review: document intended small-table boundary of prose-cell guard

Investigated the suggested row-count exemption: adding a second-prose-
cell requirement (or a non_empty_rows >= 4 exemption) resurrects
verified phantom grids in 7 corpus documents — scrambled body text and
chart/figure regions, one of which is a 4-row grid. Every observed
single-dominant-cell grid in the corpora is swallowed prose, never a
real note table, and rejection degrades gracefully to prose while
acceptance scrambles reading order. Keep the guard unconditional,
document the rationale, and pin the boundary with a test.

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

LICENSE

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

README.md

@@ -2,6 +2,7 @@
 
 [![Crates.io](https://img.shields.io/crates/v/pdf-inspector.svg)](https://crates.io/crates/pdf-inspector)
 [![npm](https://img.shields.io/npm/v/@firecrawl/pdf-inspector.svg)](https://www.npmjs.com/package/@firecrawl/pdf-inspector)
+[![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 
 Fast Rust library for PDF classification and text extraction. Detects whether a PDF is text-based or scanned, extracts text with position awareness, and converts to clean Markdown — all without OCR. Includes bindings for [Python](docs/python.md) and [Node.js](napi/README.md).
 
@@ -242,4 +243,4 @@ See [docs/debugging.md](docs/debugging.md) for `RUST_LOG` environment variable u
 
 ## License
 
-MIT
+[MIT](LICENSE)

napi/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@firecrawl/pdf-inspector",
-  "version": "1.9.11",
+  "version": "1.10.0",
   "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",

napi/src/lib.rs

@@ -83,6 +83,9 @@ pub struct TextItem {
     /// Underline detected geometrically (drawn rule/thin rect under the
     /// baseline) — PDFs carry no underline font flag.
     pub is_underline: bool,
+    /// Strikeout detected geometrically (rule crossing the glyphs at mid
+    /// x-height).
+    pub is_strikeout: bool,
     pub item_type: ItemType,
     /// URL for link items, `None` for other types.
     pub link_url: Option<String>,
@@ -294,6 +297,7 @@ pub fn extract_text_with_positions(
                     is_bold: item.is_bold,
                     is_italic: item.is_italic,
                     is_underline: item.is_underline,
+                    is_strikeout: item.is_strikeout,
                     item_type,
                     link_url,
                 }

pdf_inspector.pyi

@@ -39,6 +39,7 @@ class TextItem:
     is_bold: bool
     is_italic: bool
     is_underline: bool
+    is_strikeout: bool
     item_type: str
 
 class RegionText:

src/bin/pdf2md.rs

@@ -74,7 +74,7 @@ fn format_items_json(items: &[TextItem]) -> String {
                 _ => String::new(),
             };
             format!(
-                r#"{{"text":"{}","page":{},"x":{:.2},"y":{:.2},"width":{:.2},"height":{:.2},"font":"{}","font_size":{:.2},"is_bold":{},"is_italic":{},"is_underline":{},"item_type":"{}","mcid":{}{}}}"#,
+                r#"{{"text":"{}","page":{},"x":{:.2},"y":{:.2},"width":{:.2},"height":{:.2},"font":"{}","font_size":{:.2},"is_bold":{},"is_italic":{},"is_underline":{},"is_strikeout":{},"item_type":"{}","mcid":{}{}}}"#,
                 json_escape(&item.text),
                 item.page,
                 item.x,
@@ -86,6 +86,7 @@ fn format_items_json(items: &[TextItem]) -> String {
                 item.is_bold,
                 item.is_italic,
                 item.is_underline,
+                item.is_strikeout,
                 item_type_label(&item.item_type),
                 mcid,
                 link_url,
@@ -122,6 +123,7 @@ mod tests {
             is_bold: false,
             is_italic: true,
             is_underline: true,
+            is_strikeout: true,
             item_type: ItemType::Text,
             mcid: Some(7),
         }];

src/extractor/content_stream.rs

@@ -14,8 +14,9 @@ use lopdf::{Document, Encoding, Object, ObjectId};
 use std::collections::HashMap;
 
 use super::fonts::{
-    build_font_encodings, build_font_widths, compute_string_width_ts, extract_text_from_operand,
-    get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    build_font_encodings, build_font_widths, compute_string_width_ts, descriptor_style_flags,
+    extract_text_from_operand, get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    FontStyleCache,
 };
 use super::underline::UnderlineLine;
 use super::xobjects::{extract_form_xobject_text, get_page_xobjects, XObjectType};
@@ -106,6 +107,25 @@ fn transformed_stroke_width(
     user_width * (ndx * ndx + ndy * ndy).sqrt()
 }
 
+/// Text rise (Ts) displaces the glyph origin by (0, rise) in unscaled text
+/// space — per the rendering-matrix definition it sits left of Tm, so the
+/// offset maps through the text matrix's y column. Rise never contributes
+/// to the advance, so callers apply it only to the rendering position and
+/// keep advancing the unshifted text matrix.
+fn rise_adjusted(tm: &[f32; 6], rise: f32) -> [f32; 6] {
+    if rise == 0.0 {
+        return *tm;
+    }
+    [
+        tm[0],
+        tm[1],
+        tm[2],
+        tm[3],
+        tm[4] + rise * tm[2],
+        tm[5] + rise * tm[3],
+    ]
+}
+
 /// Returns `(page_extraction, has_gid_fonts)` where `has_gid_fonts` indicates
 /// the page uses fonts with unresolvable gid-encoded glyphs.
 pub(crate) fn extract_page_text_items(
@@ -114,6 +134,7 @@ pub(crate) fn extract_page_text_items(
     page_num: u32,
     font_cmaps: &FontCMaps,
     include_invisible: bool,
+    style_cache: &mut FontStyleCache,
 ) -> Result<(PageExtraction, bool, bool), PdfError> {
     use lopdf::content::Content;
 
@@ -153,6 +174,8 @@ pub(crate) fn extract_page_text_items(
         std::collections::HashMap::new();
     let mut inline_cmaps: std::collections::HashMap<String, crate::tounicode::CMapEntry> =
         std::collections::HashMap::new();
+    let mut font_style_flags: std::collections::HashMap<String, (bool, bool)> =
+        std::collections::HashMap::new();
     for (font_name, font_dict) in &fonts {
         let resource_name = String::from_utf8_lossy(font_name).to_string();
         if let Ok(base_font) = font_dict.get(b"BaseFont") {
@@ -161,6 +184,12 @@ pub(crate) fn extract_page_text_items(
                 font_base_names.insert(resource_name.clone(), base_name);
             }
         }
+        // Descriptor style flags rescue subset fonts whose BaseFont names
+        // are opaque tags the name heuristics can't read.
+        let style = descriptor_style_flags(doc, font_dict, style_cache);
+        if style != (false, false) {
+            font_style_flags.insert(resource_name.clone(), style);
+        }
         // Track ToUnicode object reference, with FontFile2 fallback for Identity-H/V.
         // Also handle inline ToUnicode streams.
         match font_dict.get(b"ToUnicode") {
@@ -235,6 +264,7 @@ pub(crate) fn extract_page_text_items(
         line_width: f32,
         char_spacing: f32,
         word_spacing: f32,
+        text_rise: f32,
         text_leading: f32,
         current_font: String,
         current_font_size: f32,
@@ -247,6 +277,7 @@ pub(crate) fn extract_page_text_items(
     let mut text_leading: f32 = 0.0; // TL parameter (in text-space units)
     let mut char_spacing: f32 = 0.0; // Tc parameter (extra spacing per character, unscaled)
     let mut word_spacing: f32 = 0.0; // Tw parameter (extra spacing per space char, unscaled)
+    let mut text_rise: f32 = 0.0; // Ts parameter (baseline shift for super/subscripts, unscaled)
     let mut text_matrix = [1.0f32, 0.0, 0.0, 1.0, 0.0, 0.0];
     let mut line_matrix = [1.0f32, 0.0, 0.0, 1.0, 0.0, 0.0];
     let mut in_text_block = false;
@@ -268,6 +299,10 @@ pub(crate) fn extract_page_text_items(
     let mut suppress_glyph_extraction = false;
     let mut actual_text_start_tm: Option<[f32; 6]> = None; // text matrix at BDC entry
     let mut actual_text_glyph_tm: Option<[f32; 6]> = None; // text matrix at first glyph inside BDC
+                                                           // Text rise in effect at each captured matrix — the item must render at
+                                                           // the rise of its GLYPHS, not whatever rise is set by EMC time.
+    let mut actual_text_start_rise: f32 = 0.0;
+    let mut actual_text_glyph_rise: Option<f32> = None;
     /// Get the innermost MCID from the marked content stack.
     fn current_mcid(stack: &[MarkedContentEntry]) -> Option<i64> {
         stack.iter().rev().find_map(|e| e.mcid)
@@ -284,6 +319,7 @@ pub(crate) fn extract_page_text_items(
                     line_width,
                     char_spacing,
                     word_spacing,
+                    text_rise,
                     text_leading,
                     current_font: current_font.clone(),
                     current_font_size,
@@ -297,6 +333,7 @@ pub(crate) fn extract_page_text_items(
                     line_width = saved.line_width;
                     char_spacing = saved.char_spacing;
                     word_spacing = saved.word_spacing;
+                    text_rise = saved.text_rise;
                     text_leading = saved.text_leading;
                     current_font = saved.current_font;
                     current_font_size = saved.current_font_size;
@@ -369,6 +406,12 @@ pub(crate) fn extract_page_text_items(
                     word_spacing = tw;
                 }
             }
+            "Ts" => {
+                // Set text rise (baseline shift for superscripts/subscripts)
+                if let Some(ts) = op.operands.first().and_then(get_number) {
+                    text_rise = ts;
+                }
+            }
             "Td" | "TD" => {
                 // Move text position: TLM = T(tx,ty) × TLM; Tm = TLM
                 // tx,ty are in text space — must be scaled by the text line matrix
@@ -427,6 +470,7 @@ pub(crate) fn extract_page_text_items(
                     if suppress_glyph_extraction {
                         if actual_text_glyph_tm.is_none() {
                             actual_text_glyph_tm = Some(text_matrix);
+                            actual_text_glyph_rise = Some(text_rise);
                         }
                         if let Some(w_ts) = w_ts_opt {
                             text_matrix[4] += w_ts * text_matrix[0];
@@ -456,7 +500,8 @@ pub(crate) fn extract_page_text_items(
                         &mut cmap_decisions,
                         &font_widths,
                     ) {
-                        let combined = multiply_matrices(&text_matrix, &ctm);
+                        let combined =
+                            multiply_matrices(&rise_adjusted(&text_matrix, text_rise), &ctm);
                         let rendered_size = effective_font_size(current_font_size, &combined);
                         let (x, y) = (combined[4], combined[5]);
                         if combined[0].abs() >= combined[1].abs() {
@@ -478,6 +523,10 @@ pub(crate) fn extract_page_text_items(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
@@ -487,9 +536,10 @@ pub(crate) fn extract_page_text_items(
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
                                 is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: current_mcid(&marked_content_stack),
                             });
@@ -507,6 +557,7 @@ pub(crate) fn extract_page_text_items(
                         // Capture first-glyph position for ActualText
                         if suppress_glyph_extraction && actual_text_glyph_tm.is_none() {
                             actual_text_glyph_tm = Some(text_matrix);
+                            actual_text_glyph_rise = Some(text_rise);
                         }
 
                         // Compute space threshold based on font metrics when available
@@ -627,6 +678,10 @@ pub(crate) fn extract_page_text_items(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             let scale_x = text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2];
                             for (text, start_w, end_w) in &sub_items {
                                 let offset_tm = [
@@ -637,7 +692,8 @@ pub(crate) fn extract_page_text_items(
                                     text_matrix[4] + start_w * text_matrix[0],
                                     text_matrix[5] + start_w * text_matrix[1],
                                 ];
-                                let combined = multiply_matrices(&offset_tm, &ctm);
+                                let combined =
+                                    multiply_matrices(&rise_adjusted(&offset_tm, text_rise), &ctm);
                                 let (x, y) = (combined[4], combined[5]);
                                 let width = if font_info.is_some() {
                                     ((end_w - start_w) * scale_x).abs()
@@ -653,9 +709,10 @@ pub(crate) fn extract_page_text_items(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
                                     is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: current_mcid(&marked_content_stack),
                                 });
@@ -679,6 +736,26 @@ pub(crate) fn extract_page_text_items(
                 line_matrix[4] += (-tl) * line_matrix[2];
                 line_matrix[5] += (-tl) * line_matrix[3];
                 text_matrix = line_matrix;
+                // Capture first-glyph position for ActualText AFTER the
+                // line move — the BDC-entry matrix is on the previous line.
+                if suppress_glyph_extraction && actual_text_glyph_tm.is_none() {
+                    actual_text_glyph_tm = Some(text_matrix);
+                    actual_text_glyph_rise = Some(text_rise);
+                }
+                // Advance width, as for Tj — without it the item stays
+                // zero-width and geometric underline/strikeout detection
+                // rejects it (`is_underline_candidate` needs width > 0).
+                let w_ts_opt = font_widths.get(&current_font).and_then(|fi| {
+                    op.operands.first().and_then(get_operand_bytes).map(|raw| {
+                        compute_string_width_ts(
+                            raw,
+                            fi,
+                            current_font_size,
+                            char_spacing,
+                            word_spacing,
+                        )
+                    })
+                });
                 if !((text_rendering_mode == 3 && !include_invisible)
                     || suppress_glyph_extraction
                     || op.operands.is_empty())
@@ -696,7 +773,8 @@ pub(crate) fn extract_page_text_items(
                         &font_widths,
                     ) {
                         if !text.trim().is_empty() {
-                            let combined = multiply_matrices(&text_matrix, &ctm);
+                            let combined =
+                                multiply_matrices(&rise_adjusted(&text_matrix, text_rise), &ctm);
                             if combined[0].abs() >= combined[1].abs() {
                                 rotation_votes.horizontal += 1;
                             } else {
@@ -704,28 +782,45 @@ pub(crate) fn extract_page_text_items(
                             }
                             let rendered_size = effective_font_size(current_font_size, &combined);
                             let (x, y) = (combined[4], combined[5]);
+                            let width = w_ts_opt
+                                .map(|w_ts| {
+                                    (w_ts * (text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2]))
+                                        .abs()
+                                })
+                                .unwrap_or(0.0);
                             let base_font = font_base_names
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
                                 y,
-                                width: 0.0,
+                                width,
                                 height: rendered_size,
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
                                 is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: current_mcid(&marked_content_stack),
                             });
                         }
                     }
                 }
+                // Advance regardless of visibility so later show-text
+                // operators on the same line stay positioned (as for Tj).
+                if let Some(w_ts) = w_ts_opt {
+                    text_matrix[4] += w_ts * text_matrix[0];
+                    text_matrix[5] += w_ts * text_matrix[1];
+                }
             }
             "Do" => {
                 // XObject invocation - could be an image or form
@@ -757,6 +852,7 @@ pub(crate) fn extract_page_text_items(
                                         is_bold: false,
                                         is_italic: false,
                                         is_underline: false,
+                                        is_strikeout: false,
                                         item_type: ItemType::Image,
                                         mcid: current_mcid(&marked_content_stack),
                                     });
@@ -770,6 +866,7 @@ pub(crate) fn extract_page_text_items(
                                         font_cmaps,
                                         &ctm,
                                         &mut cmap_decisions,
+                                        style_cache,
                                     );
                                     items.extend(form_items);
                                 }
@@ -810,7 +907,9 @@ pub(crate) fn extract_page_text_items(
                 if actual_text.is_some() {
                     suppress_glyph_extraction = true;
                     actual_text_start_tm = Some(text_matrix);
+                    actual_text_start_rise = text_rise;
                     actual_text_glyph_tm = None; // reset — will be captured at first Tj/TJ
+                    actual_text_glyph_rise = None;
                 }
                 marked_content_stack.push(MarkedContentEntry { actual_text, mcid });
             }
@@ -823,9 +922,11 @@ pub(crate) fn extract_page_text_items(
                         // Tj may have moved the text position to the correct line —
                         // the BDC-entry position can be on the previous line.
                         let glyph_tm = actual_text_glyph_tm.take();
+                        let glyph_rise = actual_text_glyph_rise.take();
                         let entry_tm = actual_text_start_tm.take();
                         if let Some(start_tm) = glyph_tm.or(entry_tm) {
-                            let combined = multiply_matrices(&start_tm, &ctm);
+                            let rise = glyph_rise.unwrap_or(actual_text_start_rise);
+                            let combined = multiply_matrices(&rise_adjusted(&start_tm, rise), &ctm);
                             if combined[0].abs() >= combined[1].abs() {
                                 rotation_votes.horizontal += 1;
                             } else {
@@ -842,6 +943,10 @@ pub(crate) fn extract_page_text_items(
                                     .get(&current_font)
                                     .map(|s| s.as_str())
                                     .unwrap_or(&current_font);
+                                let (desc_italic, desc_bold) = font_style_flags
+                                    .get(&current_font)
+                                    .copied()
+                                    .unwrap_or((false, false));
                                 items.push(TextItem {
                                     text: expand_ligatures(&at),
                                     x,
@@ -851,9 +956,10 @@ pub(crate) fn extract_page_text_items(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
                                     is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: entry
                                         .mcid
@@ -1376,8 +1482,15 @@ mod tests {
 
         let (doc, page_id) = simple_doc_with_content(content);
         let font_cmaps = FontCMaps::from_doc(&doc);
-        let ((items, _, _), _, _) =
-            extract_page_text_items(&doc, page_id, 1, &font_cmaps, false).unwrap();
+        let ((items, _, _), _, _) = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
         items
     }
 
@@ -1462,6 +1575,108 @@ BT /F1 12 Tf 0 1 -1 0 240 100 Tm (WORLD) Tj ET
         assert!(!world.is_underline);
     }
 
+    #[test]
+    fn quote_operator_text_carries_advance_width() {
+        // `'` (move-to-next-line-and-show-text) must retain the string's
+        // advance width like Tj — zero-width items are invisible to
+        // geometric underline/strikeout detection.
+        let content = b"BT /F1 12 Tf 12 TL 1 0 0 1 100 512 Tm (first) Tj (struck) ' ET
+1 w
+99 503 m 145 503 l S";
+
+        let items = extract_simple_items(content);
+        let struck = items.iter().find(|item| item.text == "struck").unwrap();
+
+        // 6 glyphs x 600/1000 x 12pt = 43.2pt, drawn one leading below Tm.
+        assert!((struck.width - 43.2).abs() < 0.1);
+        assert!((struck.y - 500.0).abs() < 0.1);
+        assert!(struck.is_strikeout);
+        assert!(!struck.is_underline);
+    }
+
+    #[test]
+    fn quote_operator_advances_text_matrix() {
+        // Text shown after `'` on the same line must start past the shown
+        // string: "CD" lands at x=114.4 (2 glyphs x 600/1000 x 12pt after
+        // x=100), flush against "AB", so the merge pass joins them. Without
+        // the advance "CD" overlaps "AB" at x=100 and the items stay apart.
+        let content = b"BT /F1 12 Tf 12 TL 1 0 0 1 100 512 Tm (AB) ' (CD) Tj ET";
+
+        let items = extract_simple_items(content);
+        let merged = items.iter().find(|item| item.text == "ABCD").unwrap();
+
+        assert!((merged.x - 100.0).abs() < 0.1);
+        assert!((merged.width - 28.8).abs() < 0.1);
+        assert!((merged.y - 500.0).abs() < 0.1);
+    }
+
+    #[test]
+    fn text_rise_shifts_item_baseline() {
+        // Ts displaces the glyph origin vertically without touching the
+        // advance; the next run at rise 0 must return to the original
+        // baseline and follow the raised run horizontally.
+        let content =
+            b"BT /F1 12 Tf 1 0 0 1 100 500 Tm (base) Tj 5 Ts (super) Tj 0 Ts (after) Tj ET";
+
+        let items = extract_simple_items(content);
+        let base = items.iter().find(|item| item.text == "base").unwrap();
+        let raised = items.iter().find(|item| item.text == "super").unwrap();
+        let after = items.iter().find(|item| item.text == "after").unwrap();
+
+        assert!((base.y - 500.0).abs() < 0.1);
+        assert!((raised.y - 505.0).abs() < 0.1);
+        assert!((after.y - 500.0).abs() < 0.1);
+        assert!(after.x > raised.x);
+    }
+
+    #[test]
+    fn actual_text_item_uses_glyph_rise() {
+        // The ActualText replacement item must render at the rise in
+        // effect when its glyphs were drawn — not the unshifted BDC
+        // baseline, and not whatever rise is set by EMC time.
+        let content = b"BT /F1 12 Tf 1 0 0 1 100 500 Tm \
+/Span <</ActualText (super) >> BDC 5 Ts (sup) Tj 0 Ts EMC (after) Tj ET";
+
+        let items = extract_simple_items(content);
+        let sup = items.iter().find(|item| item.text == "super").unwrap();
+        let after = items.iter().find(|item| item.text == "after").unwrap();
+
+        assert!((sup.y - 505.0).abs() < 0.1);
+        assert!((after.y - 500.0).abs() < 0.1);
+    }
+
+    #[test]
+    fn actual_text_shown_with_quote_op_uses_moved_risen_baseline() {
+        // When the tagged span's show op is `'`, the glyph position is
+        // only known AFTER its line move — falling back to the BDC-entry
+        // matrix would place the item on the previous line, unrisen.
+        let content = b"BT /F1 12 Tf 14 TL 1 0 0 1 100 500 Tm \
+/Span <</ActualText (replaced) >> BDC 3 Ts (raw) ' 0 Ts EMC ET";
+
+        let items = extract_simple_items(content);
+        let item = items.iter().find(|item| item.text == "replaced").unwrap();
+
+        // Line move: 500 - 14 = 486; rise: +3 -> 489.
+        assert!((item.y - 489.0).abs() < 0.1);
+        assert!(item.width > 0.0);
+    }
+
+    #[test]
+    fn strikeout_detected_on_risen_text() {
+        // The rule crosses the glyphs at their risen position; without the
+        // rise in item.y the strike window sits 4pt too low and misses.
+        let content = b"BT /F1 12 Tf 1 0 0 1 100 500 Tm 4 Ts (struck) Tj ET
+1 w
+99 507 m 145 507 l S";
+
+        let items = extract_simple_items(content);
+        let struck = items.iter().find(|item| item.text == "struck").unwrap();
+
+        assert!((struck.y - 504.0).abs() < 0.1);
+        assert!(struck.is_strikeout);
+        assert!(!struck.is_underline);
+    }
+
     #[test]
     fn test_skip_excessive_operations() {
         use crate::tounicode::FontCMaps;
@@ -1496,7 +1711,15 @@ BT /F1 12 Tf 0 1 -1 0 240 100 Tm (WORLD) Tj ET
         doc.add_object(catalog);
 
         let font_cmaps = FontCMaps::from_doc(&doc);
-        let result = extract_page_text_items(&doc, page_id, 1, &font_cmaps, false).unwrap();
+        let result = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
         let ((items, rects, lines), _has_gid, _coords_rotated) = result;
         assert!(items.is_empty());
         assert!(rects.is_empty());
@@ -1578,8 +1801,15 @@ 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(&doc, page_id, 1, &font_cmaps, false).unwrap();
+        let ((items, _, _), _, _) = extract_page_text_items(
+            &doc,
+            page_id,
+            1,
+            &font_cmaps,
+            false,
+            &mut FontStyleCache::new(),
+        )
+        .unwrap();
         let text = items
             .iter()
             .map(|item| item.text.as_str())

src/extractor/fonts.rs

@@ -4,7 +4,7 @@ use crate::glyph_names::glyph_to_char;
 use crate::tounicode::FontCMaps;
 use crate::types::{FontEncodingMap, FontWidthInfo, PageFontEncodings, PageFontWidths};
 use log::debug;
-use lopdf::{Document, Encoding, Object};
+use lopdf::{Document, Encoding, Object, ObjectId};
 use std::collections::HashMap;
 
 #[derive(Debug, Copy, Clone, PartialEq, Eq)]
@@ -739,6 +739,171 @@ pub(crate) fn get_font_file2_obj_num(doc: &Document, font_dict: &lopdf::Dictiona
         .map(|r| r.0)
 }
 
+/// Document-scoped memo of embedded-font style flags, keyed by the
+/// FontFile2/FontFile3 stream's object id. The same font program is
+/// referenced from every page that uses the font, and decompressing +
+/// parsing it dominates `descriptor_style_flags` — without the memo that
+/// cost repeats per page whenever the descriptor leaves a flag unset
+/// (the common case: regular fonts report neither italic nor bold).
+#[derive(Debug, Default)]
+pub(crate) struct FontStyleCache {
+    by_font_file: HashMap<ObjectId, (bool, bool)>,
+}
+
+impl FontStyleCache {
+    pub(crate) fn new() -> Self {
+        Self::default()
+    }
+}
+
+/// Style flags from the FontDescriptor, which survive subset fonts whose
+/// BaseFont names are opaque tags ("Tc1", "ABCDEF+F1") that defeat the
+/// name-based bold/italic heuristics.
+///
+/// Italic: `ItalicAngle` beyond a few degrees, or Flags bit 7 (Italic,
+/// value 64). Bold: Flags bit 19 (ForceBold, value 1<<18). The small
+/// ItalicAngle threshold skips fonts that declare a token slant.
+pub(crate) fn descriptor_style_flags(
+    doc: &Document,
+    font_dict: &lopdf::Dictionary,
+    style_cache: &mut FontStyleCache,
+) -> (bool, bool) {
+    let descriptor = font_dict
+        .get(b"FontDescriptor")
+        .ok()
+        .and_then(|obj| resolve_dict(doc, obj))
+        .or_else(|| {
+            // Type0 fonts hang the descriptor off DescendantFonts[0].
+            let desc_fonts = font_dict.get(b"DescendantFonts").ok()?;
+            let desc_fonts = resolve_array(doc, desc_fonts)?;
+            let cid_font_dict = resolve_dict(doc, desc_fonts.first()?)?;
+            resolve_dict(doc, cid_font_dict.get(b"FontDescriptor").ok()?)
+        });
+    let Some(descriptor) = descriptor else {
+        return (false, false);
+    };
+
+    let italic_angle = descriptor
+        .get(b"ItalicAngle")
+        .ok()
+        .and_then(|obj| match obj {
+            Object::Integer(i) => Some(*i as f32),
+            Object::Real(r) => Some(*r),
+            _ => None,
+        })
+        .unwrap_or(0.0);
+    let flags = descriptor
+        .get(b"Flags")
+        .ok()
+        .and_then(|obj| obj.as_i64().ok())
+        .unwrap_or(0);
+
+    let mut italic = italic_angle.abs() >= 4.0 || flags & (1 << 6) != 0;
+    let mut bold = flags & (1 << 18) != 0;
+
+    // Descriptors lie: subset generators write ItalicAngle 0 for genuinely
+    // italic faces. The embedded font file keeps the truth — OS/2
+    // fsSelection (via `Face::is_italic`) and the post table's italicAngle.
+    if !italic || !bold {
+        if let Some(ff_ref) = font_file_ref(descriptor) {
+            let (emb_italic, emb_bold) = *style_cache
+                .by_font_file
+                .entry(ff_ref)
+                .or_insert_with(|| embedded_style_flags(doc, ff_ref));
+            italic = italic || emb_italic;
+            bold = bold || emb_bold;
+        }
+    }
+    (italic, bold)
+}
+
+/// Style flags parsed from an embedded font program stream.
+fn embedded_style_flags(doc: &Document, ff_ref: ObjectId) -> (bool, bool) {
+    let Some(data) = font_file_data(doc, ff_ref) else {
+        return (false, false);
+    };
+    if let Ok(face) = ttf_parser::Face::parse(&data, 0) {
+        (
+            face.is_italic() || face.italic_angle().abs() >= 4.0,
+            face.is_bold(),
+        )
+    } else if let Some(name) = cff_font_name(&data) {
+        // FontFile3 is bare CFF (no sfnt container) — ttf_parser
+        // can't open it, but the CFF Name INDEX keeps the real
+        // PostScript name ("XXXXXX+Amplitude-LightItalic") even
+        // when the descriptor was rewritten to claim upright.
+        (
+            crate::text_utils::is_italic_font(&name),
+            crate::text_utils::is_bold_font(&name),
+        )
+    } else {
+        (false, false)
+    }
+}
+
+/// First PostScript name from a bare CFF font's Name INDEX (CFF spec §7).
+fn cff_font_name(data: &[u8]) -> Option<String> {
+    // Header: major(1) minor(1) hdrSize(1) offSize(1); major must be 1.
+    if data.len() < 4 || data[0] != 1 {
+        return None;
+    }
+    let hdr_size = data[2] as usize;
+    // Name INDEX: count(u16) offSize(u8) offsets[count+1] data
+    let count = u16::from_be_bytes([*data.get(hdr_size)?, *data.get(hdr_size + 1)?]) as usize;
+    if count == 0 {
+        return None;
+    }
+    let off_size = *data.get(hdr_size + 2)? as usize;
+    if !(1..=4).contains(&off_size) {
+        return None;
+    }
+    let read_offset = |idx: usize| -> Option<usize> {
+        let at = hdr_size + 3 + idx * off_size;
+        let bytes = data.get(at..at + off_size)?;
+        let mut v = 0usize;
+        for b in bytes {
+            v = (v << 8) | *b as usize;
+        }
+        Some(v)
+    };
+    let start = read_offset(0)?;
+    let end = read_offset(1)?;
+    if start == 0 || end < start {
+        return None;
+    }
+    // Offsets are 1-based from the byte before the object data.
+    let objects_base = hdr_size + 3 + (count + 1) * off_size - 1;
+    let name = data.get(objects_base + start..objects_base + end)?;
+    Some(String::from_utf8_lossy(name).to_string())
+}
+
+/// FontFile2/FontFile3 stream reference from a FontDescriptor.
+fn font_file_ref(descriptor: &lopdf::Dictionary) -> Option<ObjectId> {
+    descriptor
+        .get(b"FontFile2")
+        .ok()
+        .and_then(|o| o.as_reference().ok())
+        .or_else(|| {
+            descriptor
+                .get(b"FontFile3")
+                .ok()
+                .and_then(|o| o.as_reference().ok())
+        })
+}
+
+/// Decompressed embedded font program bytes.
+fn font_file_data(doc: &Document, ff_ref: ObjectId) -> Option<Vec<u8>> {
+    let stream = doc
+        .get_object(ff_ref)
+        .and_then(lopdf::Object::as_stream)
+        .ok()?;
+    Some(
+        stream
+            .decompressed_content()
+            .unwrap_or_else(|_| stream.content.clone()),
+    )
+}
+
 /// Decode text from a PDF string operand using font CMaps, encodings, and fallbacks.
 #[allow(clippy::too_many_arguments)]
 pub(crate) fn extract_text_from_operand(
@@ -1262,6 +1427,172 @@ mod tests {
         }
     }
 
+    fn doc_with_descriptor(descriptor: lopdf::Dictionary) -> (Document, lopdf::Dictionary) {
+        let mut doc = Document::with_version("1.4");
+        let desc_id = doc.add_object(descriptor);
+        let font_dict = dictionary! {
+            "Type" => "Font",
+            "Subtype" => "TrueType",
+            "BaseFont" => "Tc1",
+            "FontDescriptor" => desc_id,
+        };
+        (doc, font_dict)
+    }
+
+    #[test]
+    fn descriptor_italic_angle_sets_italic() {
+        // Subset font with an opaque BaseFont name ("Tc1") — the name
+        // heuristic sees nothing, the descriptor carries the truth.
+        let (doc, font_dict) = doc_with_descriptor(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "Tc1",
+            "ItalicAngle" => -12,
+            "Flags" => 32,
+        });
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (true, false)
+        );
+    }
+
+    #[test]
+    fn descriptor_italic_flag_bit_sets_italic() {
+        let (doc, font_dict) = doc_with_descriptor(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "Tc1",
+            "ItalicAngle" => 0,
+            "Flags" => 64, // bit 7: Italic
+        });
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (true, false)
+        );
+    }
+
+    #[test]
+    fn descriptor_force_bold_flag_sets_bold() {
+        let (doc, font_dict) = doc_with_descriptor(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "Tc1",
+            "ItalicAngle" => 0,
+            "Flags" => 1 << 18, // ForceBold
+        });
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (false, true)
+        );
+    }
+
+    #[test]
+    fn tiny_italic_angle_is_not_italic() {
+        // A token 1-degree slant is optical correction, not italic.
+        let (doc, font_dict) = doc_with_descriptor(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "Tc1",
+            "ItalicAngle" => lopdf::Object::Real(-1.0),
+            "Flags" => 32,
+        });
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (false, false)
+        );
+    }
+
+    #[test]
+    fn missing_descriptor_yields_no_flags() {
+        let doc = Document::with_version("1.4");
+        let font_dict = dictionary! { "Type" => "Font", "BaseFont" => "Tc1" };
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (false, false)
+        );
+    }
+
+    #[test]
+    fn type0_descendant_descriptor_is_resolved() {
+        let mut doc = Document::with_version("1.4");
+        let desc_id = doc.add_object(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "ABCDEF+F1",
+            "ItalicAngle" => -15,
+        });
+        let cid_id = doc.add_object(dictionary! {
+            "Type" => "Font",
+            "Subtype" => "CIDFontType2",
+            "FontDescriptor" => desc_id,
+        });
+        let font_dict = dictionary! {
+            "Type" => "Font",
+            "Subtype" => "Type0",
+            "BaseFont" => "ABCDEF+F1",
+            "DescendantFonts" => vec![lopdf::Object::Reference(cid_id)],
+        };
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (true, false)
+        );
+    }
+
+    /// Bare CFF: header + Name INDEX only — enough for `cff_font_name`.
+    fn bare_cff_with_name(name: &str) -> Vec<u8> {
+        let mut data = vec![1, 0, 4, 1]; // major, minor, hdrSize, offSize
+        data.extend_from_slice(&1u16.to_be_bytes()); // Name INDEX count
+        data.push(1); // offSize
+        data.push(1); // offset of first name
+        data.push(1 + name.len() as u8); // offset past last name
+        data.extend_from_slice(name.as_bytes());
+        data
+    }
+
+    #[test]
+    fn embedded_font_style_is_cached_by_font_file_object() {
+        use lopdf::{Object, Stream};
+
+        let mut doc = Document::with_version("1.4");
+        let ff_id = doc.add_object(Object::Stream(Stream::new(
+            dictionary! {},
+            bare_cff_with_name("ABCDEF+Test-BoldItalic"),
+        )));
+        let desc_id = doc.add_object(dictionary! {
+            "Type" => "FontDescriptor",
+            "FontName" => "ABCDEF+Test-BoldItalic",
+            "ItalicAngle" => 0,
+            "Flags" => 32,
+            "FontFile3" => ff_id,
+        });
+        let font_dict = dictionary! {
+            "Type" => "Font",
+            "Subtype" => "Type1",
+            "BaseFont" => "Tc1",
+            "FontDescriptor" => desc_id,
+        };
+
+        let mut cache = FontStyleCache::new();
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut cache),
+            (true, true)
+        );
+        assert_eq!(cache.by_font_file.len(), 1);
+
+        // Replace the font program with garbage: a repeat call must serve
+        // the memo instead of re-reading the stream — repeated per-page
+        // decompression is exactly what the cache exists to avoid.
+        doc.objects.insert(
+            ff_id,
+            Object::Stream(Stream::new(dictionary! {}, vec![0u8; 4])),
+        );
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut cache),
+            (true, true)
+        );
+        // A cold cache parses the (now garbage) stream, proving the warm
+        // call above answered from the memo.
+        assert_eq!(
+            descriptor_style_flags(&doc, &font_dict, &mut FontStyleCache::new()),
+            (false, false)
+        );
+    }
+
     #[test]
     fn compute_string_width_ts_no_tc_tw() {
         // Without Tc/Tw (both 0), width = glyph widths only

src/extractor/layout.rs

@@ -1495,6 +1495,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -1625,6 +1626,7 @@ mod tests {
                     is_bold: false,
                     is_italic: false,
                     is_underline: false,
+                    is_strikeout: false,
                     item_type: ItemType::Text,
                     mcid: None,
                 });

src/extractor/links.rs

@@ -79,6 +79,7 @@ pub fn extract_page_links(doc: &Document, page_id: ObjectId, page_num: u32) -> V
                             is_bold: false,
                             is_italic: false,
                             is_underline: false,
+                            is_strikeout: false,
                             item_type: ItemType::Link(url),
                             mcid: None,
                         });
@@ -318,6 +319,7 @@ pub(crate) fn walk_form_fields(
         is_bold: false,
         is_italic: false,
         is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::FormField,
         mcid: None,
     });

src/extractor/mod.rs

@@ -24,6 +24,7 @@ use links::{extract_form_fields, extract_page_links};
 // Re-export public types so existing `crate::extractor::X` paths keep working.
 pub use crate::text_utils::{is_bold_font, is_italic_font};
 pub use crate::types::{ItemType, TextLine};
+pub(crate) use fonts::FontStyleCache;
 pub(crate) use layout::detect_columns;
 pub use layout::group_into_lines;
 pub(crate) use layout::group_into_lines_with_thresholds;
@@ -161,6 +162,9 @@ fn extract_positioned_text_impl(
     let mut all_lines = Vec::new();
     let mut page_thresholds: PageThresholds = HashMap::new();
     let mut gid_encoded_pages: HashSet<u32> = HashSet::new();
+    // Embedded-font style flags are document-scoped: the same font program
+    // is shared across pages, so parse it once, not once per page.
+    let mut style_cache = FontStyleCache::new();
 
     // Build page ObjectId → page number map for form field extraction
     let page_id_to_num: HashMap<ObjectId, u32> =
@@ -172,8 +176,14 @@ fn extract_positioned_text_impl(
                 continue;
             }
         }
-        let ((mut items, rects, lines), has_gid_fonts, _coords_rotated) =
-            extract_page_text_items(doc, page_id, *page_num, font_cmaps, include_invisible)?;
+        let ((mut items, rects, lines), has_gid_fonts, _coords_rotated) = extract_page_text_items(
+            doc,
+            page_id,
+            *page_num,
+            font_cmaps,
+            include_invisible,
+            &mut style_cache,
+        )?;
         if has_gid_fonts {
             gid_encoded_pages.insert(*page_num);
         }
@@ -234,7 +244,10 @@ fn suppress_table_underlines(
     lines: &[PdfLine],
     page: u32,
 ) {
-    if !items.iter().any(|item| item.is_underline) {
+    if !items
+        .iter()
+        .any(|item| item.is_underline || item.is_strikeout)
+    {
         return;
     }
 
@@ -256,6 +269,7 @@ fn suppress_table_underlines(
     for index in table_item_indices {
         if let Some(item) = items.get_mut(index) {
             item.is_underline = false;
+            item.is_strikeout = false;
         }
     }
 }
@@ -576,6 +590,7 @@ pub(crate) fn merge_text_items(items: Vec<TextItem>) -> Vec<TextItem> {
                 if next.is_bold != first.is_bold
                     || next.is_italic != first.is_italic
                     || next.is_underline != first.is_underline
+                    || next.is_strikeout != first.is_strikeout
                 {
                     break;
                 }
@@ -638,6 +653,7 @@ pub(crate) fn merge_text_items(items: Vec<TextItem>) -> Vec<TextItem> {
                 is_bold: first.is_bold,
                 is_italic: first.is_italic,
                 is_underline: first.is_underline,
+                is_strikeout: first.is_strikeout,
                 item_type: first.item_type.clone(),
                 mcid: first.mcid,
             });
@@ -715,7 +731,9 @@ pub(crate) fn merge_subscript_items(items: Vec<TextItem>) -> Vec<TextItem> {
                         .chars()
                         .last()
                         .is_some_and(|c| c.is_alphabetic());
-                    if parent.font_size >= sub_threshold && ends_with_letter {
+                    let same_marks = parent.is_underline == item.is_underline
+                        && parent.is_strikeout == item.is_strikeout;
+                    if parent.font_size >= sub_threshold && ends_with_letter && same_marks {
                         let parent_right = parent.x + parent.width;
                         let gap = item.x - parent_right;
                         // Subscripts must be tightly adjacent (within ~1pt)
@@ -789,6 +807,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -991,6 +1010,7 @@ mod tests {
         items[3].y = 470.0;
         for item in &mut items {
             item.is_underline = true;
+            item.is_strikeout = true;
         }
         let lines = vec![
             make_line(100.0, 500.0, 300.0, 500.0),
@@ -1004,6 +1024,30 @@ mod tests {
         suppress_table_underlines(&mut items, &[], &lines, 1);
 
         assert!(items.iter().all(|item| !item.is_underline));
+        assert!(items.iter().all(|item| !item.is_strikeout));
+    }
+
+    #[test]
+    fn subscript_digit_with_different_marks_is_not_absorbed() {
+        // A struck-out word followed by an unmarked footnote digit: merging
+        // would widen the parent's strikeout claim over the digit (and the
+        // reverse would drop the digit's own mark). Style boundaries break
+        // the merge, as in merge_text_items.
+        let mut word = make_merge_item("word", 100.0, 24.0);
+        word.font_size = 10.0;
+        word.is_strikeout = true;
+        let mut digit = make_merge_item("2", 124.5, 4.0);
+        digit.font_size = 6.0;
+        digit.y = word.y + 3.0;
+
+        let merged = merge_subscript_items(vec![word.clone(), digit.clone()]);
+        assert_eq!(merged.len(), 2);
+
+        // Same marks still merge (footnote ref inside the strike).
+        digit.is_strikeout = true;
+        let merged = merge_subscript_items(vec![word, digit]);
+        assert_eq!(merged.len(), 1);
+        assert!(merged[0].text.starts_with("word"));
     }
 
     #[test]
@@ -1021,6 +1065,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1036,6 +1081,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1051,6 +1097,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1106,6 +1153,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1121,6 +1169,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1136,6 +1185,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1162,6 +1212,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1177,6 +1228,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1192,6 +1244,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1220,6 +1273,7 @@ mod tests {
                 is_bold: true,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             }
@@ -1255,6 +1309,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             }
@@ -1291,6 +1346,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1306,6 +1362,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1321,6 +1378,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1344,6 +1402,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -1457,6 +1516,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1472,6 +1532,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1497,6 +1558,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1512,6 +1574,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             },
@@ -1553,6 +1616,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             }],
@@ -1598,6 +1662,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             }],
@@ -1643,6 +1708,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             }],
@@ -1681,6 +1747,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/extractor/underline.rs

@@ -231,8 +231,27 @@ fn rule_matches_item(rule: &Rule, item: &TextItem) -> bool {
     overlap >= min_overlap
 }
 
+/// Strikeout window: a rule crossing the glyphs. Strikethroughs sit at
+/// roughly 20-35% of the em above the baseline (about half the x-height);
+/// accept a band well inside the glyph body so baseline underlines and
+/// overlines never qualify.
+fn rule_strikes_item(rule: &Rule, item: &TextItem) -> bool {
+    let y_min = item.y + item.font_size * 0.12;
+    let y_max = item.y + item.font_size * 0.55;
+    if rule.y < y_min || rule.y > y_max {
+        return false;
+    }
+
+    let ix1 = item.x;
+    let ix2 = item.x + item.width;
+    let min_overlap = item.width * MIN_X_OVERLAP;
+    let overlap = rule.x2.min(ix2) - rule.x1.max(ix1);
+    overlap >= min_overlap
+}
+
 /// Mark `is_underline` on text items that have a horizontal rule just
-/// below their baseline. `items`, `rects`, and `lines` are a single
+/// below their baseline, and `is_strikeout` on items whose glyphs a rule
+/// crosses at mid x-height. `items`, `rects`, and `lines` are a single
 /// page's extraction output (all in PDF coordinates, y-up, where
 /// `TextItem::y` is the text baseline).
 pub(crate) fn mark_underlined_items(
@@ -258,6 +277,11 @@ pub(crate) fn mark_underlined_items(
             }
             if rule_matches_item(rule, item) {
                 item.is_underline = true;
+            }
+            if rule_strikes_item(rule, item) {
+                item.is_strikeout = true;
+            }
+            if item.is_underline && item.is_strikeout {
                 break;
             }
         }
@@ -282,6 +306,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -358,6 +383,44 @@ mod tests {
         assert!(!items[0].is_underline);
     }
 
+    #[test]
+    fn mid_glyph_rule_marks_strikeout_not_underline() {
+        // Rule at ~30% of the em above the baseline crosses the glyphs.
+        let mut items = vec![item("struck out", 100.0, 500.0, 60.0, 10.0)];
+        let lines = vec![hline(99.0, 161.0, 503.0)];
+        mark_underlined_items(&mut items, &[], &lines, 1);
+        assert!(items[0].is_strikeout);
+        assert!(!items[0].is_underline);
+    }
+
+    #[test]
+    fn baseline_rule_marks_underline_not_strikeout() {
+        let mut items = vec![item("underlined", 100.0, 500.0, 60.0, 10.0)];
+        let lines = vec![hline(99.0, 161.0, 498.5)];
+        mark_underlined_items(&mut items, &[], &lines, 1);
+        assert!(items[0].is_underline);
+        assert!(!items[0].is_strikeout);
+    }
+
+    #[test]
+    fn overline_is_neither_underline_nor_strikeout() {
+        // Rule just above the cap height (overline / next line's rule).
+        let mut items = vec![item("text", 100.0, 500.0, 60.0, 10.0)];
+        let lines = vec![hline(99.0, 161.0, 507.0)];
+        mark_underlined_items(&mut items, &[], &lines, 1);
+        assert!(!items[0].is_underline);
+        assert!(!items[0].is_strikeout);
+    }
+
+    #[test]
+    fn thin_filled_rect_at_mid_glyph_marks_strikeout() {
+        let mut items = vec![item("struck out", 100.0, 500.0, 60.0, 10.0)];
+        let rects = vec![thin_rect(100.0, 502.6, 60.0)];
+        mark_underlined_items(&mut items, &rects, &[], 1);
+        assert!(items[0].is_strikeout);
+        assert!(!items[0].is_underline);
+    }
+
     #[test]
     fn line_above_baseline_is_not_an_underline() {
         // Strikethrough / overline geometry must not mark.

src/extractor/xobjects.rs

@@ -1,5 +1,6 @@
 //! Form XObject and image XObject extraction.
 
+use super::fonts::descriptor_style_flags;
 use crate::text_utils::{effective_font_size, expand_ligatures, is_bold_font, is_italic_font};
 use crate::tounicode::FontCMaps;
 use crate::types::{ItemType, TextItem};
@@ -8,7 +9,7 @@ use std::collections::HashMap;
 
 use super::fonts::{
     build_font_encodings, build_font_widths, compute_string_width_ts, extract_text_from_operand,
-    get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache,
+    get_font_file2_obj_num, get_operand_bytes, CMapDecisionCache, FontStyleCache,
 };
 use super::{get_number, image_bbox_from_ctm, multiply_matrices};
 
@@ -114,6 +115,7 @@ pub(crate) fn extract_form_xobject_text(
     font_cmaps: &FontCMaps,
     parent_ctm: &[f32; 6],
     cmap_decisions: &mut CMapDecisionCache,
+    style_cache: &mut FontStyleCache,
 ) -> Vec<TextItem> {
     extract_form_xobject_text_inner(
         doc,
@@ -122,10 +124,12 @@ pub(crate) fn extract_form_xobject_text(
         font_cmaps,
         parent_ctm,
         cmap_decisions,
+        style_cache,
         0,
     )
 }
 
+#[allow(clippy::too_many_arguments)]
 fn extract_form_xobject_text_inner(
     doc: &Document,
     form_id: ObjectId,
@@ -133,6 +137,7 @@ fn extract_form_xobject_text_inner(
     font_cmaps: &FontCMaps,
     parent_ctm: &[f32; 6],
     cmap_decisions: &mut CMapDecisionCache,
+    style_cache: &mut FontStyleCache,
     depth: u8,
 ) -> Vec<TextItem> {
     use lopdf::content::Content;
@@ -167,6 +172,7 @@ fn extract_form_xobject_text_inner(
     let mut font_tounicode_refs: HashMap<String, u32> = HashMap::new();
     let mut inline_cmaps: HashMap<String, crate::tounicode::CMapEntry> = HashMap::new();
 
+    let mut font_style_flags: HashMap<String, (bool, bool)> = HashMap::new();
     for (font_name, font_dict) in &form_fonts {
         let resource_name = String::from_utf8_lossy(font_name).to_string();
         if let Ok(base_font) = font_dict.get(b"BaseFont") {
@@ -175,6 +181,10 @@ fn extract_form_xobject_text_inner(
                 font_base_names.insert(resource_name.clone(), base_name);
             }
         }
+        let style = descriptor_style_flags(doc, font_dict, style_cache);
+        if style != (false, false) {
+            font_style_flags.insert(resource_name.clone(), style);
+        }
         match font_dict.get(b"ToUnicode") {
             Ok(tounicode) => {
                 if let Ok(obj_ref) = tounicode.as_reference() {
@@ -272,6 +282,7 @@ fn extract_form_xobject_text_inner(
                                         font_cmaps,
                                         &ctm,
                                         cmap_decisions,
+                                        style_cache,
                                         depth + 1,
                                     );
                                     items.extend(nested_items);
@@ -295,6 +306,7 @@ fn extract_form_xobject_text_inner(
                                     is_bold: false,
                                     is_italic: false,
                                     is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Image,
                                     mcid: None,
                                 });
@@ -428,6 +440,10 @@ fn extract_form_xobject_text_inner(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             items.push(TextItem {
                                 text: expand_ligatures(&text),
                                 x,
@@ -437,9 +453,10 @@ fn extract_form_xobject_text_inner(
                                 font: current_font.clone(),
                                 font_size: rendered_size,
                                 page: page_num,
-                                is_bold: is_bold_font(base_font),
-                                is_italic: is_italic_font(base_font),
+                                is_bold: is_bold_font(base_font) || desc_bold,
+                                is_italic: is_italic_font(base_font) || desc_italic,
                                 is_underline: false,
+                                is_strikeout: false,
                                 item_type: ItemType::Text,
                                 mcid: None,
                             });
@@ -560,6 +577,10 @@ fn extract_form_xobject_text_inner(
                                 .get(&current_font)
                                 .map(|s| s.as_str())
                                 .unwrap_or(&current_font);
+                            let (desc_italic, desc_bold) = font_style_flags
+                                .get(&current_font)
+                                .copied()
+                                .unwrap_or((false, false));
                             let scale_x = text_matrix[0] * ctm[0] + text_matrix[1] * ctm[2];
                             for (text, start_w, end_w) in &sub_items {
                                 let offset_tm = [
@@ -586,9 +607,10 @@ fn extract_form_xobject_text_inner(
                                     font: current_font.clone(),
                                     font_size: rendered_size,
                                     page: page_num,
-                                    is_bold: is_bold_font(base_font),
-                                    is_italic: is_italic_font(base_font),
+                                    is_bold: is_bold_font(base_font) || desc_bold,
+                                    is_italic: is_italic_font(base_font) || desc_italic,
                                     is_underline: false,
+                                    is_strikeout: false,
                                     item_type: ItemType::Text,
                                     mcid: None,
                                 });

src/lib.rs

@@ -579,6 +579,7 @@ pub fn extract_text_in_regions_mem(
     let mut gid_pages: HashSet<u32> = HashSet::new();
     let mut page_thresholds: HashMap<u32, f32> = HashMap::new();
     let mut rotated_pages: HashSet<u32> = HashSet::new();
+    let mut style_cache = extractor::FontStyleCache::new();
 
     for (page_num, &page_id) in pages.iter() {
         if !needed_pages.contains(page_num) {
@@ -597,6 +598,7 @@ pub fn extract_text_in_regions_mem(
                 *page_num,
                 &font_cmaps,
                 false,
+                &mut style_cache,
             )?;
         let threshold = text_utils::fix_letterspaced_items(&mut items);
         if threshold > 0.10 {
@@ -699,6 +701,7 @@ pub fn extract_tables_in_regions_mem(
     let mut gid_pages: HashSet<u32> = HashSet::new();
     let mut page_thresholds: HashMap<u32, f32> = HashMap::new();
     let mut rotated_pages: HashSet<u32> = HashSet::new();
+    let mut style_cache = extractor::FontStyleCache::new();
 
     for (page_num, &page_id) in pages.iter() {
         if !needed_pages.contains(page_num) {
@@ -714,6 +717,7 @@ pub fn extract_tables_in_regions_mem(
                 *page_num,
                 &font_cmaps,
                 false,
+                &mut style_cache,
             )?;
         let threshold = text_utils::fix_letterspaced_items(&mut items);
         if threshold > 0.10 {
@@ -1019,6 +1023,7 @@ pub fn detect_vector_grid_in_region_mem(
             page_1idx,
             &font_cmaps,
             false,
+            &mut extractor::FontStyleCache::new(),
         )?;
     text_utils::fix_letterspaced_items(&mut items);
 
@@ -1205,8 +1210,15 @@ 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).unwrap();
+        let ((items, rects, _lines), _has_gid, _rotated) = 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");
@@ -1240,8 +1252,15 @@ 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).unwrap();
+        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 (rect_tables, _) = detect_tables_from_rects(&items, &rects, page_num);
         rect_tables
@@ -1966,6 +1985,7 @@ pub fn extract_tables_with_structure_cells_mem(
     let mut page_heights: HashMap<u32, f32> = HashMap::new();
     let mut page_thresholds: HashMap<u32, f32> = HashMap::new();
     let mut rotated_pages: HashSet<u32> = HashSet::new();
+    let mut style_cache = extractor::FontStyleCache::new();
 
     for (page_num, &page_id) in pages.iter() {
         if !needed_pages.contains(page_num) {
@@ -1981,6 +2001,7 @@ pub fn extract_tables_with_structure_cells_mem(
                 *page_num,
                 &font_cmaps,
                 false,
+                &mut style_cache,
             )?;
         let threshold = text_utils::fix_letterspaced_items(&mut items);
         if threshold > 0.10 {
@@ -2782,6 +2803,7 @@ fn detect_tsr_quality_issue(
             page_1idx,
             &font_cmaps,
             false,
+            &mut extractor::FontStyleCache::new(),
         )?;
     let adaptive_threshold = text_utils::fix_letterspaced_items(&mut items);
     let coords = if coords_rotated {
@@ -5056,6 +5078,7 @@ mod text_cluster_column_undercount_tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -5331,6 +5354,7 @@ mod table_candidate_selection_tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -6078,6 +6102,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/markdown/analysis.rs

@@ -130,6 +130,29 @@ pub(crate) fn has_dot_leaders(text: &str) -> bool {
     dot_groups >= 2
 }
 
+/// Detect a table-of-contents entry: a line ending in a page number preceded by
+/// a dot-leader group (e.g. "Measurement Lab worksheet ... 3"). `has_dot_leaders`
+/// misses single-group leaders ("..."), but a trailing "<dots> <number>" is a
+/// strong TOC signal on its own. Such lines must never be promoted to headings.
+pub(crate) fn is_toc_entry_line(text: &str) -> bool {
+    let trimmed = text.trim_end();
+    let digits = trimmed
+        .chars()
+        .rev()
+        .take_while(|c| c.is_ascii_digit())
+        .count();
+    if digits == 0 || digits > 4 {
+        return false;
+    }
+    let before_number = trimmed[..trimmed.len() - digits].trim_end();
+    let dots = before_number
+        .chars()
+        .rev()
+        .take_while(|c| *c == '.')
+        .count();
+    dots >= 3
+}
+
 /// Compute the Y-gap threshold for paragraph break detection.
 ///
 /// Instead of using a fixed multiple of base_size (which fails for double-spaced
@@ -320,3 +343,28 @@ pub(crate) fn detect_header_level(
         Some(4)
     }
 }
+
+#[cfg(test)]
+mod tests {
+    use super::*;
+
+    #[test]
+    fn toc_entry_with_single_dot_group() {
+        assert!(is_toc_entry_line("Measurement Lab worksheet ... 3"));
+        assert!(is_toc_entry_line("Results ........ 12"));
+        assert!(is_toc_entry_line("Appendix B...42"));
+    }
+
+    #[test]
+    fn non_toc_lines_pass() {
+        assert!(!is_toc_entry_line(
+            "6.2. Expectations for Re-Hiring Employees"
+        ));
+        assert!(!is_toc_entry_line("What happened in 2020"));
+        assert!(!is_toc_entry_line("IMPLEMENTATION"));
+        // Ellipsis without a trailing page number
+        assert!(!is_toc_entry_line("and so it goes ..."));
+        // Long numbers are data, not page refs
+        assert!(!is_toc_entry_line("ISBN ... 97814"));
+    }
+}

src/markdown/convert.rs

@@ -7,7 +7,7 @@ use crate::types::TextLine;
 
 use super::analysis::{
     bold_heading_level, calculate_font_stats, compute_heading_tiers, compute_paragraph_threshold,
-    detect_header_level, font_size_rarity, has_dot_leaders,
+    detect_header_level, font_size_rarity, has_dot_leaders, is_toc_entry_line,
 };
 use super::classify::{
     format_list_item, is_caption_line, is_list_item, is_monospace_font, starts_with_bullet_marker,
@@ -703,6 +703,7 @@ pub(super) fn to_markdown_from_lines_with_tables_and_images(
             && plain_trimmed.len() > 3
             && plain_trimmed.split_whitespace().count() <= 15
             && !starts_with_bullet_marker(plain_trimmed)
+            && !is_toc_entry_line(plain_trimmed)
         {
             let line_font_size = line.items.first().map(|i| i.font_size).unwrap_or(base_size);
             detect_header_level(line_font_size, base_size, &heading_tiers).or_else(|| {
@@ -738,7 +739,11 @@ pub(super) fn to_markdown_from_lines_with_tables_and_images(
                 // paragraph continuity and minor font-size variation
                 // inflates rarity scores.
                 let has_strong_signal = all_bold || isolated || (rarity >= 0.97 && word_count <= 8);
-                if score >= 0.5 && standalone && word_count >= 2 && has_strong_signal {
+                // Single-word headings ("IMPLEMENTATION", "CONTENTS") are common;
+                // accept them only with the strongest signal combination.
+                let enough_words =
+                    word_count >= 2 || (all_bold && isolated && plain_trimmed.len() >= 4);
+                if score >= 0.5 && standalone && enough_words && has_strong_signal {
                     Some(bold_heading_level(&heading_tiers))
                 } else {
                     None
@@ -1037,6 +1042,7 @@ pub fn to_markdown_from_lines(lines: Vec<TextLine>, options: MarkdownOptions) ->
         if options.detect_headers
             && plain_trimmed.len() > 3
             && plain_trimmed.split_whitespace().count() <= 15
+            && !is_toc_entry_line(plain_trimmed)
         {
             let line_font_size = line.items.first().map(|i| i.font_size).unwrap_or(base_size);
             if let Some(header_level) =
@@ -1059,7 +1065,9 @@ pub fn to_markdown_from_lines(lines: Vec<TextLine>, options: MarkdownOptions) ->
                         + if all_bold { 0.3 } else { 0.0 }
                         + if standalone { 0.2 } else { 0.0 }
                         + if isolated { 0.3 } else { 0.0 };
-                    if score >= 0.5 && standalone && word_count >= 2 {
+                    let enough_words =
+                        word_count >= 2 || (all_bold && isolated && plain_trimmed.len() >= 4);
+                    if score >= 0.5 && standalone && enough_words {
                         return Some(bold_heading_level(&heading_tiers));
                     }
                     None
@@ -1189,6 +1197,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: crate::types::ItemType::Text,
             mcid,
         }

src/markdown/mod.rs

@@ -1221,6 +1221,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: crate::types::ItemType::Text,
             mcid: None,
         }

src/markdown/preprocess.rs

@@ -543,6 +543,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid,
         }

src/python.rs

@@ -268,6 +268,8 @@ pub struct PyTextItem {
     #[pyo3(get)]
     pub is_underline: bool,
     #[pyo3(get)]
+    pub is_strikeout: bool,
+    #[pyo3(get)]
     pub item_type: String,
 }
 
@@ -352,6 +354,7 @@ fn convert_text_items(items: Vec<crate::TextItem>) -> Vec<PyTextItem> {
             is_bold: item.is_bold,
             is_italic: item.is_italic,
             is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item_type_str(&item.item_type),
         })
         .collect()

src/tables/detect_heuristic.rs

@@ -105,6 +105,7 @@ pub(crate) fn merge_adjacent_items(items: &[TextItem]) -> (Vec<TextItem>, Vec<Ve
                 is_bold: first_item.is_bold,
                 is_italic: first_item.is_italic,
                 is_underline: first_item.is_underline,
+                is_strikeout: first_item.is_strikeout,
                 item_type: first_item.item_type.clone(),
                 mcid: first_item.mcid,
             });

src/tables/detect_lines.rs

@@ -394,6 +394,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/tables/detect_rects.rs

@@ -1477,6 +1477,10 @@ fn detect_row_stripe_table(
         debug!("  row-stripe rejected: sparse outline/prose continuation shape");
         return None;
     }
+    if has_dominant_prose_cell(&cells) {
+        debug!("  row-stripe rejected: dominant prose cell (chart/figure region over body text)");
+        return None;
+    }
 
     let column_centers: Vec<f32> = (0..num_cols)
         .map(|c| (col_edges[c] + col_edges[c + 1]) / 2.0)
@@ -1495,6 +1499,34 @@ fn detect_row_stripe_table(
     Some(Table::new(column_centers, row_centers, cells, item_indices))
 }
 
+/// Detect a grid that swallowed body text instead of tabular data.
+///
+/// Charts (bar graphs, axis gridlines) emit fields of drawing rects that can
+/// pass the row-stripe shape test; the resulting "table" then captures the
+/// page's prose. The signature: one cell holds an entire paragraph — ≥60 words
+/// AND at least a third of all words in the table.
+///
+/// There is deliberately no row-count exemption. A small table whose single
+/// long cell dominates its word count is indistinguishable by content from a
+/// phantom grid over body text, and across the regression corpora every such
+/// grid observed has been swallowed prose, never a real note table. The costs
+/// are also asymmetric: rejecting a real table degrades it to readable prose,
+/// while accepting a phantom scrambles the page into Y-interleaved cells.
+/// Larger legitimate tables are safe because the one-third-of-total threshold
+/// scales with table size.
+fn has_dominant_prose_cell(cells: &[Vec<String>]) -> bool {
+    let mut total_words = 0usize;
+    let mut max_cell_words = 0usize;
+    for row in cells {
+        for cell in row {
+            let words = cell.split_whitespace().count();
+            total_words += words;
+            max_cell_words = max_cell_words.max(words);
+        }
+    }
+    max_cell_words >= 60 && max_cell_words * 3 >= total_words
+}
+
 fn row_stripe_is_sparse_prose_outline(cells: &[Vec<String>]) -> bool {
     let Some(num_cols) = cells.first().map(|row| row.len()) else {
         return false;
@@ -2294,6 +2326,12 @@ fn detect_merged_cluster_table(
         );
         return None;
     }
+    if has_dominant_prose_cell(&cells) {
+        debug!(
+            "  merged-cluster rejected: dominant prose cell (chart/figure region over body text)"
+        );
+        return None;
+    }
 
     // No empty columns
     for col in 0..num_cols {
@@ -2392,11 +2430,95 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
     }
 
+    // --- has_dominant_prose_cell ---
+
+    fn cells_of(rows: &[&[&str]]) -> Vec<Vec<String>> {
+        rows.iter()
+            .map(|r| r.iter().map(|c| c.to_string()).collect())
+            .collect()
+    }
+
+    #[test]
+    fn dominant_prose_cell_rejects_swallowed_paragraph() {
+        // Two cells hold paragraphs (the shape every observed phantom grid
+        // has: swallowed body text spans multiple cells), rest are chart labels
+        let para = ["word"; 70].join(" ");
+        let para2 = ["word"; 35].join(" ");
+        let cells = cells_of(&[
+            &[para.as_str(), "81", "76"],
+            &[para2.as_str(), "56", "9"],
+            &["2019", "2020", ""],
+        ]);
+        assert!(has_dominant_prose_cell(&cells));
+    }
+
+    #[test]
+    fn dominant_prose_cell_rejects_small_table_dominated_by_one_cell() {
+        // Boundary case, documented as INTENDED: a small grid whose single
+        // long cell dominates the word count is rejected even at 4+ rows.
+        // By content alone this shape is indistinguishable from a phantom
+        // grid over body text, and every observed instance in the regression
+        // corpora was swallowed prose (chart/figure regions), not a real
+        // note table. Rejection degrades gracefully — the text is still
+        // extracted as prose — while accepting a phantom scrambles reading
+        // order.
+        let note = ["word"; 70].join(" ");
+        let cells = cells_of(&[
+            &["Purpose", note.as_str()],
+            &["Owner", "Facilities team"],
+            &["Date", "2024-06-01"],
+            &["Status", "Active"],
+        ]);
+        assert!(has_dominant_prose_cell(&cells));
+    }
+
+    #[test]
+    fn dominant_prose_cell_allows_description_column() {
+        // Long-ish description cells, but text is spread across the table
+        let desc = ["word"; 25].join(" ");
+        let cells = cells_of(&[
+            &["Item A", desc.as_str(), "100"],
+            &["Item B", desc.as_str(), "200"],
+            &["Item C", desc.as_str(), "300"],
+            &["Item D", desc.as_str(), "400"],
+        ]);
+        assert!(!has_dominant_prose_cell(&cells));
+    }
+
+    #[test]
+    fn dominant_prose_cell_allows_short_tables() {
+        let cells = cells_of(&[&["Name", "Value"], &["Total", "42"]]);
+        assert!(!has_dominant_prose_cell(&cells));
+    }
+
+    #[test]
+    fn dominant_prose_cell_allows_data_table_with_long_note() {
+        // A real 4+ row table with one verbose remark cell: the note is ≥60
+        // words but the table's other content carries more than 2× its word
+        // count, so concentration stays below the 1/3 threshold. The
+        // denominator scales with table size — this is what keeps large
+        // legitimate tables safe where a bare length cap would not.
+        let note = ["word"; 60].join(" ");
+        let row_text = ["data"; 12].join(" ");
+        let mut rows: Vec<Vec<String>> = (0..11)
+            .map(|i| {
+                vec![
+                    format!("Item {i}"),
+                    row_text.clone(),
+                    format!("{}", i * 100),
+                ]
+            })
+            .collect();
+        rows.push(vec!["Note".into(), note, String::new()]);
+        assert!(!has_dominant_prose_cell(&rows));
+    }
+
     // --- rects_overlap ---
 
     #[test]
@@ -3366,6 +3488,7 @@ mod tests {
                     is_bold: false,
                     is_italic: false,
                     is_underline: false,
+                    is_strikeout: false,
                     item_type: crate::types::ItemType::Text,
                     mcid: None,
                 });
@@ -3676,6 +3799,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: crate::types::ItemType::Text,
                 mcid: None,
             });

src/tables/detect_struct.rs

@@ -587,6 +587,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid,
         }

src/tables/financial.rs

@@ -109,6 +109,7 @@ pub(crate) fn try_split_financial_item(item: &TextItem) -> Option<Vec<TextItem>>
             is_bold: item.is_bold,
             is_italic: item.is_italic,
             is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item.item_type.clone(),
             mcid: item.mcid,
         });

src/tables/grid.rs

@@ -521,6 +521,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -886,6 +887,7 @@ mod tests {
                         is_bold: false,
                         is_italic: false,
                         is_underline: false,
+                        is_strikeout: false,
                         item_type: ItemType::Text,
                         mcid: None,
                         page: 1,
@@ -923,6 +925,7 @@ mod tests {
                         is_bold: false,
                         is_italic: false,
                         is_underline: false,
+                        is_strikeout: false,
                         item_type: ItemType::Text,
                         mcid: None,
                         page: 1,

src/tables/mod.rs

@@ -236,6 +236,7 @@ fn split_merged_numbers(item: &TextItem, col_boundaries: &[f32]) -> Vec<TextItem
             is_bold: item.is_bold,
             is_italic: item.is_italic,
             is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item.item_type.clone(),
             mcid: item.mcid,
         });
@@ -257,6 +258,7 @@ fn split_merged_numbers(item: &TextItem, col_boundaries: &[f32]) -> Vec<TextItem
             is_bold: item.is_bold,
             is_italic: item.is_italic,
             is_underline: item.is_underline,
+            is_strikeout: item.is_strikeout,
             item_type: item.item_type.clone(),
             mcid: item.mcid,
         });
@@ -1432,6 +1434,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -1450,6 +1453,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/text_utils.rs

@@ -93,6 +93,9 @@ pub fn is_bold_font(font_name: &str) -> bool {
         || lower.contains("extrabold")
         || lower.contains("ultrabold")
         || lower.contains("medium") && !lower.contains("mediumitalic") // Some fonts use Medium for semi-bold
+        // URW Type 1 fonts abbreviate Medium as "Medi" (e.g. NimbusRomNo9L-Medi,
+        // the Times-Bold substitute in LaTeX documents; -MediItal is bold italic).
+        || lower.contains("-medi") && !lower.contains("mediumital")
 }
 
 /// Detect if a font name indicates italic/oblique style
@@ -762,6 +765,17 @@ mod tests {
     use super::*;
     use crate::types::ItemType;
 
+    #[test]
+    fn bold_font_urw_medi_abbreviation() {
+        // URW Type 1 fonts (LaTeX default Times) abbreviate Medium as "Medi"
+        assert!(is_bold_font("NROFIU+NimbusRomNo9L-Medi"));
+        assert!(is_bold_font("NimbusRomNo9L-MediItal"));
+        assert!(!is_bold_font("DSSZWN+NimbusRomNo9L-Regu"));
+        assert!(!is_bold_font("NimbusRomNo9L-ReguItal"));
+        // Medium-Italic exclusion still holds
+        assert!(!is_bold_font("Foo-MediumItalic"));
+    }
+
     #[test]
     fn strip_soft_hyphen() {
         assert_eq!(expand_ligatures("con\u{00AD}tent"), "content");
@@ -884,6 +898,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }
@@ -1004,6 +1019,7 @@ mod tests {
                 is_bold: false,
                 is_italic: false,
                 is_underline: false,
+                is_strikeout: false,
                 item_type: ItemType::Text,
                 mcid: None,
             });
@@ -1081,6 +1097,7 @@ mod tests {
             is_bold: false,
             is_italic: false,
             is_underline: false,
+            is_strikeout: false,
             item_type: ItemType::Text,
             mcid: None,
         }

src/types.rs

@@ -120,6 +120,10 @@ pub struct TextItem {
     /// baseline — PDFs have no underline font flag, so this is detected
     /// geometrically after extraction; see `extractor::underline`).
     pub is_underline: bool,
+    /// Whether the text is struck out (drawn rule/thin rect crossing the
+    /// glyphs at mid x-height). Same geometric detection as underline,
+    /// different vertical window; see `extractor::underline`.
+    pub is_strikeout: bool,
     /// Type of item (text, image, link)
     pub item_type: ItemType,
     /// Marked Content ID from the content stream's BDC/BMC operator.

tests/integration_tests.rs

@@ -105,6 +105,7 @@ fn make_text_item(text: &str, x: f32, y: f32, font_size: f32, page: u32) -> Text
         is_bold: false,
         is_italic: false,
         is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::Text,
         mcid: None,
     }
@@ -131,6 +132,7 @@ fn make_text_item_with_font(
         is_bold: is_bold_font(font),
         is_italic: is_italic_font(font),
         is_underline: false,
+        is_strikeout: false,
         item_type: ItemType::Text,
         mcid: None,
     }