_geometry(group_rects) {
+            if has_external_segmented_bar_labels(items, page, &geometry) {
+                return true;
+            }
+        }
+    }
+    if repeated_cell_grid_overrides_bar_hypothesis(group_rects) {
+        return false;
+    }
+
+    has_bar_signature
+}
+
 fn detect_row_stripe_table_from_cell_rects(
     items: &[TextItem],
     group_rects: &[(f32, f32, f32, f32)],
@@ -3083,6 +3339,194 @@ mod tests {
         assert!(detect_chart_regions(&items, &rects, 1).is_empty());
     }
 
+    #[test]
+    fn variable_height_ruled_grid_overrides_bar_hypothesis() {
+        let edge_sets = [
+            [80.0, 140.0, 200.0, 260.0, 320.0, 380.0, 440.0, 500.0, 560.0],
+            [80.0, 140.0, 210.0, 260.0, 320.0, 380.0, 450.0, 500.0, 560.0],
+        ];
+        let heights = [20.0, 34.0, 26.0, 42.0, 20.0, 34.0];
+        let edge_variants = [0, 0, 0, 0, 1, 1];
+        let mut rects = Vec::new();
+        let mut y = 650.0;
+        for (row, height) in heights.into_iter().enumerate() {
+            let edges = edge_sets[edge_variants[row]];
+            rects.extend(
+                edges
+                    .windows(2)
+                    .map(|edge| (edge[0], y, edge[1] - edge[0], height)),
+            );
+            y -= height;
+        }
+
+        assert!(is_repeated_cell_grid(&rects));
+        assert!(has_chart_bar_signature(&[], &rects, 1));
+        assert!(repeated_cell_grid_overrides_bar_hypothesis(&rects));
+        assert!(segmented_stacked_bar_geometry(&rects).is_none());
+        assert!(!is_chart_bar_cluster(&[], &rects, 1));
+
+        let mut with_page_fills =
+            vec![(0.0, 0.0, 600.0, 800.0); DOMINANT_PAGE_BACKGROUND_MIN_REPETITIONS];
+        with_page_fills.extend(rects);
+        assert!(!repeated_cell_grid_overrides_bar_hypothesis(
+            &with_page_fills
+        ));
+    }
+
+    #[test]
+    fn touching_segments_with_spaced_rows_remain_a_chart() {
+        let row_edges = [
+            [100.0, 140.0, 180.0, 220.0, 260.0],
+            [100.0, 140.0, 180.0, 228.0, 260.0],
+            [100.0, 140.0, 180.0, 214.0, 260.0],
+            [100.0, 140.0, 180.0, 232.0, 260.0],
+        ];
+        let mut raw_rects = vec![(90.0, 530.0, 190.0, 100.0)];
+        for (row, edges) in row_edges.into_iter().enumerate() {
+            let y = 540.0 + row as f32 * 20.0;
+            raw_rects.extend(
+                edges
+                    .windows(2)
+                    .map(|edge| (edge[0], y, edge[1] - edge[0], 12.0)),
+            );
+        }
+        let items: Vec<TextItem> = (0..4)
+            .map(|row| make_item("Category", 62.0, 541.0 + row as f32 * 20.0, 9.0))
+            .collect();
+
+        assert!(is_repeated_cell_grid(&raw_rects));
+        assert!(has_chart_bar_signature(&items, &raw_rects, 1));
+        let geometry = segmented_stacked_bar_geometry(&raw_rects).expect("segmented stack");
+        assert!(has_external_segmented_bar_labels(&items, 1, &geometry));
+        assert!(is_chart_bar_cluster(&items, &raw_rects, 1));
+
+        let numeric_items: Vec<TextItem> = (0..4)
+            .map(|row| make_item("2024", 80.0, 541.0 + row as f32 * 18.0, 9.0))
+            .collect();
+        assert!(has_external_segmented_bar_labels(
+            &numeric_items,
+            1,
+            &geometry
+        ));
+        assert!(is_chart_bar_cluster(&numeric_items, &raw_rects, 1));
+
+        let edge_adjacent_items: Vec<TextItem> = (0..4)
+            .map(|row| make_item("2024", 92.0, 541.0 + row as f32 * 18.0, 9.0))
+            .collect();
+        assert!(has_external_segmented_bar_labels(
+            &edge_adjacent_items,
+            1,
+            &geometry
+        ));
+        assert!(is_chart_bar_cluster(&edge_adjacent_items, &raw_rects, 1));
+
+        let far_items: Vec<TextItem> = (0..4)
+            .map(|row| make_item("Category", 20.0, 541.0 + row as f32 * 18.0, 9.0))
+            .collect();
+        assert!(!has_external_segmented_bar_labels(&far_items, 1, &geometry));
+        assert!(!is_chart_bar_cluster(&far_items, &raw_rects, 1));
+
+        let rects: Vec<PdfRect> = raw_rects
+            .into_iter()
+            .map(|(x, y, width, height)| PdfRect {
+                x,
+                y,
+                width,
+                height,
+                page: 1,
+            })
+            .collect();
+        assert_eq!(detect_chart_regions(&items, &rects, 1).len(), 1);
+        let (tables, hints) = detect_tables_from_rects(&items, &rects, 1);
+        assert!(tables.is_empty());
+        assert!(hints.is_empty());
+    }
+
+    #[test]
+    fn padded_numeric_grid_frame_remains_a_table() {
+        let row_edges = [
+            [100.0, 140.0, 180.0, 220.0, 260.0],
+            [100.0, 140.0, 180.0, 228.0, 260.0],
+            [100.0, 140.0, 180.0, 214.0, 260.0],
+            [100.0, 140.0, 180.0, 232.0, 260.0],
+        ];
+        let mut raw_rects = vec![(96.0, 536.0, 168.0, 80.0)];
+        let mut items = Vec::new();
+        for (row, edges) in row_edges.into_iter().enumerate() {
+            let y = 540.0 + row as f32 * 20.0;
+            for edge in edges.windows(2) {
+                raw_rects.push((edge[0], y, edge[1] - edge[0], 12.0));
+                items.push(make_item("42", edge[0] + 8.0, y + 1.0, 9.0));
+            }
+        }
+
+        assert!(is_repeated_cell_grid(&raw_rects));
+        assert!(has_chart_bar_signature(&items, &raw_rects, 1));
+        let geometry = segmented_stacked_bar_geometry(&raw_rects).expect("segmented rows");
+        assert!(!has_external_segmented_bar_labels(&items, 1, &geometry));
+        assert!(!is_chart_bar_cluster(&items, &raw_rects, 1));
+
+        let flush_items: Vec<TextItem> = (0..4)
+            .map(|row| make_item("1", 100.0, 541.0 + row as f32 * 20.0, 9.0))
+            .collect();
+        assert!(!has_external_segmented_bar_labels(
+            &flush_items,
+            1,
+            &geometry
+        ));
+        assert!(!is_chart_bar_cluster(&flush_items, &raw_rects, 1));
+
+        let rects: Vec<PdfRect> = raw_rects
+            .into_iter()
+            .map(|(x, y, width, height)| PdfRect {
+                x,
+                y,
+                width,
+                height,
+                page: 1,
+            })
+            .collect();
+        assert!(detect_chart_regions(&items, &rects, 1).is_empty());
+        assert!(!detect_tables_from_rects(&items, &rects, 1).0.is_empty());
+    }
+
+    #[test]
+    fn frameless_segmented_chart_with_category_labels_remains_a_chart() {
+        let row_edges = [
+            [100.0, 140.0, 180.0, 220.0, 260.0],
+            [100.0, 140.0, 180.0, 228.0, 260.0],
+            [100.0, 140.0, 180.0, 214.0, 260.0],
+            [100.0, 140.0, 180.0, 232.0, 260.0],
+        ];
+        let mut raw_rects = Vec::new();
+        let mut items = Vec::new();
+        for (row, edges) in row_edges.into_iter().enumerate() {
+            let y = 540.0 + row as f32 * 18.0;
+            raw_rects.extend(
+                edges
+                    .windows(2)
+                    .map(|edge| (edge[0], y, edge[1] - edge[0], 12.0)),
+            );
+            items.push(make_item("Category", 62.0, y + 1.0, 9.0));
+        }
+
+        let geometry = segmented_stacked_bar_geometry(&raw_rects).expect("segmented stack");
+        assert!(has_external_segmented_bar_labels(&items, 1, &geometry));
+        assert!(is_chart_bar_cluster(&items, &raw_rects, 1));
+
+        let rects: Vec<PdfRect> = raw_rects
+            .into_iter()
+            .map(|(x, y, width, height)| PdfRect {
+                x,
+                y,
+                width,
+                height,
+                page: 1,
+            })
+            .collect();
+        assert_eq!(detect_chart_regions(&items, &rects, 1).len(), 1);
+    }
+
     // --- detect_stacked_box_table ---
 
     /// N stacked boxes at x=100, w=300, h=22, top-to-bottom from y=600.

src/tables/mod.rs

@@ -16,7 +16,9 @@ pub(crate) use detect_heuristic::{
     content_width, detect_tables_with_page_width, is_table_of_contents,
 };
 pub use detect_lines::detect_tables_from_lines;
-pub(crate) use detect_lines::detect_vector_grid_tables_from_lines;
+pub(crate) use detect_lines::{
+    detect_dense_line_chart_regions, detect_vector_grid_tables_from_lines,
+};
 pub(crate) use detect_rects::cluster_rects;
 pub use detect_rects::{detect_chart_regions, detect_tables_from_rects, RectHintRegion};
 pub use detect_struct::detect_tables_from_struct_tree;

src/text_utils.rs

@@ -70,10 +70,17 @@ pub(crate) fn is_page_number_line(text: &str) -> bool {
 
     let lowercase = text.trim().to_ascii_lowercase();
     lowercase.strip_prefix("page").is_some_and(|rest| {
-        rest.trim_start()
-            .chars()
-            .next()
+        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)
     })
 }
 

src/tounicode.rs

@@ -594,7 +594,7 @@ fn hex_to_unicode_string(hex: &str) -> Option<String> {
 
     let bytes: Option<Vec<u8>> = (0..hex.len())
         .step_by(2)
-        .map(|i| u8::from_str_radix(&hex[i..i + 2], 16).ok())
+        .map(|i| u8::from_str_radix(hex.get(i..i + 2)?, 16).ok())
         .collect();
     let bytes = bytes?;
 
@@ -880,6 +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) {
@@ -2587,6 +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>
@@ -3159,4 +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"
+        );
+    }
 }

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

@@ -1429,6 +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
@@ -1583,6 +1654,282 @@ fn test_extract_regions_mem_basic_text_pdf() {
     assert_eq!(regions[0].page, 0);
 }
 
+/// Build a synthetic "scanned page" PDF: a full-page image XObject with a
+/// text layer drawn in the given render mode (3 = invisible OCR overlay,
+/// 0 = normal visible fill). `visible_extra` optionally adds a normally
+/// rendered line so double-layer behavior can be tested; `layer_lines`
+/// overrides the layer content (default: three pangram lines);
+/// `quote_ops` shows every layer line via the `'` operator instead of Tj
+/// (both are standard show-text encodings for OCR layers).
+fn make_pdf_with_custom_text_layer(
+    text_render_mode: i32,
+    visible_extra: Option<&str>,
+    layer_lines: Option<&[&str]>,
+    quote_ops: bool,
+) -> 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");
+    }
+    fn add_stream_object(
+        pdf: &mut Vec<u8>,
+        offsets: &mut Vec<usize>,
+        id: usize,
+        dict: &str,
+        stream_bytes: &[u8],
+    ) {
+        offsets.push(pdf.len());
+        pdf.extend_from_slice(format!("{id} 0 obj\n").as_bytes());
+        pdf.extend_from_slice(
+            format!("<< {} /Length {} >>\nstream\n", dict, stream_bytes.len()).as_bytes(),
+        );
+        pdf.extend_from_slice(stream_bytes);
+        pdf.extend_from_slice(b"\nendstream\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 0 612 792] \
+         /Resources << /Font << /F1 5 0 R >> /XObject << /Im0 6 0 R >> >> \
+         /Contents 4 0 R >>",
+    );
+    // Full-page raster, then the text layer in the requested render mode —
+    // several lines so the OCR-layer gate's alnum floor (40) is well cleared.
+    let mut content = String::from("q 612 0 0 792 0 0 cm /Im0 Do Q\n");
+    let default_layer = [
+        "The quick brown fox jumps over the lazy dog",
+        "Pack my box with five dozen liquor jugs tonight",
+        "Sphinx of black quartz judge my vow carefully",
+    ];
+    let layer: &[&str] = layer_lines.unwrap_or(&default_layer);
+    if quote_ops {
+        // Every line shown via `'` (move-to-next-line + show) — nothing on
+        // this layer goes through Tj, pinning the `'` suppression path.
+        content.push_str(&format!(
+            "BT /F1 12 Tf {text_render_mode} Tr 16 TL 72 716 Td "
+        ));
+        for line in layer {
+            content.push_str(&format!("({line}) ' "));
+        }
+    } else {
+        content.push_str(&format!("BT /F1 12 Tf {text_render_mode} Tr 72 700 Td "));
+        for (i, line) in layer.iter().enumerate() {
+            if i > 0 {
+                content.push_str("0 -16 Td ");
+            }
+            content.push_str(&format!("({line}) Tj "));
+        }
+    }
+    content.push_str("ET\n");
+    if let Some(extra) = visible_extra {
+        content.push_str(&format!("BT /F1 12 Tf 0 Tr 72 500 Td ({extra}) Tj ET\n"));
+    }
+    add_stream_object(&mut pdf, &mut offsets, 4, "", content.as_bytes());
+    add_object(
+        &mut pdf,
+        &mut offsets,
+        5,
+        "<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica >>",
+    );
+    let image_pixel = [128u8];
+    add_stream_object(
+        &mut pdf,
+        &mut offsets,
+        6,
+        "/Type /XObject /Subtype /Image /Width 1 /Height 1 \
+         /ColorSpace /DeviceGray /BitsPerComponent 8",
+        &image_pixel,
+    );
+
+    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_text_layer(text_render_mode: i32, visible_extra: Option<&str>) -> Vec<u8> {
+    make_pdf_with_custom_text_layer(text_render_mode, visible_extra, None, false)
+}
+
+/// A scanned page whose only text is an invisible (Tr 3) OCR layer behind
+/// the raster must serve that layer from the region extractor instead of
+/// reporting the region as needs_ocr — the exact text is already in the PDF.
+#[test]
+fn test_extract_regions_mem_recovers_invisible_ocr_layer() {
+    let buf = make_pdf_with_text_layer(3, None);
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    assert_eq!(regions.len(), 1);
+    let region = &regions[0].regions[0];
+    assert!(
+        region.text.contains("quick brown fox"),
+        "invisible OCR layer should be served as region text, got: {:?}",
+        region.text
+    );
+    assert!(
+        !region.needs_ocr,
+        "recovered OCR layer must not fall back to GPU OCR"
+    );
+}
+
+/// ANY visible text on the page — even a single short line — must block the
+/// invisible-layer adoption entirely: the invisible pass returns visible
+/// items too, so adopting it alongside visible text would duplicate the
+/// visible words. Strict zero-visible gate, no fuzzy dedupe.
+#[test]
+fn test_extract_regions_mem_visible_text_blocks_invisible_layer() {
+    let buf = make_pdf_with_text_layer(3, Some("Folio 142"));
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert!(
+        region.text.contains("Folio 142"),
+        "visible text should be extracted, got: {:?}",
+        region.text
+    );
+    assert!(
+        !region.text.contains("quick brown fox"),
+        "invisible layer must not be adopted when any visible text exists, got: {:?}",
+        region.text
+    );
+    assert_eq!(
+        region.text.matches("Folio 142").count(),
+        1,
+        "visible text must appear exactly once, got: {:?}",
+        region.text
+    );
+}
+
+/// An invisible OCR layer shown entirely via the `'` show-text operator
+/// (move-to-next-line + show) must also be recovered — the skipped_invisible
+/// signal has to fire on every show-text path, not just Tj/TJ.
+#[test]
+fn test_extract_regions_mem_recovers_quote_operator_layer() {
+    let buf = make_pdf_with_custom_text_layer(3, None, None, true);
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert!(
+        region.text.contains("quick brown fox"),
+        "'-operator OCR layer should be recovered, got: {:?}",
+        region.text
+    );
+    assert!(!region.needs_ocr);
+}
+
+/// An invisible layer below the 40-alnum floor (a stray watermark line)
+/// must NOT be adopted — the region keeps its needs_ocr fallback.
+#[test]
+fn test_extract_regions_mem_tiny_invisible_layer_not_adopted() {
+    let buf = make_pdf_with_custom_text_layer(3, None, Some(&["Scanned by ACME"]), false);
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert!(
+        !region.text.contains("Scanned by ACME"),
+        "below-floor invisible layer must not be adopted, got: {:?}",
+        region.text
+    );
+    // Only the raster placeholder remains — needs_ocr stays whatever main
+    // reports for placeholder-only regions (false today; downstream
+    // pipelines route placeholder-only text to OCR themselves, and this PR
+    // deliberately does not change that contract).
+    assert!(
+        region.text.trim().starts_with("[Image:"),
+        "region should hold only the raster placeholder, got: {:?}",
+        region.text
+    );
+}
+
+/// An invisible layer that clears the alnum floor but is mostly symbol
+/// garbage (a broken OCR run) must be rejected by the garbage gate.
+#[test]
+fn test_extract_regions_mem_garbage_invisible_layer_not_adopted() {
+    // Each line: 5 alphanumerics among 15 symbol chars. Ten lines clear the
+    // 40-alnum floor (50 alnum) while staying well under the half-alnum
+    // ratio is_garbage_text requires.
+    let garbage_lines: Vec<&str> = vec!["a@@b%%c&&d==e~~"; 10];
+    let buf = make_pdf_with_custom_text_layer(3, None, Some(&garbage_lines), false);
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert!(
+        !region.text.contains("a@@b"),
+        "garbage invisible layer must not be adopted, got: {:?}",
+        region.text
+    );
+    assert!(
+        region.text.trim().starts_with("[Image:"),
+        "region should hold only the raster placeholder, got: {:?}",
+        region.text
+    );
+}
+
+/// Punctuation-only visible text (zero alphanumerics) must ALSO block
+/// adoption — the gate is item-presence, not alphanumeric mass. (Real-world
+/// rationale: an invisible OCR layer transcribes the raster, so visible
+/// glyphs typically have invisible twins there; this fixture's layers are
+/// disjoint, so it pins the gate itself, not the duplication scenario.)
+#[test]
+fn test_extract_regions_mem_punctuation_visible_blocks_invisible_layer() {
+    let buf = make_pdf_with_text_layer(3, Some("... --- ..."));
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert!(
+        !region.text.contains("quick brown fox"),
+        "invisible layer must not be adopted over punctuation-only visible text, got: {:?}",
+        region.text
+    );
+    assert_eq!(
+        region.text.matches("... --- ...").count(),
+        1,
+        "visible punctuation must be preserved exactly once, got: {:?}",
+        region.text
+    );
+}
+
+/// Regression guard: a normal visible-text page (render mode 0) is served
+/// once and only once — if the fallback ever mis-fired here and merged a
+/// second pass, the phrase would duplicate.
+#[test]
+fn test_extract_regions_mem_visible_layer_unchanged() {
+    let buf = make_pdf_with_text_layer(0, None);
+    let regions = extract_text_in_regions_mem(&buf, &full_page_regions(1)).unwrap();
+    let region = &regions[0].regions[0];
+    assert_eq!(
+        region.text.matches("quick brown fox").count(),
+        1,
+        "visible text must appear exactly once, got: {:?}",
+        region.text
+    );
+    assert!(!region.needs_ocr);
+}
+
 #[test]
 fn test_extract_regions_mem_identity_h_needs_ocr() {
     let buf = std::fs::read("tests/fixtures/shinagawa_identity_h.pdf").unwrap();
@@ -3926,6 +4273,65 @@ fn encrypted_pdf_decrypts_with_correct_password() {
     );
 }
 
+/// 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");
@@ -3936,3 +4342,60 @@ fn pdf_options_debug_redacts_password() {
     );
     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/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.lock

@@ -724,7 +724,7 @@ checksum = "d6790f58c7ff633d8771f42965289203411a5e5c68388703c06e14f24770b41e"
 
 [[package]]
 name = "pdf-inspector"
-version = "0.1.7"
+version = "1.14.0"
 dependencies = [
  "env_logger",
  "include_dir",
@@ -740,7 +740,7 @@ dependencies = [
 
 [[package]]
 name = "pdf-inspector-wasm"
-version = "0.1.3"
+version = "1.14.0"
 dependencies = [
  "console_error_panic_hook",
  "js-sys",

wasm/Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "pdf-inspector-wasm"
-version = "0.1.3"
+version = "1.14.0"
 edition = "2021"
 authors = ["Firecrawl Team"]
 description = "Browser WebAssembly bindings for pdf-inspector"