@effortlessmotion/pdf-to-md 0.1.8 → 0.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (108) hide show
  1. package/README.md +5 -5
  2. package/dist/cjs/analyze.d.cts +3 -1
  3. package/dist/cjs/analyze.d.ts +3 -1
  4. package/dist/cjs/analyze.js +38 -100
  5. package/dist/cjs/analyze.js.map +1 -1
  6. package/dist/cjs/audit.d.cts +6 -5
  7. package/dist/cjs/audit.d.ts +6 -5
  8. package/dist/cjs/audit.js +368 -60
  9. package/dist/cjs/audit.js.map +1 -1
  10. package/dist/cjs/cli/audit.js.map +1 -1
  11. package/dist/cjs/cli/extract.js +13 -48
  12. package/dist/cjs/cli/extract.js.map +1 -1
  13. package/dist/cjs/cli/index.js.map +1 -1
  14. package/dist/cjs/cli/render.js.map +1 -1
  15. package/dist/cjs/cli/types.js.map +1 -1
  16. package/dist/cjs/cli.js.map +1 -1
  17. package/dist/cjs/collect.js +91 -35
  18. package/dist/cjs/collect.js.map +1 -1
  19. package/dist/cjs/dict.js.map +1 -1
  20. package/dist/cjs/emit.js.map +1 -1
  21. package/dist/cjs/engines/tesseract.js.map +1 -1
  22. package/dist/cjs/errors.d.cts +2 -0
  23. package/dist/cjs/errors.d.ts +2 -0
  24. package/dist/cjs/errors.js.map +1 -1
  25. package/dist/cjs/extract.d.cts +20 -9
  26. package/dist/cjs/extract.d.ts +20 -9
  27. package/dist/cjs/extract.js +333 -106
  28. package/dist/cjs/extract.js.map +1 -1
  29. package/dist/cjs/geometry.js.map +1 -1
  30. package/dist/cjs/index.js.map +1 -1
  31. package/dist/cjs/lines.js +21 -10
  32. package/dist/cjs/lines.js.map +1 -1
  33. package/dist/cjs/markdown.js +140 -54
  34. package/dist/cjs/markdown.js.map +1 -1
  35. package/dist/cjs/page-marker.d.cts +1 -1
  36. package/dist/cjs/page-marker.d.ts +1 -1
  37. package/dist/cjs/page-marker.js +39 -8
  38. package/dist/cjs/page-marker.js.map +1 -1
  39. package/dist/cjs/pdf-open.js.map +1 -1
  40. package/dist/cjs/raster.js.map +1 -1
  41. package/dist/cjs/render-analysis.d.cts +4 -2
  42. package/dist/cjs/render-analysis.d.ts +4 -2
  43. package/dist/cjs/render-analysis.js +5 -3
  44. package/dist/cjs/render-analysis.js.map +1 -1
  45. package/dist/cjs/report.d.cts +2 -2
  46. package/dist/cjs/report.d.ts +2 -2
  47. package/dist/cjs/report.js +27 -8
  48. package/dist/cjs/report.js.map +1 -1
  49. package/dist/cjs/resolve.js.map +1 -1
  50. package/dist/cjs/text-analysis.d.cts +12 -0
  51. package/dist/cjs/text-analysis.d.ts +12 -0
  52. package/dist/cjs/text-analysis.js +175 -0
  53. package/dist/cjs/text-analysis.js.map +1 -0
  54. package/dist/cjs/types.d.cts +40 -3
  55. package/dist/cjs/types.d.ts +40 -3
  56. package/dist/cjs/types.js +2 -3
  57. package/dist/cjs/types.js.map +1 -1
  58. package/dist/cjs/validate.js.map +1 -1
  59. package/dist/esm/analyze.d.ts +3 -1
  60. package/dist/esm/analyze.js +18 -89
  61. package/dist/esm/analyze.js.map +1 -1
  62. package/dist/esm/audit.d.ts +6 -5
  63. package/dist/esm/audit.js +236 -29
  64. package/dist/esm/audit.js.map +1 -1
  65. package/dist/esm/cli/audit.js.map +1 -1
  66. package/dist/esm/cli/extract.js +10 -23
  67. package/dist/esm/cli/extract.js.map +1 -1
  68. package/dist/esm/cli/index.js.map +1 -1
  69. package/dist/esm/cli/render.js.map +1 -1
  70. package/dist/esm/cli/types.js.map +1 -1
  71. package/dist/esm/cli.js.map +1 -1
  72. package/dist/esm/collect.js +64 -15
  73. package/dist/esm/collect.js.map +1 -1
  74. package/dist/esm/dict.js.map +1 -1
  75. package/dist/esm/emit.js.map +1 -1
  76. package/dist/esm/engines/tesseract.js.map +1 -1
  77. package/dist/esm/errors.d.ts +2 -0
  78. package/dist/esm/errors.js.map +1 -1
  79. package/dist/esm/extract.d.ts +20 -9
  80. package/dist/esm/extract.js +167 -76
  81. package/dist/esm/extract.js.map +1 -1
  82. package/dist/esm/geometry.js.map +1 -1
  83. package/dist/esm/index.js.map +1 -1
  84. package/dist/esm/lines.js +14 -10
  85. package/dist/esm/lines.js.map +1 -1
  86. package/dist/esm/markdown.js +110 -27
  87. package/dist/esm/markdown.js.map +1 -1
  88. package/dist/esm/page-marker.d.ts +1 -1
  89. package/dist/esm/page-marker.js +15 -9
  90. package/dist/esm/page-marker.js.map +1 -1
  91. package/dist/esm/pdf-open.js.map +1 -1
  92. package/dist/esm/raster.js.map +1 -1
  93. package/dist/esm/render-analysis.d.ts +4 -2
  94. package/dist/esm/render-analysis.js +9 -5
  95. package/dist/esm/render-analysis.js.map +1 -1
  96. package/dist/esm/report.d.ts +2 -2
  97. package/dist/esm/report.js +12 -7
  98. package/dist/esm/report.js.map +1 -1
  99. package/dist/esm/resolve.js.map +1 -1
  100. package/dist/esm/text-analysis.d.ts +12 -0
  101. package/dist/esm/text-analysis.js +81 -0
  102. package/dist/esm/text-analysis.js.map +1 -0
  103. package/dist/esm/types.d.ts +40 -3
  104. package/dist/esm/types.js +4 -4
  105. package/dist/esm/types.js.map +1 -1
  106. package/dist/esm/validate.js.map +1 -1
  107. package/package.json +1 -1
  108. package/skills/pdf-to-md/SKILL.md +39 -66
package/dist/esm/types.js CHANGED
@@ -1,6 +1,6 @@
1
1
  // Shared shapes threaded through the collect -> lines -> markdown -> report
2
2
  // pipeline; kept in one place since every downstream stage uses at least one.
3
- // A positioned glyph run from the PDF's text layer: `s` may be more than one
4
- // character; `y` is flipped so ascending sorts top-down; `h` is font height rounded to 0.1pt.
5
- /** analyze()'s output: plain, JSON-serializable per-page words/blocks plus
6
- * the same QA `report` the other entry points return. */ export { };
3
+ /** Viewport-point AABB for a source text run's transformed advance by
4
+ * font-height box. This is a run box, not fabricated per-token geometry. */ /** One page's blocks from the text Markdown builder. This is internal
5
+ * rendering evidence, but it is declared here because the shared text
6
+ * analysis consumes it without parsing Markdown back into structure. */ export { };
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/types.ts"],"sourcesContent":["// Shared shapes threaded through the collect -> lines -> markdown -> report\n// pipeline; kept in one place since every downstream stage uses at least one.\n\n// A positioned glyph run from the PDF's text layer: `s` may be more than one\n// character; `y` is flipped so ascending sorts top-down; `h` is font height rounded to 0.1pt.\nexport interface Glyph {\n s: string;\n x: number;\n y: number;\n w: number;\n h: number;\n}\n\n// One page's worth of collected glyphs plus the geometry collect.ts needs\n// downstream (lines.ts's two-column detection, report.ts's hybrid-page flag).\nexport interface CollectedPage {\n page: number;\n glyphs: Glyph[];\n width: number;\n height: number;\n textChars: number;\n largeImage: boolean;\n}\n\n// collect.ts's return value: selected pages plus the char-weighted\n// font-height histogram, built ONLY from those pages (matters for a `--pages` re-extraction — see collect.ts).\nexport interface CollectResult {\n numPages: number;\n pages: CollectedPage[];\n heightChars: Map<number, number>;\n}\n\n// A visual line assembled from glyphs (lines.ts's `toLines`): `h` is the\n// char-weighted dominant height (not max); `col` is 0 (left/only) or 1 (right, de-braided page).\nexport interface Line {\n text: string;\n x: number;\n y: number;\n h: number;\n col: number;\n}\n\n// One page's lines, as src/extract.ts assembles them for buildMarkdown.\nexport interface PageLines {\n page: number;\n lines: Line[];\n}\n\nexport interface BuildMarkdownOptions {\n pageMarkers?: boolean;\n}\n\n// Running counters src/markdown.ts accumulates while building the output,\n// surfaced in the QA report (src/report.ts's buildReport).\nexport interface MarkdownStats {\n headings: number;\n listItems: number;\n paragraphs: number;\n joinedPairs: number;\n reflowed: number;\n callouts: number;\n closedHyphens: number;\n}\n\nexport interface MarkdownResult {\n md: string;\n stats: MarkdownStats;\n bodyH: number;\n headingSizes: number[];\n}\n\n/** One row of the per-page QA breakdown (\"### pN\" blocks) — used to decide\n * which pages are worth rendering and re-checking (README's \"Design contract\"). */\nexport interface PageStatsRow {\n page: number;\n chars: number;\n headings: number;\n junkHeadings: number;\n floats: number;\n paragraphs: number;\n danglingLong: number;\n danglingShort: number;\n /** count of `> [review: ...]` lines — blocks the geometry could not order\n * and labeled instead of guessing (README's \"Design contract\"). */\n reviewBlocks: number;\n /** present only on the text path's rows; omitted (not `false`) on the OCR\n * path, which has no per-page hybrid-image signal. */\n largeImages?: boolean;\n /** count of words below `report.ts`'s LOW_CONFIDENCE_THRESHOLD; omitted (not\n * 0) when the page had no confidence data at all, so \"measured 0\" and \"nothing to measure\" stay distinct. */\n lowConfidenceWords?: number;\n /** the suspect tokens behind `lowConfidenceWords`, worst-confidence first\n * and capped (`report.ts`'s LOW_CONFIDENCE_SAMPLE_CAP) so a chart-heavy page can't bloat the JSON. */\n lowConfidenceSample?: string[];\n}\n\n// buildReport's params (src/report.ts) — everything the text (default) path\n// hands over after running collect -> lines -> markdown.\nexport interface BuildReportInput {\n numPages: number;\n md: string;\n /** identical to `md` but always carries \"### pN\" markers, so pageStats is\n * measured independent of the caller's `--page-markers` choice; same string as `md` when markers were requested. */\n statsMd: string;\n stats: MarkdownStats;\n bodyH: number;\n headingSizes: number[];\n hybridPages?: number[];\n}\n\n/** One recognized word's page/text/confidence — `report.ts`'s low-confidence\n * detector input; a structural subset of OcrWordInput kept here so `types.ts` doesn't import the library-layer module that depends on it. */\nexport interface WordConfidenceInput {\n page: number;\n text: string;\n confidence?: number;\n}\n\nexport interface BuildOcrReportInput {\n md: string;\n fallback?: boolean;\n /** every recognized word, for `report.ts`'s low-confidence detector.\n * Defaults to [] (a pre-confidence --words-json replay), reporting the detector absent, never wrong. */\n words?: WordConfidenceInput[];\n}\n\n/** Per-document and per-page QA metrics returned by every extraction entry\n * point, discriminated by `path`; OCR-only and text-only fields are grouped below. */\nexport interface Report {\n /** manually-bumped integer (not the npm package version) — see\n * `report.ts`'s OUTPUT_VERSION for the bump rule; present on every Report so a frozen baseline's staleness is self-announcing. */\n outputVersion: number;\n path: 'text' | 'ocr';\n pages: number;\n chars: number;\n charsPerPage: number;\n sparse: boolean;\n headings: number;\n junkHeadings: number;\n paragraphs: number;\n callouts: number;\n danglingLong: number;\n danglingShort: number;\n /** total `> [review: ...]` blocks across all pages (sum of pageStats'\n * reviewBlocks); always 0 on the text path — only the OCR path emits review markers. */\n reviewBlocks: number;\n /** document-level sum of pageStats' `lowConfidenceWords` — see that\n * field for why 0 can mean \"measured 0\" or \"nothing to measure\". */\n lowConfidenceWords: number;\n pageStats: PageStatsRow[];\n\n // --- OCR-only ---\n floats?: number;\n ocrFallback?: boolean;\n\n // --- text-only ---\n bodyFontHeight?: number;\n headingSizes?: number[];\n listItems?: number;\n joinedPairs?: number;\n closedHyphens?: number;\n hybridPages?: number[];\n}\n\n// --- Analysis: the raw-mode IR (analyze() produces it; toMarkdown()/toText()\n// in render-analysis.ts render it — see toMarkdown's doc comment for the fail-fast validation contract) ---\n\n/** One recognized/collected word, page-relative; `box` is always present.\n * See `AnalysisPage.words` for the coordinate contract and per-path granularity difference. */\nexport interface AnalysisWord {\n text: string;\n box: { x: number; y: number; w: number; h: number };\n confidence?: number;\n}\n\n/** One structural block; `wordIndexes` links it to its page's `words` list (blocks -> words only).\n * `text` is bare content for `'heading'` (level separate), else the LITERAL markdown line (decoration included, round-trippable) — a pull-quote is `'paragraph'`, not a 5th type. */\nexport interface AnalysisBlock {\n type: 'heading' | 'paragraph' | 'float' | 'review';\n text: string;\n /** Heading level (integer 1-6); REQUIRED when `type` is `'heading'`,\n * FORBIDDEN otherwise — toMarkdown()/toText() validate this fail-fast (throws PdfToMdError/'ANALYSIS_INPUT'), never silently ignored. */\n level?: number;\n wordIndexes: number[];\n}\n\n/** One page's flat words plus its structural blocks. */\nexport interface AnalysisPage {\n page: number;\n /** Flat, page-relative word list; `AnalysisBlock.wordIndexes` indexes into it. `box` is always normalized [0,1], origin bottom-left, y up, regardless of path.\n * OCR-path is genuinely per-word; TEXT-path is per visual LINE (`box.w` always 0, `box.y` the line's baseline) — coarser but honest, not faked. */\n words: AnalysisWord[];\n blocks: AnalysisBlock[];\n}\n\n/** analyze()'s output: plain, JSON-serializable per-page words/blocks plus\n * the same QA `report` the other entry points return. */\nexport interface Analysis {\n /** manually-bumped integer — see `Report.outputVersion` / `report.ts`'s OUTPUT_VERSION. */\n outputVersion: number;\n /** the same QA report extractText/extractOcr/pdfToMarkdown return, always\n * measured off real markdown (OCR: the shared core's output; TEXT: buildMarkdown's reflowed output, not this coarser IR) — never stale relative to the IR. */\n report: Report;\n pages: AnalysisPage[];\n}\n"],"names":[],"mappings":"AAAA,4EAA4E;AAC5E,8EAA8E;AAE9E,6EAA6E;AAC7E,8FAA8F;AA+L9F;uDACuD,GACvD,WAOC"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/types.ts"],"sourcesContent":["// Shared shapes threaded through the collect -> lines -> markdown -> report\n// pipeline; kept in one place since every downstream stage uses at least one.\n\n/** Viewport-point AABB for a source text run's transformed advance by\n * font-height box. This is a run box, not fabricated per-token geometry. */\nexport interface GlyphBounds {\n x: number;\n y: number;\n w: number;\n h: number;\n}\n\n// A positioned glyph run from the PDF's text layer: `s` may be more than one\n// character. x/y are the transformed baseline origin in viewport points;\n// w/h are its advance and font height, and bounds is their viewport AABB.\nexport interface Glyph {\n s: string;\n x: number;\n y: number;\n w: number;\n h: number;\n bounds: GlyphBounds;\n}\n\n// One page's worth of collected glyphs plus the geometry collect.ts needs\n// downstream (lines.ts's two-column detection, report.ts's hybrid-page flag).\nexport interface CollectedPage {\n page: number;\n glyphs: Glyph[];\n width: number;\n height: number;\n textChars: number;\n /** any painted image operator, independent of text-layer density */\n hasImage: boolean;\n largeImage: boolean;\n}\n\n// collect.ts's return value: selected pages plus the char-weighted\n// font-height histogram, built ONLY from those pages (matters for a `--pages` re-extraction — see collect.ts).\nexport interface CollectResult {\n numPages: number;\n pages: CollectedPage[];\n heightChars: Map<number, number>;\n}\n\n// A visual line assembled from glyphs (lines.ts's `toLines`): `h` is the\n// char-weighted dominant height (not max); `col` is 0 (left/only) or 1 (right, de-braided page).\nexport interface Line {\n text: string;\n x: number;\n y: number;\n h: number;\n col: number;\n /** indexes of the collected PDF text runs that produced this line, in\n * display order. OCR has its own source-word linkage. */\n wordIndexes: number[];\n}\n\n// One page's lines, as src/extract.ts assembles them for buildMarkdown.\nexport interface PageLines {\n page: number;\n lines: Line[];\n}\n\nexport interface BuildMarkdownOptions {\n pageMarkers?: boolean;\n}\n\n// Running counters src/markdown.ts accumulates while building the output,\n// surfaced in the QA report (src/report.ts's buildReport).\nexport interface MarkdownStats {\n headings: number;\n listItems: number;\n paragraphs: number;\n joinedPairs: number;\n reflowed: number;\n callouts: number;\n closedHyphens: number;\n}\n\nexport interface MarkdownResult {\n md: string;\n stats: MarkdownStats;\n bodyH: number;\n headingSizes: number[];\n /** The structural blocks that produced `md`, grouped by source page.\n * Markers are rendering provenance and are deliberately not blocks. */\n pages: MarkdownPage[];\n}\n\n/** One row of the per-page QA breakdown (\"### pN\" blocks) — used to decide\n * which pages are worth rendering and re-checking (README's \"Design contract\"). */\nexport interface PageStatsRow {\n page: number;\n /** producer retained for this page, including in a mixed result */\n path: 'text' | 'ocr';\n chars: number;\n headings: number;\n junkHeadings: number;\n floats: number;\n paragraphs: number;\n danglingLong: number;\n danglingShort: number;\n /** count of `> [review: ...]` lines — blocks the geometry could not order\n * and labeled instead of guessing (README's \"Design contract\"). */\n reviewBlocks: number;\n /** present only on the text path's rows; omitted (not `false`) on the OCR\n * path, which has no per-page hybrid-image signal. */\n largeImages?: boolean;\n /** count of words below `report.ts`'s LOW_CONFIDENCE_THRESHOLD; omitted (not\n * 0) when the page had no confidence data at all, so \"measured 0\" and \"nothing to measure\" stay distinct. */\n lowConfidenceWords?: number;\n /** the suspect tokens behind `lowConfidenceWords`, worst-confidence first\n * and capped (`report.ts`'s LOW_CONFIDENCE_SAMPLE_CAP) so a chart-heavy page can't bloat the JSON. */\n lowConfidenceSample?: string[];\n}\n\n// buildReport's params (src/report.ts) — everything the text (default) path\n// hands over after running collect -> lines -> markdown.\nexport interface BuildReportInput {\n numPages: number;\n md: string;\n /** identical to `md` but always carries \"### pN\" markers, so pageStats is\n * measured independent of the caller's `--page-markers` choice; same string as `md` when markers were requested. */\n statsMd: string;\n stats: MarkdownStats;\n bodyH: number;\n headingSizes: number[];\n hybridPages?: number[];\n}\n\n/** One recognized word's page/text/confidence — `report.ts`'s low-confidence\n * detector input; a structural subset of OcrWordInput kept here so `types.ts` doesn't import the library-layer module that depends on it. */\nexport interface WordConfidenceInput {\n page: number;\n text: string;\n confidence?: number;\n}\n\nexport interface BuildOcrReportInput {\n md: string;\n fallback?: boolean;\n /** Per-page producer map for a mixed result. Omit for OCR-only output. */\n pagePaths?: Map<number, 'text' | 'ocr'>;\n /** text pages retaining a large embedded image, for mixed per-page QA */\n hybridPages?: number[];\n /** every recognized word, for `report.ts`'s low-confidence detector.\n * Defaults to [] (a pre-confidence --words-json replay), reporting the detector absent, never wrong. */\n words?: WordConfidenceInput[];\n}\n\n/** Per-document and per-page QA metrics returned by every extraction entry\n * point, discriminated by `path`; OCR-only and text-only fields are grouped below. */\nexport interface Report {\n /** manually-bumped integer (not the npm package version) — see\n * `report.ts`'s OUTPUT_VERSION for the bump rule; present on every Report so a frozen baseline's staleness is self-announcing. */\n outputVersion: number;\n path: 'text' | 'ocr' | 'mixed';\n pages: number;\n chars: number;\n charsPerPage: number;\n sparse: boolean;\n headings: number;\n junkHeadings: number;\n paragraphs: number;\n callouts: number;\n danglingLong: number;\n danglingShort: number;\n /** total `> [review: ...]` blocks across all pages (sum of pageStats'\n * reviewBlocks); always 0 on the text path — only the OCR path emits review markers. */\n reviewBlocks: number;\n /** document-level sum of pageStats' `lowConfidenceWords` — see that\n * field for why 0 can mean \"measured 0\" or \"nothing to measure\". */\n lowConfidenceWords: number;\n pageStats: PageStatsRow[];\n\n // --- OCR-only ---\n floats?: number;\n ocrFallback?: boolean;\n\n // --- text-only ---\n bodyFontHeight?: number;\n headingSizes?: number[];\n listItems?: number;\n joinedPairs?: number;\n closedHyphens?: number;\n hybridPages?: number[];\n}\n\n// --- Analysis: the raw-mode IR (analyze() produces it; toMarkdown()/toText()\n// in render-analysis.ts render it — see toMarkdown's doc comment for the fail-fast validation contract) ---\n\n/** One recognized/collected word, page-relative; `box` is always present.\n * See `AnalysisPage.words` for the coordinate contract and per-path granularity difference. */\nexport interface AnalysisWord {\n text: string;\n box: { x: number; y: number; w: number; h: number };\n confidence?: number;\n}\n\n/** One structural block; `wordIndexes` links it to its page's `words` list (blocks -> words only).\n * `text` is bare content for `'heading'` (level separate), else the LITERAL markdown line (decoration included, round-trippable) — a pull-quote is `'paragraph'`, not a 5th type. */\nexport interface AnalysisBlock {\n type: 'heading' | 'paragraph' | 'float' | 'review';\n text: string;\n /** Heading level (integer 1-6); REQUIRED when `type` is `'heading'`,\n * FORBIDDEN otherwise — toMarkdown()/toText() validate this fail-fast (throws PdfToMdError/'ANALYSIS_INPUT'), never silently ignored. */\n level?: number;\n /** adjacent Markdown line with no blank separator, used for list items */\n tight?: boolean;\n wordIndexes: number[];\n}\n\n/** One page's flat words plus its structural blocks. */\nexport interface AnalysisPage {\n page: number;\n /** producer retained for this page; `Analysis.report.path` is `'mixed'`\n * when the document has more than one page producer. */\n path: 'text' | 'ocr';\n /** Flat, page-relative source items; `AnalysisBlock.wordIndexes` indexes\n * into it. OCR boxes are normalized [0,1]. Text-run boxes are page\n * fractions and may extend outside that range when the PDF run extends past\n * the viewport. Both use a bottom-left origin with y up. */\n words: AnalysisWord[];\n blocks: AnalysisBlock[];\n}\n\n/** analyze()'s output: plain, JSON-serializable per-page words/blocks plus\n * the same QA `report` the other entry points return. */\nexport interface Analysis {\n /** manually-bumped integer — see `Report.outputVersion` / `report.ts`'s OUTPUT_VERSION. */\n outputVersion: number;\n /** the same QA report extractText/extractOcr/pdfToMarkdown return, always\n * measured off real markdown (OCR: the shared core's output; TEXT: buildMarkdown's reflowed output, not this coarser IR) — never stale relative to the IR. */\n report: Report;\n pages: AnalysisPage[];\n}\n\n/** One page's blocks from the text Markdown builder. This is internal\n * rendering evidence, but it is declared here because the shared text\n * analysis consumes it without parsing Markdown back into structure. */\nexport interface MarkdownPage {\n page: number;\n blocks: AnalysisBlock[];\n}\n"],"names":[],"mappings":"AAAA,4EAA4E;AAC5E,8EAA8E;AAE9E;0EAC0E,GA0O1E;;sEAEsE,GACtE,WAGC"}
@@ -1 +1 @@
1
- {"version":3,"sources":["/Users/kevin/Dev/OpenSource/formats/pdf-to-md/src/validate.ts"],"sourcesContent":["// validate.ts — input-shape validation shared by analyze.ts and extract.ts,\n// kept out of both to dodge their value-level import cycle (extract.ts imports analyze.ts's buildOcrAnalysisPages; both need this).\nimport { PdfToMdError } from './errors.ts';\nimport type { OcrWordInput } from './extract.ts';\n\n/** Validates the one option every entry point accepts before touching the\n * PDF. Type-only import from `extract.ts` keeps this cycle-free. */\nexport function validatePages(pages: unknown): void {\n if (pages === undefined) return;\n const isRecord = typeof pages === 'object' && pages !== null;\n const first = isRecord ? (pages as Record<string, unknown>).first : undefined;\n const last = isRecord ? (pages as Record<string, unknown>).last : undefined;\n const firstOk = typeof first === 'number' && Number.isFinite(first) && first >= 1;\n const lastOk = last === undefined || (typeof last === 'number' && Number.isFinite(last) && last >= (first as number));\n if (!isRecord || !firstOk || !lastOk) {\n throw new PdfToMdError('PAGE_RANGE', `pages option must be { first: number, last?: number } with first >= 1 and last >= first (got ${JSON.stringify(pages)})`);\n }\n}\n\n/** Validates analyze()'s non-string `input` branch (`{ words }`) before\n * `.words` is read, so a plain-JS mistake like `analyze(null)` throws PdfToMdError/WORDS_INPUT instead of a raw TypeError. */\nexport function validateWordsInput(input: unknown): OcrWordInput[] {\n const isRecord = typeof input === 'object' && input !== null;\n const words = isRecord ? (input as Record<string, unknown>).words : undefined;\n if (!Array.isArray(words)) {\n throw new PdfToMdError('WORDS_INPUT', `analyze() input must be a pdf path (string) or { words: OcrWordInput[] } (got ${isRecord ? JSON.stringify(input) : String(input)})`);\n }\n return words as OcrWordInput[];\n}\n\n/** Runtime shape guard for `buildOcrAnalysisPages`'s input: a plain-JS\n * caller bypasses the TS union, so this turns a raw TypeError into PdfToMdError before any property is read off it. */\nexport function validateOcrAnalysisInput(input: unknown): { pdfPath: string } | { words: OcrWordInput[] } {\n if (typeof input === 'string') {\n throw new PdfToMdError('PDF_OPEN', `extractOcr/analyze input must be { pdfPath: string } or { words: OcrWordInput[] }, not a bare path string (got ${JSON.stringify(input)}) — did you mean { pdfPath: ${JSON.stringify(input)} }?`);\n }\n const isRecord = typeof input === 'object' && input !== null;\n const rawPdfPath = isRecord ? (input as Record<string, unknown>).pdfPath : undefined;\n const rawWords = isRecord ? (input as Record<string, unknown>).words : undefined;\n if (!isRecord || (rawPdfPath === undefined && rawWords === undefined)) {\n throw new PdfToMdError('WORDS_INPUT', `extractOcr/analyze input must be { pdfPath: string } or { words: OcrWordInput[] } (got ${isRecord ? JSON.stringify(input) : String(input)})`);\n }\n if (rawWords !== undefined && !Array.isArray(rawWords)) {\n throw new PdfToMdError('WORDS_INPUT', `extractOcr/analyze { words } input must be an array (got ${JSON.stringify(rawWords)})`);\n }\n if (rawPdfPath !== undefined && typeof rawPdfPath !== 'string') {\n throw new PdfToMdError('PDF_OPEN', `extractOcr/analyze { pdfPath } input must be a string (got ${JSON.stringify(rawPdfPath)})`);\n }\n if (rawWords !== undefined) {\n for (const w of rawWords as unknown[]) {\n const rec = w as Record<string, unknown>;\n if (typeof rec.page !== 'number' || typeof rec.text !== 'string' || typeof rec.x !== 'number' || typeof rec.y !== 'number' || typeof rec.w !== 'number' || typeof rec.h !== 'number') {\n throw new PdfToMdError('WORDS_INPUT', `malformed word entry (need {page, text, x, y, w, h}): ${JSON.stringify(w)}`);\n }\n }\n return { words: rawWords as OcrWordInput[] };\n }\n return { pdfPath: rawPdfPath as string };\n}\n"],"names":["PdfToMdError","validatePages","pages","undefined","isRecord","first","last","firstOk","Number","isFinite","lastOk","JSON","stringify","validateWordsInput","input","words","Array","isArray","String","validateOcrAnalysisInput","rawPdfPath","pdfPath","rawWords","w","rec","page","text","x","y","h"],"mappings":"AAAA,4EAA4E;AAC5E,oIAAoI;AACpI,SAASA,YAAY,QAAQ,cAAc;AAG3C;kEACkE,GAClE,OAAO,SAASC,cAAcC,KAAc;IAC1C,IAAIA,UAAUC,WAAW;IACzB,MAAMC,WAAW,OAAOF,UAAU,YAAYA,UAAU;IACxD,MAAMG,QAAQD,WAAW,AAACF,MAAkCG,KAAK,GAAGF;IACpE,MAAMG,OAAOF,WAAW,AAACF,MAAkCI,IAAI,GAAGH;IAClE,MAAMI,UAAU,OAAOF,UAAU,YAAYG,OAAOC,QAAQ,CAACJ,UAAUA,SAAS;IAChF,MAAMK,SAASJ,SAASH,aAAc,OAAOG,SAAS,YAAYE,OAAOC,QAAQ,CAACH,SAASA,QAASD;IACpG,IAAI,CAACD,YAAY,CAACG,WAAW,CAACG,QAAQ;QACpC,MAAM,IAAIV,aAAa,cAAc,CAAC,6FAA6F,EAAEW,KAAKC,SAAS,CAACV,OAAO,CAAC,CAAC;IAC/J;AACF;AAEA;4HAC4H,GAC5H,OAAO,SAASW,mBAAmBC,KAAc;IAC/C,MAAMV,WAAW,OAAOU,UAAU,YAAYA,UAAU;IACxD,MAAMC,QAAQX,WAAW,AAACU,MAAkCC,KAAK,GAAGZ;IACpE,IAAI,CAACa,MAAMC,OAAO,CAACF,QAAQ;QACzB,MAAM,IAAIf,aAAa,eAAe,CAAC,8EAA8E,EAAEI,WAAWO,KAAKC,SAAS,CAACE,SAASI,OAAOJ,OAAO,CAAC,CAAC;IAC5K;IACA,OAAOC;AACT;AAEA;qHACqH,GACrH,OAAO,SAASI,yBAAyBL,KAAc;IACrD,IAAI,OAAOA,UAAU,UAAU;QAC7B,MAAM,IAAId,aAAa,YAAY,CAAC,+GAA+G,EAAEW,KAAKC,SAAS,CAACE,OAAO,4BAA4B,EAAEH,KAAKC,SAAS,CAACE,OAAO,GAAG,CAAC;IACrO;IACA,MAAMV,WAAW,OAAOU,UAAU,YAAYA,UAAU;IACxD,MAAMM,aAAahB,WAAW,AAACU,MAAkCO,OAAO,GAAGlB;IAC3E,MAAMmB,WAAWlB,WAAW,AAACU,MAAkCC,KAAK,GAAGZ;IACvE,IAAI,CAACC,YAAagB,eAAejB,aAAamB,aAAanB,WAAY;QACrE,MAAM,IAAIH,aAAa,eAAe,CAAC,uFAAuF,EAAEI,WAAWO,KAAKC,SAAS,CAACE,SAASI,OAAOJ,OAAO,CAAC,CAAC;IACrL;IACA,IAAIQ,aAAanB,aAAa,CAACa,MAAMC,OAAO,CAACK,WAAW;QACtD,MAAM,IAAItB,aAAa,eAAe,CAAC,yDAAyD,EAAEW,KAAKC,SAAS,CAACU,UAAU,CAAC,CAAC;IAC/H;IACA,IAAIF,eAAejB,aAAa,OAAOiB,eAAe,UAAU;QAC9D,MAAM,IAAIpB,aAAa,YAAY,CAAC,2DAA2D,EAAEW,KAAKC,SAAS,CAACQ,YAAY,CAAC,CAAC;IAChI;IACA,IAAIE,aAAanB,WAAW;QAC1B,KAAK,MAAMoB,KAAKD,SAAuB;YACrC,MAAME,MAAMD;YACZ,IAAI,OAAOC,IAAIC,IAAI,KAAK,YAAY,OAAOD,IAAIE,IAAI,KAAK,YAAY,OAAOF,IAAIG,CAAC,KAAK,YAAY,OAAOH,IAAII,CAAC,KAAK,YAAY,OAAOJ,IAAID,CAAC,KAAK,YAAY,OAAOC,IAAIK,CAAC,KAAK,UAAU;gBACpL,MAAM,IAAI7B,aAAa,eAAe,CAAC,sDAAsD,EAAEW,KAAKC,SAAS,CAACW,IAAI;YACpH;QACF;QACA,OAAO;YAAER,OAAOO;QAA2B;IAC7C;IACA,OAAO;QAAED,SAASD;IAAqB;AACzC"}
1
+ {"version":3,"sources":["/Users/kevin/Dev/open-source/formats/pdf-to-md/src/validate.ts"],"sourcesContent":["// validate.ts — input-shape validation shared by analyze.ts and extract.ts,\n// kept out of both to dodge their value-level import cycle (extract.ts imports analyze.ts's buildOcrAnalysisPages; both need this).\nimport { PdfToMdError } from './errors.ts';\nimport type { OcrWordInput } from './extract.ts';\n\n/** Validates the one option every entry point accepts before touching the\n * PDF. Type-only import from `extract.ts` keeps this cycle-free. */\nexport function validatePages(pages: unknown): void {\n if (pages === undefined) return;\n const isRecord = typeof pages === 'object' && pages !== null;\n const first = isRecord ? (pages as Record<string, unknown>).first : undefined;\n const last = isRecord ? (pages as Record<string, unknown>).last : undefined;\n const firstOk = typeof first === 'number' && Number.isFinite(first) && first >= 1;\n const lastOk = last === undefined || (typeof last === 'number' && Number.isFinite(last) && last >= (first as number));\n if (!isRecord || !firstOk || !lastOk) {\n throw new PdfToMdError('PAGE_RANGE', `pages option must be { first: number, last?: number } with first >= 1 and last >= first (got ${JSON.stringify(pages)})`);\n }\n}\n\n/** Validates analyze()'s non-string `input` branch (`{ words }`) before\n * `.words` is read, so a plain-JS mistake like `analyze(null)` throws PdfToMdError/WORDS_INPUT instead of a raw TypeError. */\nexport function validateWordsInput(input: unknown): OcrWordInput[] {\n const isRecord = typeof input === 'object' && input !== null;\n const words = isRecord ? (input as Record<string, unknown>).words : undefined;\n if (!Array.isArray(words)) {\n throw new PdfToMdError('WORDS_INPUT', `analyze() input must be a pdf path (string) or { words: OcrWordInput[] } (got ${isRecord ? JSON.stringify(input) : String(input)})`);\n }\n return words as OcrWordInput[];\n}\n\n/** Runtime shape guard for `buildOcrAnalysisPages`'s input: a plain-JS\n * caller bypasses the TS union, so this turns a raw TypeError into PdfToMdError before any property is read off it. */\nexport function validateOcrAnalysisInput(input: unknown): { pdfPath: string } | { words: OcrWordInput[] } {\n if (typeof input === 'string') {\n throw new PdfToMdError('PDF_OPEN', `extractOcr/analyze input must be { pdfPath: string } or { words: OcrWordInput[] }, not a bare path string (got ${JSON.stringify(input)}) — did you mean { pdfPath: ${JSON.stringify(input)} }?`);\n }\n const isRecord = typeof input === 'object' && input !== null;\n const rawPdfPath = isRecord ? (input as Record<string, unknown>).pdfPath : undefined;\n const rawWords = isRecord ? (input as Record<string, unknown>).words : undefined;\n if (!isRecord || (rawPdfPath === undefined && rawWords === undefined)) {\n throw new PdfToMdError('WORDS_INPUT', `extractOcr/analyze input must be { pdfPath: string } or { words: OcrWordInput[] } (got ${isRecord ? JSON.stringify(input) : String(input)})`);\n }\n if (rawWords !== undefined && !Array.isArray(rawWords)) {\n throw new PdfToMdError('WORDS_INPUT', `extractOcr/analyze { words } input must be an array (got ${JSON.stringify(rawWords)})`);\n }\n if (rawPdfPath !== undefined && typeof rawPdfPath !== 'string') {\n throw new PdfToMdError('PDF_OPEN', `extractOcr/analyze { pdfPath } input must be a string (got ${JSON.stringify(rawPdfPath)})`);\n }\n if (rawWords !== undefined) {\n for (const w of rawWords as unknown[]) {\n const rec = w as Record<string, unknown>;\n if (typeof rec.page !== 'number' || typeof rec.text !== 'string' || typeof rec.x !== 'number' || typeof rec.y !== 'number' || typeof rec.w !== 'number' || typeof rec.h !== 'number') {\n throw new PdfToMdError('WORDS_INPUT', `malformed word entry (need {page, text, x, y, w, h}): ${JSON.stringify(w)}`);\n }\n }\n return { words: rawWords as OcrWordInput[] };\n }\n return { pdfPath: rawPdfPath as string };\n}\n"],"names":["PdfToMdError","validatePages","pages","undefined","isRecord","first","last","firstOk","Number","isFinite","lastOk","JSON","stringify","validateWordsInput","input","words","Array","isArray","String","validateOcrAnalysisInput","rawPdfPath","pdfPath","rawWords","w","rec","page","text","x","y","h"],"mappings":"AAAA,4EAA4E;AAC5E,oIAAoI;AACpI,SAASA,YAAY,QAAQ,cAAc;AAG3C;kEACkE,GAClE,OAAO,SAASC,cAAcC,KAAc;IAC1C,IAAIA,UAAUC,WAAW;IACzB,MAAMC,WAAW,OAAOF,UAAU,YAAYA,UAAU;IACxD,MAAMG,QAAQD,WAAW,AAACF,MAAkCG,KAAK,GAAGF;IACpE,MAAMG,OAAOF,WAAW,AAACF,MAAkCI,IAAI,GAAGH;IAClE,MAAMI,UAAU,OAAOF,UAAU,YAAYG,OAAOC,QAAQ,CAACJ,UAAUA,SAAS;IAChF,MAAMK,SAASJ,SAASH,aAAc,OAAOG,SAAS,YAAYE,OAAOC,QAAQ,CAACH,SAASA,QAASD;IACpG,IAAI,CAACD,YAAY,CAACG,WAAW,CAACG,QAAQ;QACpC,MAAM,IAAIV,aAAa,cAAc,CAAC,6FAA6F,EAAEW,KAAKC,SAAS,CAACV,OAAO,CAAC,CAAC;IAC/J;AACF;AAEA;4HAC4H,GAC5H,OAAO,SAASW,mBAAmBC,KAAc;IAC/C,MAAMV,WAAW,OAAOU,UAAU,YAAYA,UAAU;IACxD,MAAMC,QAAQX,WAAW,AAACU,MAAkCC,KAAK,GAAGZ;IACpE,IAAI,CAACa,MAAMC,OAAO,CAACF,QAAQ;QACzB,MAAM,IAAIf,aAAa,eAAe,CAAC,8EAA8E,EAAEI,WAAWO,KAAKC,SAAS,CAACE,SAASI,OAAOJ,OAAO,CAAC,CAAC;IAC5K;IACA,OAAOC;AACT;AAEA;qHACqH,GACrH,OAAO,SAASI,yBAAyBL,KAAc;IACrD,IAAI,OAAOA,UAAU,UAAU;QAC7B,MAAM,IAAId,aAAa,YAAY,CAAC,+GAA+G,EAAEW,KAAKC,SAAS,CAACE,OAAO,4BAA4B,EAAEH,KAAKC,SAAS,CAACE,OAAO,GAAG,CAAC;IACrO;IACA,MAAMV,WAAW,OAAOU,UAAU,YAAYA,UAAU;IACxD,MAAMM,aAAahB,WAAW,AAACU,MAAkCO,OAAO,GAAGlB;IAC3E,MAAMmB,WAAWlB,WAAW,AAACU,MAAkCC,KAAK,GAAGZ;IACvE,IAAI,CAACC,YAAagB,eAAejB,aAAamB,aAAanB,WAAY;QACrE,MAAM,IAAIH,aAAa,eAAe,CAAC,uFAAuF,EAAEI,WAAWO,KAAKC,SAAS,CAACE,SAASI,OAAOJ,OAAO,CAAC,CAAC;IACrL;IACA,IAAIQ,aAAanB,aAAa,CAACa,MAAMC,OAAO,CAACK,WAAW;QACtD,MAAM,IAAItB,aAAa,eAAe,CAAC,yDAAyD,EAAEW,KAAKC,SAAS,CAACU,UAAU,CAAC,CAAC;IAC/H;IACA,IAAIF,eAAejB,aAAa,OAAOiB,eAAe,UAAU;QAC9D,MAAM,IAAIpB,aAAa,YAAY,CAAC,2DAA2D,EAAEW,KAAKC,SAAS,CAACQ,YAAY,CAAC,CAAC;IAChI;IACA,IAAIE,aAAanB,WAAW;QAC1B,KAAK,MAAMoB,KAAKD,SAAuB;YACrC,MAAME,MAAMD;YACZ,IAAI,OAAOC,IAAIC,IAAI,KAAK,YAAY,OAAOD,IAAIE,IAAI,KAAK,YAAY,OAAOF,IAAIG,CAAC,KAAK,YAAY,OAAOH,IAAII,CAAC,KAAK,YAAY,OAAOJ,IAAID,CAAC,KAAK,YAAY,OAAOC,IAAIK,CAAC,KAAK,UAAU;gBACpL,MAAM,IAAI7B,aAAa,eAAe,CAAC,sDAAsD,EAAEW,KAAKC,SAAS,CAACW,IAAI;YACpH;QACF;QACA,OAAO;YAAER,OAAOO;QAA2B;IAC7C;IACA,OAAO;QAAED,SAASD;IAAqB;AACzC"}
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@effortlessmotion/pdf-to-md",
3
- "version": "0.1.8",
3
+ "version": "0.2.1",
4
4
  "description": "Extract PDFs to structured markdown: text layer via pdf.js, OCR fallback for image-only pages, pure npm and cross-platform",
5
5
  "keywords": [
6
6
  "pdf",
@@ -1,102 +1,75 @@
1
1
  ---
2
2
  name: pdf-to-md
3
- description: "Convert a PDF to markdown with pdf-to-md and drive its QA loop to a lossless result: read the --json report's pageStats to find suspect pages, render pages to PNG as ground truth, re-extract targeted page ranges with tuning flags, and run the severity-1 word-loss audit before trusting the output. Use when converting a PDF to markdown with pdf-to-md, when a pdf-to-md conversion looks wrong (garbled reading order, misplaced text, misread words, a suspicious page), or when asked to verify a pdf-to-md conversion is lossless."
3
+ description: "Convert PDFs to Markdown with pdf-to-md, investigate missing or misordered text, and verify conversions against source pages and extraction evidence. Use for conversion, targeted re-extraction, or quality checks of pdf-to-md output."
4
4
  ---
5
5
 
6
6
  # pdf-to-md
7
7
 
8
- pdf-to-md converts PDFs to markdown: text layer via pdf.js, OCR (tesseract.js) fallback for
9
- image-only or sparse pages. Its contract: recognized text is never silently dropped,
10
- unrecoverable structure is marked (`> [review: ...]`) rather than guessed, and every claim the
11
- output makes is mechanically checkable. This skill is the QA loop that contract is built for —
12
- convert, read the report, verify against the real page, re-extract, audit.
8
+ pdf-to-md reads embedded text and uses local OCR where needed. A preservation audit checks extracted text against Markdown. Recognition accuracy, reading order, tables and diagrams require comparison with rendered source pages.
13
9
 
14
- ## 1. Convert
10
+ ## Convert with evidence
15
11
 
16
- ```
17
- pdf-to-md file.pdf --json
12
+ ```sh
13
+ pdf-to-md file.pdf out.md --json --page-markers --debug-words=words.jsonl
18
14
  ```
19
15
 
20
- Auto-detects the OCR need (falls back when the text layer is too sparse) and prints the QA report
21
- alongside the markdown. `--ocr` forces the OCR path; `--no-ocr` suppresses the fallback and keeps
22
- the sparse text-layer output. `--pages N[-M]` limits to a page range; `--stdout` writes markdown
23
- to stdout instead of a file.
16
+ Page markers connect the output to its source pages. `--debug-words` captures text-layer items or recognized OCR words before layout reconstruction. Text items can contain multiple words. Parse the dump as JSON Lines; key order is not fixed.
24
17
 
25
- ## 2. Read the QA report
18
+ `--ocr` forces recognition and `--no-ocr` retains the text path. `--pages N[-M]` selects source pages. A mixed document can use different extraction paths per page. Keep the input identity, tool version, effective options, report and dump with the run's working artifacts so later corrections use matching evidence.
26
19
 
27
- `report.pageStats` (one row per page) flags pages worth checking by hand:
20
+ ## Find pages to inspect
28
21
 
29
- - `danglingLong` — a paragraph broke off without terminal punctuation before the next one starts:
30
- a likely invisible sentence split the geometry got wrong.
31
- - `reviewBlocks` — count of `> [review: reason]` blocks: the geometry declined to guess an order
32
- and labeled the block instead of guessing. Always worth reading.
33
- - `lowConfidenceWords` / `lowConfidenceSample` — words the OCR engine itself scored below its
34
- confidence threshold; the sample is the actual suspect tokens, worst first. Probable misreads.
35
- - `junkHeadings` — lines promoted to headings by a font-size heuristic that are probably not
36
- headings.
37
- - `largeImages` (text path only) — a real text layer plus a large embedded image: text baked into
38
- the image may duplicate or replace what the text layer extracted. Don't assume the text layer
39
- is complete — compare it against `--pages N --ocr` on that page.
22
+ Read `report.pageStats` and warnings:
40
23
 
41
- A page with none of these flags isn't proof it's correct, only that these particular detectors
42
- found nothing — pair with step 3 for anything the report doesn't cover.
24
+ - `danglingLong` identifies prose that may have broken at the wrong place.
25
+ - `reviewBlocks` counts OCR blocks whose structure could not be recovered. The text path does not provide this detector; zero is not a layout guarantee.
26
+ - `lowConfidenceWords` and `lowConfidenceSample` identify uncertain OCR tokens. Missing confidence data is not a high-confidence result.
27
+ - `junkHeadings` identifies possible font-size classification errors.
28
+ - `largeImages` on text pages identifies images that may contain additional text. Compare a targeted OCR pass before replacing usable embedded text.
43
29
 
44
- ## 3. Ground truth: render the page
30
+ Also inspect sparse or empty pages and any layout the detectors do not cover. An unflagged page can still have incorrect columns, character mappings, tables or missing visual material.
45
31
 
46
- ```
32
+ ## Compare against the page
33
+
34
+ ```sh
47
35
  pdf-to-md render file.pdf N
48
36
  ```
49
37
 
50
- Renders page N to PNG at 288dpi — the same resolution the OCR path reads at (override with
51
- `--dpi`) — into the current directory (or `--out DIR`). View the PNG next to the flagged markdown;
52
- this is the only real ground truth, everything in the report is inference on top of it.
38
+ The default render is 288 DPI; use `--out DIR` for its destination. Compare the rendered page with the output, including item numbers, captions, symbols and reading order. Word presence alone does not validate these relationships.
53
39
 
54
- ## 4. Targeted re-extraction
40
+ For an OCR comparison or tuning experiment:
55
41
 
56
- Re-run just the flagged pages with an adjusted knob:
57
-
58
- ```
59
- pdf-to-md file.pdf --pages 27-31 --stdout --dpi 400
42
+ ```sh
43
+ pdf-to-md file.pdf page-N.md --pages N --ocr --dpi 400 --debug-words=page-N.words.jsonl
60
44
  ```
61
45
 
62
- Tuning flags (`--float-max-words`, `--para-gap`, `--heading-scale`, and the rest) are measured
63
- defaults — read README's tuning table for what each one controls and its default rather than
64
- guessing a value. Raising `--dpi` fixes chart/label corruption but can make scattered numerals on
65
- the same page worse, not better — try it on the flagged page only and compare, don't assume higher
66
- is strictly better.
46
+ OCR can recover image text and also damage correct embedded text. Higher DPI can improve small labels and worsen ordinary prose. Compare the result before accepting it. Read the package's `QUALITY.md` for tuning controls; their effects apply to OCR geometry.
67
47
 
68
- ## 5. Severity-1 audit
48
+ If a page needs manual repair, retain its source dump and page number. An image can preserve a diagram visually but does not recover editable text. Record unresolved character mappings or flattened tables alongside the affected content.
69
49
 
70
- Before trusting a conversion, run the audit — the only mechanically-checkable proof that no
71
- recognized word was silently dropped:
50
+ ## Audit the final Markdown
72
51
 
73
- ```
74
- pdf-to-md file.pdf out.md --debug-words=words.jsonl
52
+ ```sh
75
53
  pdf-to-md audit words.jsonl out.md
76
54
  ```
77
55
 
78
- Exits 1 if any word from the dump is missing from the markdown; `MISSING=0` in the printed summary
79
- line is the pass signal. `--debug-words` output is JSON Lines with no guaranteed key order —
80
- parse each line as JSON, never grep the file for text.
56
+ A successful audit with `MISSING=0` means the supplied extracted text was accounted for under the auditor's matching rules. Empty or malformed evidence is an error. It does not prove that the dump captured every visible word or that the result is an accurate transcription.
81
57
 
82
- ## 6. Machine-parseable markers
58
+ The auditor accepts `### pN` headings and `<!-- pN -->` comments. Audit the final file after edits and marker changes. Use the matching extraction dump for any pages replaced with a different extraction. Keep editorial additions separate while checking preservation, because added text can satisfy a missing occurrence.
83
59
 
84
- Two regex-matchable line prefixes carry content the geometry parked rather than lost:
60
+ Report preservation results, visual/structural checks, and unresolved issues separately. Avoid a blanket "lossless" claim based on a word audit.
85
61
 
86
- ```
87
- ^> \[(floats|review: [^\]]+)\] (.*)$
88
- ```
62
+ ## Programmatic repair
89
63
 
90
- `> [floats]` is a fragment (short line, margin text, etc.) set aside as decoration; `> [review:
91
- reason]` is a block the geometry couldn't order, with the reason inline. Both are visible in the
92
- markdown and safe to post-process (move, drop, re-attach) once matched.
64
+ `--format raw` returns an `Analysis` with per-page source items, blocks and a report. Text pages contain PDF text runs; OCR pages contain recognized words. Boxes use page-normalized coordinates, origin bottom-left, y increasing upward. Block `wordIndexes` link to the page's source list. Preserve that evidence when reordering blocks; rebuilding the source list from corrected output would make the audit circular.
93
65
 
94
- For programmatic convergence instead of eyeballing, `--format raw` returns the full `Analysis` as
95
- JSON: every word and block with its coordinates (normalized `[0,1]`, origin bottom-left, y up)
96
- plus the same `report` `--json` prints. Feed that into your own re-ordering or merge logic instead
97
- of re-parsing the markdown.
66
+ OCR uncertainty markers remain visible in Markdown:
67
+
68
+ ```text
69
+ > [floats] ...
70
+ > [review: reason] ...
71
+ ```
98
72
 
99
- ## Reference
73
+ Moving or deleting these blocks can remove real content. Check their source pages before changing them.
100
74
 
101
- Full flag list: `pdf-to-md --help`. Tuning table and design contract: the package README. Every
102
- exported function and option is documented at its declaration, visible in editor hover.
75
+ PDF compression is a separate operation. If requested, retain an identifiable original and verify text and visual material independently before replacing it. Lossy image compression can affect subsequent OCR and diagram checks even when extracted text stays identical.