@effortlessmotion/pdf-to-md 0.1.0

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 (189) hide show
  1. package/LICENSE +21 -0
  2. package/README.md +157 -0
  3. package/bin/cli.js +9 -0
  4. package/dist/cjs/analyze.d.cts +35 -0
  5. package/dist/cjs/analyze.d.ts +35 -0
  6. package/dist/cjs/analyze.js +620 -0
  7. package/dist/cjs/analyze.js.map +1 -0
  8. package/dist/cjs/audit.d.cts +38 -0
  9. package/dist/cjs/audit.d.ts +38 -0
  10. package/dist/cjs/audit.js +252 -0
  11. package/dist/cjs/audit.js.map +1 -0
  12. package/dist/cjs/cli/audit.d.cts +3 -0
  13. package/dist/cjs/cli/audit.d.ts +3 -0
  14. package/dist/cjs/cli/audit.js +269 -0
  15. package/dist/cjs/cli/audit.js.map +1 -0
  16. package/dist/cjs/cli/extract.d.cts +3 -0
  17. package/dist/cjs/cli/extract.d.ts +3 -0
  18. package/dist/cjs/cli/extract.js +704 -0
  19. package/dist/cjs/cli/extract.js.map +1 -0
  20. package/dist/cjs/cli/index.d.cts +5 -0
  21. package/dist/cjs/cli/index.d.ts +5 -0
  22. package/dist/cjs/cli/index.js +58 -0
  23. package/dist/cjs/cli/index.js.map +1 -0
  24. package/dist/cjs/cli/render.d.cts +3 -0
  25. package/dist/cjs/cli/render.d.ts +3 -0
  26. package/dist/cjs/cli/render.js +361 -0
  27. package/dist/cjs/cli/render.js.map +1 -0
  28. package/dist/cjs/cli/types.d.cts +10 -0
  29. package/dist/cjs/cli/types.d.ts +10 -0
  30. package/dist/cjs/cli/types.js +7 -0
  31. package/dist/cjs/cli/types.js.map +1 -0
  32. package/dist/cjs/cli.d.cts +2 -0
  33. package/dist/cjs/cli.d.ts +2 -0
  34. package/dist/cjs/cli.js +345 -0
  35. package/dist/cjs/cli.js.map +1 -0
  36. package/dist/cjs/collect.d.cts +6 -0
  37. package/dist/cjs/collect.d.ts +6 -0
  38. package/dist/cjs/collect.js +370 -0
  39. package/dist/cjs/collect.js.map +1 -0
  40. package/dist/cjs/dict.d.cts +8 -0
  41. package/dist/cjs/dict.d.ts +8 -0
  42. package/dist/cjs/dict.js +55 -0
  43. package/dist/cjs/dict.js.map +1 -0
  44. package/dist/cjs/emit.d.cts +27 -0
  45. package/dist/cjs/emit.d.ts +27 -0
  46. package/dist/cjs/emit.js +187 -0
  47. package/dist/cjs/emit.js.map +1 -0
  48. package/dist/cjs/engines/tesseract.d.cts +30 -0
  49. package/dist/cjs/engines/tesseract.d.ts +30 -0
  50. package/dist/cjs/engines/tesseract.js +650 -0
  51. package/dist/cjs/engines/tesseract.js.map +1 -0
  52. package/dist/cjs/errors.d.cts +21 -0
  53. package/dist/cjs/errors.d.ts +21 -0
  54. package/dist/cjs/errors.js +124 -0
  55. package/dist/cjs/errors.js.map +1 -0
  56. package/dist/cjs/extract.d.cts +88 -0
  57. package/dist/cjs/extract.d.ts +88 -0
  58. package/dist/cjs/extract.js +436 -0
  59. package/dist/cjs/extract.js.map +1 -0
  60. package/dist/cjs/geometry.d.cts +63 -0
  61. package/dist/cjs/geometry.d.ts +63 -0
  62. package/dist/cjs/geometry.js +521 -0
  63. package/dist/cjs/geometry.js.map +1 -0
  64. package/dist/cjs/index.d.cts +15 -0
  65. package/dist/cjs/index.d.ts +15 -0
  66. package/dist/cjs/index.js +56 -0
  67. package/dist/cjs/index.js.map +1 -0
  68. package/dist/cjs/lines.d.cts +3 -0
  69. package/dist/cjs/lines.d.ts +3 -0
  70. package/dist/cjs/lines.js +181 -0
  71. package/dist/cjs/lines.js.map +1 -0
  72. package/dist/cjs/markdown.d.cts +2 -0
  73. package/dist/cjs/markdown.d.ts +2 -0
  74. package/dist/cjs/markdown.js +222 -0
  75. package/dist/cjs/markdown.js.map +1 -0
  76. package/dist/cjs/package.json +1 -0
  77. package/dist/cjs/page-marker.d.cts +16 -0
  78. package/dist/cjs/page-marker.d.ts +16 -0
  79. package/dist/cjs/page-marker.js +85 -0
  80. package/dist/cjs/page-marker.js.map +1 -0
  81. package/dist/cjs/pdf-open.d.cts +12 -0
  82. package/dist/cjs/pdf-open.d.ts +12 -0
  83. package/dist/cjs/pdf-open.js +257 -0
  84. package/dist/cjs/pdf-open.js.map +1 -0
  85. package/dist/cjs/raster.d.cts +26 -0
  86. package/dist/cjs/raster.d.ts +26 -0
  87. package/dist/cjs/raster.js +262 -0
  88. package/dist/cjs/raster.js.map +1 -0
  89. package/dist/cjs/render-analysis.d.cts +25 -0
  90. package/dist/cjs/render-analysis.d.ts +25 -0
  91. package/dist/cjs/render-analysis.js +319 -0
  92. package/dist/cjs/render-analysis.js.map +1 -0
  93. package/dist/cjs/report.d.cts +12 -0
  94. package/dist/cjs/report.d.ts +12 -0
  95. package/dist/cjs/report.js +392 -0
  96. package/dist/cjs/report.js.map +1 -0
  97. package/dist/cjs/resolve.d.cts +1 -0
  98. package/dist/cjs/resolve.d.ts +1 -0
  99. package/dist/cjs/resolve.js +18 -0
  100. package/dist/cjs/resolve.js.map +1 -0
  101. package/dist/cjs/types.d.cts +171 -0
  102. package/dist/cjs/types.d.ts +171 -0
  103. package/dist/cjs/types.js +9 -0
  104. package/dist/cjs/types.js.map +1 -0
  105. package/dist/cjs/validate.d.cts +14 -0
  106. package/dist/cjs/validate.d.ts +14 -0
  107. package/dist/cjs/validate.js +96 -0
  108. package/dist/cjs/validate.js.map +1 -0
  109. package/dist/esm/analyze.d.ts +35 -0
  110. package/dist/esm/analyze.js +270 -0
  111. package/dist/esm/analyze.js.map +1 -0
  112. package/dist/esm/audit.d.ts +38 -0
  113. package/dist/esm/audit.js +113 -0
  114. package/dist/esm/audit.js.map +1 -0
  115. package/dist/esm/cli/audit.d.ts +3 -0
  116. package/dist/esm/cli/audit.js +20 -0
  117. package/dist/esm/cli/audit.js.map +1 -0
  118. package/dist/esm/cli/extract.d.ts +3 -0
  119. package/dist/esm/cli/extract.js +313 -0
  120. package/dist/esm/cli/extract.js.map +1 -0
  121. package/dist/esm/cli/index.d.ts +5 -0
  122. package/dist/esm/cli/index.js +7 -0
  123. package/dist/esm/cli/index.js.map +1 -0
  124. package/dist/esm/cli/render.d.ts +3 -0
  125. package/dist/esm/cli/render.js +77 -0
  126. package/dist/esm/cli/render.js.map +1 -0
  127. package/dist/esm/cli/types.d.ts +10 -0
  128. package/dist/esm/cli/types.js +3 -0
  129. package/dist/esm/cli/types.js.map +1 -0
  130. package/dist/esm/cli.d.ts +2 -0
  131. package/dist/esm/cli.js +90 -0
  132. package/dist/esm/cli.js.map +1 -0
  133. package/dist/esm/collect.d.ts +6 -0
  134. package/dist/esm/collect.js +80 -0
  135. package/dist/esm/collect.js.map +1 -0
  136. package/dist/esm/dict.d.ts +8 -0
  137. package/dist/esm/dict.js +28 -0
  138. package/dist/esm/dict.js.map +1 -0
  139. package/dist/esm/emit.d.ts +27 -0
  140. package/dist/esm/emit.js +131 -0
  141. package/dist/esm/emit.js.map +1 -0
  142. package/dist/esm/engines/tesseract.d.ts +30 -0
  143. package/dist/esm/engines/tesseract.js +190 -0
  144. package/dist/esm/engines/tesseract.js.map +1 -0
  145. package/dist/esm/errors.d.ts +21 -0
  146. package/dist/esm/errors.js +9 -0
  147. package/dist/esm/errors.js.map +1 -0
  148. package/dist/esm/extract.d.ts +88 -0
  149. package/dist/esm/extract.js +147 -0
  150. package/dist/esm/extract.js.map +1 -0
  151. package/dist/esm/geometry.d.ts +63 -0
  152. package/dist/esm/geometry.js +279 -0
  153. package/dist/esm/geometry.js.map +1 -0
  154. package/dist/esm/index.d.ts +15 -0
  155. package/dist/esm/index.js +10 -0
  156. package/dist/esm/index.js.map +1 -0
  157. package/dist/esm/lines.d.ts +3 -0
  158. package/dist/esm/lines.js +67 -0
  159. package/dist/esm/lines.js.map +1 -0
  160. package/dist/esm/markdown.d.ts +2 -0
  161. package/dist/esm/markdown.js +153 -0
  162. package/dist/esm/markdown.js.map +1 -0
  163. package/dist/esm/package.json +1 -0
  164. package/dist/esm/page-marker.d.ts +16 -0
  165. package/dist/esm/page-marker.js +40 -0
  166. package/dist/esm/page-marker.js.map +1 -0
  167. package/dist/esm/pdf-open.d.ts +12 -0
  168. package/dist/esm/pdf-open.js +61 -0
  169. package/dist/esm/pdf-open.js.map +1 -0
  170. package/dist/esm/raster.d.ts +26 -0
  171. package/dist/esm/raster.js +60 -0
  172. package/dist/esm/raster.js.map +1 -0
  173. package/dist/esm/render-analysis.d.ts +25 -0
  174. package/dist/esm/render-analysis.js +115 -0
  175. package/dist/esm/render-analysis.js.map +1 -0
  176. package/dist/esm/report.d.ts +12 -0
  177. package/dist/esm/report.js +182 -0
  178. package/dist/esm/report.js.map +1 -0
  179. package/dist/esm/resolve.d.ts +1 -0
  180. package/dist/esm/resolve.js +7 -0
  181. package/dist/esm/resolve.js.map +1 -0
  182. package/dist/esm/types.d.ts +171 -0
  183. package/dist/esm/types.js +6 -0
  184. package/dist/esm/types.js.map +1 -0
  185. package/dist/esm/validate.d.ts +14 -0
  186. package/dist/esm/validate.js +56 -0
  187. package/dist/esm/validate.js.map +1 -0
  188. package/package.json +67 -0
  189. package/skills/pdf-to-md/SKILL.md +102 -0
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Projects/pdf-to-md/src/analyze.ts"],"sourcesContent":["// analyze() plus the OCR-path core (buildOcrAnalysisPages) that extractOcr\n// shares with it — same buildPageOCR -> emitPage -> renderPagesToMarkdown pipeline, so OCR-path markdown is byte-identical by construction, not two implementations kept in sync by hand. The TEXT path is NOT shared this way (see analyzeText).\nimport { collect, pageCount } from './collect.ts';\nimport { dictAvailable, dictUnavailableWarning } from './dict.ts';\nimport type { EmitBlock } from './emit.ts';\nimport { emitPage, renderEmitBlock } from './emit.ts';\n// Type-only: erased at compile time, so this does NOT create a runtime\n// circular dependency even though src/extract.ts imports (value-level) from this module.\nimport type { ExtractOcrOptions, ExtractTextOptions, OcrWordInput } from './extract.ts';\nimport type { PageOCR, Tuning, Word } from './geometry.ts';\nimport { buildPageOCR, DEFAULT_TUNING, headingLevels } from './geometry.ts';\nimport { toLines } from './lines.ts';\nimport { buildMarkdown } from './markdown.ts';\nimport { renderPagesToMarkdown } from './render-analysis.ts';\nimport { buildOcrReport, buildReport, OUTPUT_VERSION } from './report.ts';\nimport type { Analysis, AnalysisBlock, AnalysisPage, AnalysisWord, Line, PageLines } from './types.ts';\nimport { validateOcrAnalysisInput, validatePages, validateWordsInput } from './validate.ts';\n\n/** analyze(pdf) with the default/'text' path — OCR-only fields don't exist\n * on this branch, so passing e.g. `tuning` without `path: 'ocr'` is a compile error. */\nexport type AnalyzeTextOptions = ExtractTextOptions & { path?: 'text' };\n\n/** analyze(pdf, {path:'ocr'}) — same fields as {@link ExtractOcrOptions},\n * unlocked only once `path: 'ocr'` is set. */\nexport type AnalyzeOcrOptions = ExtractOcrOptions & { path: 'ocr' };\n\n/** Options for {@link analyze}: which fields type-check depends on `path`\n * (ignored, forced 'ocr', when `input` is a `{ words }` replay). */\nexport type AnalyzeOptions = AnalyzeTextOptions | AnalyzeOcrOptions;\n\nfunction toAnalysisWord(w: OcrWordInput): AnalysisWord {\n const box = { x: w.x, y: w.y, w: w.w, h: w.h };\n return w.confidence === undefined ? { text: w.text, box } : { text: w.text, box, confidence: w.confidence };\n}\n\n// EmitBlock -> AnalysisBlock: collapses emit.ts's 'quote' kind (a pull-quote)\n// into Analysis's 'paragraph' type, `text` keeping its literal '> ' prefix — see AnalysisBlock's doc comment (src/types.ts) for why.\nfunction toAnalysisBlock(b: EmitBlock, wordIndex: Map<Word, number>): AnalysisBlock {\n const wordIndexes = b.words.map((w) => wordIndex.get(w) as number);\n if (b.kind === 'heading') return { type: 'heading', text: b.text, level: b.level, wordIndexes };\n if (b.kind === 'review') return { type: 'review', text: renderEmitBlock(b), wordIndexes };\n return { type: 'paragraph', text: renderEmitBlock(b), wordIndexes };\n}\n\n/** The OCR-path core shared by extractOcr and analyze(..., {path:'ocr'}):\n * recognizes (or replays) words and runs the unchanged buildPageOCR/emitPage pipeline. Word IDENTITY (object reference), not value, carries indices through into the emitted blocks — see `geometry.ts`'s OcrLine.words. */\nexport async function buildOcrAnalysisPages(input: { pdfPath: string } | { words: OcrWordInput[] }, opts: ExtractOcrOptions, warn: (m: string) => void): Promise<{ pages: AnalysisPage[]; words: OcrWordInput[] }> {\n const tuning: Tuning = { ...DEFAULT_TUNING, ...opts.tuning };\n const validated = validateOcrAnalysisInput(input);\n\n let words: OcrWordInput[];\n let first: number;\n let last: number;\n if ('words' in validated) {\n // Replay: page range from --pages if given, else every page up to the highest one the dump mentions.\n words = validated.words;\n first = opts.pages?.first ?? 1;\n last = opts.pages?.last ?? opts.pages?.first ?? Math.max(1, ...words.map((w) => w.page));\n } else {\n const { recognizePdf } = await import('./engines/tesseract.ts');\n const totalPages = await pageCount(validated.pdfPath);\n const rawFirst = opts.pages?.first ?? 1;\n const rawLast = opts.pages?.last ?? (opts.pages ? rawFirst : totalPages);\n first = Math.min(Math.max(1, rawFirst), Math.max(1, totalPages));\n last = Math.min(Math.max(first, rawLast), totalPages);\n const result = await recognizePdf(validated.pdfPath, first, last, {\n dpi: opts.dpi,\n onProgress: opts.onProgress,\n onWarning: warn,\n });\n words = result.words;\n }\n\n opts.onWords?.(words);\n\n // Stable per-page index, assigned in recognition/replay order — this IS\n // the page's flat `words` list order; `wordIndex` recovers it later since geometry.ts never clones a Word (object identity survives row-building).\n const byPage = new Map<number, OcrWordInput[]>();\n const wordIndex = new Map<Word, number>();\n for (const w of words) {\n const list = byPage.get(w.page);\n if (list) {\n wordIndex.set(w, list.length);\n list.push(w);\n } else {\n wordIndex.set(w, 0);\n byPage.set(w.page, [w]);\n }\n }\n\n const pageOcrs = new Map<number, PageOCR>();\n for (let pn = first; pn <= last; pn++) pageOcrs.set(pn, buildPageOCR(byPage.get(pn) ?? [], tuning));\n const { levelOf } = headingLevels([...pageOcrs.values()], tuning);\n\n const pages: AnalysisPage[] = [];\n for (let pn = first; pn <= last; pn++) {\n const p = pageOcrs.get(pn) as PageOCR;\n // An in-range page with no recognized words stays visible (marker +\n // warning), never silently absent — same rule as extractOcr.\n if (!byPage.has(pn)) warn(`p${pn}: no recognized words, page emitted with no OCR text`);\n\n const blocks: AnalysisBlock[] = emitPage(p, levelOf, tuning).map((b) => toAnalysisBlock(b, wordIndex));\n // Decorative fragments always render as one visible \"> [floats]\" aside\n // per page (matching the one printed line) — see geometry.ts's PageOCR.floats.\n if (p.floats.length) {\n blocks.push({\n type: 'float',\n text: `> [floats] ${p.floats.map((f) => f.text).join(' · ')}`,\n wordIndexes: p.floats.flatMap((f) => f.words).map((w) => wordIndex.get(w) as number),\n });\n }\n\n pages.push({\n page: pn,\n words: (byPage.get(pn) ?? []).map(toAnalysisWord),\n blocks,\n });\n }\n\n return { pages, words };\n}\n\nasync function analyzeOcr(input: { pdfPath: string } | { words: OcrWordInput[] }, opts: ExtractOcrOptions): Promise<Analysis> {\n const warn = (m: string) => opts.onWarning?.(m);\n if (!dictAvailable()) warn(dictUnavailableWarning());\n const { pages, words } = await buildOcrAnalysisPages(input, opts, warn);\n const md = renderPagesToMarkdown(pages, { forcePageMarkers: true, pageMarkers: true });\n const report = buildOcrReport({ md, fallback: opts.fallback, words });\n return { outputVersion: OUTPUT_VERSION, report, pages };\n}\n\n// TEXT path: deliberately NOT routed through the OCR shared-core pattern —\n// buildMarkdown's reflow/hyphen-rejoin/list rules don't fit the 4-type Analysis block union, so this is an independent, intentionally coarser reconstruction (one word/block per visual LINE) — honest, not byte-identical to extractText (see AnalysisPage, src/types.ts).\n\n// Converts a text-path Line's page-point geometry (src/lines.ts) into the\n// same normalized box contract the OCR path uses (see AnalysisPage.words) — only the coordinate SYSTEM changes; `box.y` anchors on the line's baseline, `box.w` stays 0.\nfunction textLineBox(l: Line, dims: { width: number; height: number }): AnalysisWord['box'] {\n const { width, height } = dims;\n return {\n x: width > 0 ? l.x / width : 0,\n y: height > 0 ? (height - l.y) / height : 0,\n w: 0,\n h: height > 0 ? l.h / height : 0,\n };\n}\n\nasync function analyzeText(pdfPath: string, opts: ExtractTextOptions): Promise<Analysis> {\n const warn = (m: string) => opts.onWarning?.(m);\n if (!dictAvailable()) warn(dictUnavailableWarning());\n const range = opts.pages ? { first: opts.pages.first, last: opts.pages.last ?? opts.pages.first } : {};\n const { numPages, pages, heightChars } = await collect(pdfPath, range);\n const pageDims = new Map(pages.map((p) => [p.page, { width: p.width, height: p.height }]));\n const pageLines: PageLines[] = pages.map((p) => ({ page: p.page, lines: toLines(p) }));\n const pageMarkers = opts.pageMarkers ?? false;\n const { md, stats, bodyH, headingSizes } = buildMarkdown(pageLines, heightChars, { pageMarkers });\n const statsMd = pageMarkers ? md : buildMarkdown(pageLines, heightChars, { pageMarkers: true }).md;\n\n const hybridPages = pages.filter((p) => p.largeImage).map((p) => p.page);\n for (const n of hybridPages) {\n warn(`p${n}: text layer used but page carries large image(s); if it has baked-in text this output won't have it — compare (don't blindly replace) against --pages ${n} --ocr, which can itself introduce errors`);\n }\n\n // Report is measured from the REAL markdown above — accurate regardless of\n // the coarser block/word reconstruction below.\n const report = buildReport({ numPages, md, statsMd, stats, bodyH, headingSizes, hybridPages });\n\n // headingSizes is offset-independent — reused directly, then re-leveled\n // here with offset 0 regardless of `opts.pageMarkers` (see AnalysisPage).\n const levelOf = new Map(headingSizes.slice(0, 6).map((h, i) => [h, Math.min(6, i + 1)]));\n const analysisPages: AnalysisPage[] = pageLines.map(({ page, lines }) => {\n const dims = pageDims.get(page) as { width: number; height: number };\n const words: AnalysisWord[] = lines.map((l) => ({ text: l.text, box: textLineBox(l, dims) }));\n const blocks: AnalysisBlock[] = lines.map((l, i) => {\n const level = levelOf.get(l.h);\n return level !== undefined ? { type: 'heading' as const, text: l.text, level, wordIndexes: [i] } : { type: 'paragraph' as const, text: l.text, wordIndexes: [i] };\n });\n return { page, words, blocks };\n });\n\n return { outputVersion: OUTPUT_VERSION, report, pages: analysisPages };\n}\n\n/**\n * Analyze a PDF (or replay an already-recognized OCR word dump) into\n * `Analysis`: plain, JSON-serializable per-page `words`/`blocks` plus the same QA `report` the other entry points return. Render with toMarkdown()/toText(), or hand-edit/splice it (no merge API — see toMarkdown's doc comment).\n *\n * @param input a bare pdf path, or `{ words }` to replay a dump (always analyzed as OCR) — unlike extractOcr, never `{ pdfPath }`\n * @param opts see {@link AnalyzeOptions}; for a pdf path, `opts.path` picks 'text' (default) or 'ocr' — analyze() makes no auto-fallback decision (that's pdfToMarkdown's job)\n */\nexport async function analyze(input: string | { words: OcrWordInput[] }, opts: AnalyzeOptions = {}): Promise<Analysis> {\n validatePages(opts.pages);\n if (typeof input !== 'string') return analyzeOcr({ words: validateWordsInput(input) }, opts);\n if (opts.path === 'ocr') return analyzeOcr({ pdfPath: input }, opts);\n return analyzeText(input, opts);\n}\n"],"names":["collect","pageCount","dictAvailable","dictUnavailableWarning","emitPage","renderEmitBlock","buildPageOCR","DEFAULT_TUNING","headingLevels","toLines","buildMarkdown","renderPagesToMarkdown","buildOcrReport","buildReport","OUTPUT_VERSION","validateOcrAnalysisInput","validatePages","validateWordsInput","toAnalysisWord","w","box","x","y","h","confidence","undefined","text","toAnalysisBlock","b","wordIndex","wordIndexes","words","map","get","kind","type","level","buildOcrAnalysisPages","input","opts","warn","byPage","tuning","validated","first","last","pages","Math","max","page","recognizePdf","totalPages","pdfPath","rawFirst","rawLast","min","result","dpi","onProgress","onWarning","onWords","Map","list","set","length","push","pageOcrs","pn","levelOf","values","p","has","blocks","floats","f","join","flatMap","analyzeOcr","m","md","forcePageMarkers","pageMarkers","report","fallback","outputVersion","textLineBox","l","dims","width","height","analyzeText","range","numPages","heightChars","pageDims","pageLines","lines","stats","bodyH","headingSizes","statsMd","hybridPages","filter","largeImage","n","slice","i","analysisPages","analyze","path"],"mappings":"AAAA,2EAA2E;AAC3E,kPAAkP;AAClP,SAASA,OAAO,EAAEC,SAAS,QAAQ,eAAe;AAClD,SAASC,aAAa,EAAEC,sBAAsB,QAAQ,YAAY;AAElE,SAASC,QAAQ,EAAEC,eAAe,QAAQ,YAAY;AAKtD,SAASC,YAAY,EAAEC,cAAc,EAAEC,aAAa,QAAQ,gBAAgB;AAC5E,SAASC,OAAO,QAAQ,aAAa;AACrC,SAASC,aAAa,QAAQ,gBAAgB;AAC9C,SAASC,qBAAqB,QAAQ,uBAAuB;AAC7D,SAASC,cAAc,EAAEC,WAAW,EAAEC,cAAc,QAAQ,cAAc;AAE1E,SAASC,wBAAwB,EAAEC,aAAa,EAAEC,kBAAkB,QAAQ,gBAAgB;AAc5F,SAASC,eAAeC,CAAe;IACrC,MAAMC,MAAM;QAAEC,GAAGF,EAAEE,CAAC;QAAEC,GAAGH,EAAEG,CAAC;QAAEH,GAAGA,EAAEA,CAAC;QAAEI,GAAGJ,EAAEI,CAAC;IAAC;IAC7C,OAAOJ,EAAEK,UAAU,KAAKC,YAAY;QAAEC,MAAMP,EAAEO,IAAI;QAAEN;IAAI,IAAI;QAAEM,MAAMP,EAAEO,IAAI;QAAEN;QAAKI,YAAYL,EAAEK,UAAU;IAAC;AAC5G;AAEA,8EAA8E;AAC9E,qIAAqI;AACrI,SAASG,gBAAgBC,CAAY,EAAEC,SAA4B;IACjE,MAAMC,cAAcF,EAAEG,KAAK,CAACC,GAAG,CAAC,CAACb,IAAMU,UAAUI,GAAG,CAACd;IACrD,IAAIS,EAAEM,IAAI,KAAK,WAAW,OAAO;QAAEC,MAAM;QAAWT,MAAME,EAAEF,IAAI;QAAEU,OAAOR,EAAEQ,KAAK;QAAEN;IAAY;IAC9F,IAAIF,EAAEM,IAAI,KAAK,UAAU,OAAO;QAAEC,MAAM;QAAUT,MAAMrB,gBAAgBuB;QAAIE;IAAY;IACxF,OAAO;QAAEK,MAAM;QAAaT,MAAMrB,gBAAgBuB;QAAIE;IAAY;AACpE;AAEA;0NAC0N,GAC1N,OAAO,eAAeO,sBAAsBC,KAAsD,EAAEC,IAAuB,EAAEC,IAAyB;QA6C/EC;QAlBrEF;IA1BA,MAAMG,SAAiB;QAAE,GAAGnC,cAAc;QAAE,GAAGgC,KAAKG,MAAM;IAAC;IAC3D,MAAMC,YAAY5B,yBAAyBuB;IAE3C,IAAIP;IACJ,IAAIa;IACJ,IAAIC;IACJ,IAAI,WAAWF,WAAW;kBAIjBJ;YADCA,aACDA,cAAoBA;QAH3B,qGAAqG;QACrGR,QAAQY,UAAUZ,KAAK;QACvBa,iBAAQL,cAAAA,KAAKO,KAAK,cAAVP,kCAAAA,YAAYK,KAAK,uCAAI;QAC7BC,QAAON,kBAAAA,eAAAA,KAAKO,KAAK,cAAVP,mCAAAA,aAAYM,IAAI,0CAAIN,eAAAA,KAAKO,KAAK,cAAVP,mCAAAA,aAAYK,KAAK,cAArCL,mBAAAA,QAAyCQ,KAAKC,GAAG,CAAC,MAAMjB,MAAMC,GAAG,CAAC,CAACb,IAAMA,EAAE8B,IAAI;IACxF,OAAO;;YAGYV,cACDA;QAHhB,MAAM,EAAEW,YAAY,EAAE,GAAG,MAAM,MAAM,CAAC;QACtC,MAAMC,aAAa,MAAMlD,UAAU0C,UAAUS,OAAO;QACpD,MAAMC,qBAAWd,eAAAA,KAAKO,KAAK,cAAVP,mCAAAA,aAAYK,KAAK,yCAAI;QACtC,MAAMU,oBAAUf,eAAAA,KAAKO,KAAK,cAAVP,mCAAAA,aAAYM,IAAI,yCAAKN,KAAKO,KAAK,GAAGO,WAAWF;QAC7DP,QAAQG,KAAKQ,GAAG,CAACR,KAAKC,GAAG,CAAC,GAAGK,WAAWN,KAAKC,GAAG,CAAC,GAAGG;QACpDN,OAAOE,KAAKQ,GAAG,CAACR,KAAKC,GAAG,CAACJ,OAAOU,UAAUH;QAC1C,MAAMK,SAAS,MAAMN,aAAaP,UAAUS,OAAO,EAAER,OAAOC,MAAM;YAChEY,KAAKlB,KAAKkB,GAAG;YACbC,YAAYnB,KAAKmB,UAAU;YAC3BC,WAAWnB;QACb;QACAT,QAAQyB,OAAOzB,KAAK;IACtB;KAEAQ,gBAAAA,KAAKqB,OAAO,cAAZrB,oCAAAA,mBAAAA,MAAeR;IAEf,wEAAwE;IACxE,mJAAmJ;IACnJ,MAAMU,SAAS,IAAIoB;IACnB,MAAMhC,YAAY,IAAIgC;IACtB,KAAK,MAAM1C,KAAKY,MAAO;QACrB,MAAM+B,OAAOrB,OAAOR,GAAG,CAACd,EAAE8B,IAAI;QAC9B,IAAIa,MAAM;YACRjC,UAAUkC,GAAG,CAAC5C,GAAG2C,KAAKE,MAAM;YAC5BF,KAAKG,IAAI,CAAC9C;QACZ,OAAO;YACLU,UAAUkC,GAAG,CAAC5C,GAAG;YACjBsB,OAAOsB,GAAG,CAAC5C,EAAE8B,IAAI,EAAE;gBAAC9B;aAAE;QACxB;IACF;IAEA,MAAM+C,WAAW,IAAIL;IACrB,IAAK,IAAIM,KAAKvB,OAAOuB,MAAMtB,MAAMsB,KAAMD,SAASH,GAAG,CAACI,IAAI7D,cAAamC,cAAAA,OAAOR,GAAG,CAACkC,iBAAX1B,yBAAAA,cAAkB,EAAE,EAAEC;IAC3F,MAAM,EAAE0B,OAAO,EAAE,GAAG5D,cAAc;WAAI0D,SAASG,MAAM;KAAG,EAAE3B;IAE1D,MAAMI,QAAwB,EAAE;IAChC,IAAK,IAAIqB,KAAKvB,OAAOuB,MAAMtB,MAAMsB,KAAM;YAmB3B1B;QAlBV,MAAM6B,IAAIJ,SAASjC,GAAG,CAACkC;QACvB,oEAAoE;QACpE,6DAA6D;QAC7D,IAAI,CAAC1B,OAAO8B,GAAG,CAACJ,KAAK3B,KAAK,CAAC,CAAC,EAAE2B,GAAG,oDAAoD,CAAC;QAEtF,MAAMK,SAA0BpE,SAASkE,GAAGF,SAAS1B,QAAQV,GAAG,CAAC,CAACJ,IAAMD,gBAAgBC,GAAGC;QAC3F,uEAAuE;QACvE,+EAA+E;QAC/E,IAAIyC,EAAEG,MAAM,CAACT,MAAM,EAAE;YACnBQ,OAAOP,IAAI,CAAC;gBACV9B,MAAM;gBACNT,MAAM,CAAC,WAAW,EAAE4C,EAAEG,MAAM,CAACzC,GAAG,CAAC,CAAC0C,IAAMA,EAAEhD,IAAI,EAAEiD,IAAI,CAAC,QAAQ;gBAC7D7C,aAAawC,EAAEG,MAAM,CAACG,OAAO,CAAC,CAACF,IAAMA,EAAE3C,KAAK,EAAEC,GAAG,CAAC,CAACb,IAAMU,UAAUI,GAAG,CAACd;YACzE;QACF;QAEA2B,MAAMmB,IAAI,CAAC;YACThB,MAAMkB;YACNpC,OAAO,EAACU,eAAAA,OAAOR,GAAG,CAACkC,iBAAX1B,0BAAAA,eAAkB,EAAE,EAAET,GAAG,CAACd;YAClCsD;QACF;IACF;IAEA,OAAO;QAAE1B;QAAOf;IAAM;AACxB;AAEA,eAAe8C,WAAWvC,KAAsD,EAAEC,IAAuB;IACvG,MAAMC,OAAO,CAACsC;YAAcvC;gBAAAA,kBAAAA,KAAKoB,SAAS,cAAdpB,sCAAAA,qBAAAA,MAAiBuC;;IAC7C,IAAI,CAAC5E,iBAAiBsC,KAAKrC;IAC3B,MAAM,EAAE2C,KAAK,EAAEf,KAAK,EAAE,GAAG,MAAMM,sBAAsBC,OAAOC,MAAMC;IAClE,MAAMuC,KAAKpE,sBAAsBmC,OAAO;QAAEkC,kBAAkB;QAAMC,aAAa;IAAK;IACpF,MAAMC,SAAStE,eAAe;QAAEmE;QAAII,UAAU5C,KAAK4C,QAAQ;QAAEpD;IAAM;IACnE,OAAO;QAAEqD,eAAetE;QAAgBoE;QAAQpC;IAAM;AACxD;AAEA,2EAA2E;AAC3E,4QAA4Q;AAE5Q,0EAA0E;AAC1E,yKAAyK;AACzK,SAASuC,YAAYC,CAAO,EAAEC,IAAuC;IACnE,MAAM,EAAEC,KAAK,EAAEC,MAAM,EAAE,GAAGF;IAC1B,OAAO;QACLlE,GAAGmE,QAAQ,IAAIF,EAAEjE,CAAC,GAAGmE,QAAQ;QAC7BlE,GAAGmE,SAAS,IAAI,AAACA,CAAAA,SAASH,EAAEhE,CAAC,AAADA,IAAKmE,SAAS;QAC1CtE,GAAG;QACHI,GAAGkE,SAAS,IAAIH,EAAE/D,CAAC,GAAGkE,SAAS;IACjC;AACF;AAEA,eAAeC,YAAYtC,OAAe,EAAEb,IAAwB;QAGNA,kBAIxCA;IANpB,MAAMC,OAAO,CAACsC;YAAcvC;gBAAAA,kBAAAA,KAAKoB,SAAS,cAAdpB,sCAAAA,qBAAAA,MAAiBuC;;IAC7C,IAAI,CAAC5E,iBAAiBsC,KAAKrC;IAC3B,MAAMwF,QAAQpD,KAAKO,KAAK,GAAG;QAAEF,OAAOL,KAAKO,KAAK,CAACF,KAAK;QAAEC,IAAI,GAAEN,mBAAAA,KAAKO,KAAK,CAACD,IAAI,cAAfN,8BAAAA,mBAAmBA,KAAKO,KAAK,CAACF,KAAK;IAAC,IAAI,CAAC;IACrG,MAAM,EAAEgD,QAAQ,EAAE9C,KAAK,EAAE+C,WAAW,EAAE,GAAG,MAAM7F,QAAQoD,SAASuC;IAChE,MAAMG,WAAW,IAAIjC,IAAIf,MAAMd,GAAG,CAAC,CAACsC,IAAM;YAACA,EAAErB,IAAI;YAAE;gBAAEuC,OAAOlB,EAAEkB,KAAK;gBAAEC,QAAQnB,EAAEmB,MAAM;YAAC;SAAE;IACxF,MAAMM,YAAyBjD,MAAMd,GAAG,CAAC,CAACsC,IAAO,CAAA;YAAErB,MAAMqB,EAAErB,IAAI;YAAE+C,OAAOvF,QAAQ6D;QAAG,CAAA;IACnF,MAAMW,eAAc1C,oBAAAA,KAAK0C,WAAW,cAAhB1C,+BAAAA,oBAAoB;IACxC,MAAM,EAAEwC,EAAE,EAAEkB,KAAK,EAAEC,KAAK,EAAEC,YAAY,EAAE,GAAGzF,cAAcqF,WAAWF,aAAa;QAAEZ;IAAY;IAC/F,MAAMmB,UAAUnB,cAAcF,KAAKrE,cAAcqF,WAAWF,aAAa;QAAEZ,aAAa;IAAK,GAAGF,EAAE;IAElG,MAAMsB,cAAcvD,MAAMwD,MAAM,CAAC,CAAChC,IAAMA,EAAEiC,UAAU,EAAEvE,GAAG,CAAC,CAACsC,IAAMA,EAAErB,IAAI;IACvE,KAAK,MAAMuD,KAAKH,YAAa;QAC3B7D,KAAK,CAAC,CAAC,EAAEgE,EAAE,uJAAuJ,EAAEA,EAAE,yCAAyC,CAAC;IAClN;IAEA,2EAA2E;IAC3E,+CAA+C;IAC/C,MAAMtB,SAASrE,YAAY;QAAE+E;QAAUb;QAAIqB;QAASH;QAAOC;QAAOC;QAAcE;IAAY;IAE5F,wEAAwE;IACxE,0EAA0E;IAC1E,MAAMjC,UAAU,IAAIP,IAAIsC,aAAaM,KAAK,CAAC,GAAG,GAAGzE,GAAG,CAAC,CAACT,GAAGmF,IAAM;YAACnF;YAAGwB,KAAKQ,GAAG,CAAC,GAAGmD,IAAI;SAAG;IACtF,MAAMC,gBAAgCZ,UAAU/D,GAAG,CAAC,CAAC,EAAEiB,IAAI,EAAE+C,KAAK,EAAE;QAClE,MAAMT,OAAOO,SAAS7D,GAAG,CAACgB;QAC1B,MAAMlB,QAAwBiE,MAAMhE,GAAG,CAAC,CAACsD,IAAO,CAAA;gBAAE5D,MAAM4D,EAAE5D,IAAI;gBAAEN,KAAKiE,YAAYC,GAAGC;YAAM,CAAA;QAC1F,MAAMf,SAA0BwB,MAAMhE,GAAG,CAAC,CAACsD,GAAGoB;YAC5C,MAAMtE,QAAQgC,QAAQnC,GAAG,CAACqD,EAAE/D,CAAC;YAC7B,OAAOa,UAAUX,YAAY;gBAAEU,MAAM;gBAAoBT,MAAM4D,EAAE5D,IAAI;gBAAEU;gBAAON,aAAa;oBAAC4E;iBAAE;YAAC,IAAI;gBAAEvE,MAAM;gBAAsBT,MAAM4D,EAAE5D,IAAI;gBAAEI,aAAa;oBAAC4E;iBAAE;YAAC;QAClK;QACA,OAAO;YAAEzD;YAAMlB;YAAOyC;QAAO;IAC/B;IAEA,OAAO;QAAEY,eAAetE;QAAgBoE;QAAQpC,OAAO6D;IAAc;AACvE;AAEA;;;;;;CAMC,GACD,OAAO,eAAeC,QAAQtE,KAAyC,EAAEC,OAAuB,CAAC,CAAC;IAChGvB,cAAcuB,KAAKO,KAAK;IACxB,IAAI,OAAOR,UAAU,UAAU,OAAOuC,WAAW;QAAE9C,OAAOd,mBAAmBqB;IAAO,GAAGC;IACvF,IAAIA,KAAKsE,IAAI,KAAK,OAAO,OAAOhC,WAAW;QAAEzB,SAASd;IAAM,GAAGC;IAC/D,OAAOmD,YAAYpD,OAAOC;AAC5B"}
@@ -0,0 +1,38 @@
1
+ /** auditWords()'s result — the severity-1 no-silent-loss check. */
2
+ export interface AuditResult {
3
+ /** JSON lines successfully parsed from the word dump */
4
+ parsed: number;
5
+ /** lines in the dump that were not parseable JSON, skipped rather than
6
+ * failing the whole audit (same policy `--words-json` itself uses) */
7
+ nonJsonSkipped: number;
8
+ /** distinct page numbers present in the dump */
9
+ pages: number;
10
+ /** total recognized words across all pages in the dump */
11
+ words: number;
12
+ /** words present in the dump but not found verbatim on their page, BEFORE
13
+ * hyphen-join/split repair is applied; always `>= missing` */
14
+ strictDeficits: number;
15
+ /** words genuinely absent from the markdown after repair-checking; `0` is
16
+ * the only passing value (`pdf-to-md audit` exits 1 otherwise) */
17
+ missing: number;
18
+ /** per-page detail for every page with at least one missing word (capped
19
+ * sample of the missing tokens per page) */
20
+ perPage: {
21
+ page: number;
22
+ missing: string[];
23
+ }[];
24
+ /** the one-line `parsed=… pages=… words=… strict_deficits=… MISSING=…`
25
+ * summary, also the last entry of `lines` */
26
+ summaryLine: string;
27
+ /** every line `pdf-to-md audit` prints, in order: one per page with
28
+ * missing words, then `summaryLine` */
29
+ lines: string[];
30
+ }
31
+ /**
32
+ * Audit a --debug-words dump against a markdown file: every recognized word
33
+ * must land somewhere in its page's markdown. `result.missing === 0` is the pass/fail signal.
34
+ *
35
+ * @param wordsPath path to a --debug-words JSON-lines dump (`{page, text, ...}` per line)
36
+ * @param mdPath path to the markdown file to check the dump against
37
+ */
38
+ export declare function auditWords(wordsPath: string, mdPath: string): AuditResult;
@@ -0,0 +1,113 @@
1
+ // audit.ts — the severity-1 audit instrument (README's "Design contract"),
2
+ // shipped as part of the product, not just a path into the source tree.
3
+ //
4
+ // Every recognized word (a --debug-words dump, or an Analysis page's
5
+ // `words`) must appear in its page's markdown — normalized, compared as a per-page multiset, with substring repair for hyphen joins.
6
+ //
7
+ // NOTE: no producer guarantees JSON key order in a --debug-words dump —
8
+ // parse, never grep.
9
+ import { readFileSync } from 'node:fs';
10
+ import { bodyByPage } from './page-marker.js';
11
+ function norm(s) {
12
+ return s.normalize('NFKD').replace(/\p{M}/gu, '').toLowerCase().replace(/[^a-z0-9]/g, '');
13
+ }
14
+ function counter(items) {
15
+ var _m_get;
16
+ const m = new Map();
17
+ for (const item of items)m.set(item, ((_m_get = m.get(item)) !== null && _m_get !== void 0 ? _m_get : 0) + 1);
18
+ return m;
19
+ }
20
+ /**
21
+ * Audit a --debug-words dump against a markdown file: every recognized word
22
+ * must land somewhere in its page's markdown. `result.missing === 0` is the pass/fail signal.
23
+ *
24
+ * @param wordsPath path to a --debug-words JSON-lines dump (`{page, text, ...}` per line)
25
+ * @param mdPath path to the markdown file to check the dump against
26
+ */ export function auditWords(wordsPath, mdPath) {
27
+ const rec = new Map();
28
+ let parsed = 0;
29
+ let skipped = 0;
30
+ const wordsRaw = readFileSync(wordsPath, 'utf8');
31
+ // A trailing "\n" at EOF must not yield a spurious final empty line; a
32
+ // genuine blank line mid-file does count, as skipped.
33
+ const rawLines = wordsRaw.split('\n');
34
+ if (rawLines.length > 0 && rawLines[rawLines.length - 1] === '') rawLines.pop();
35
+ for (let line of rawLines){
36
+ line = line.trim();
37
+ if (!line.startsWith('{')) {
38
+ skipped++;
39
+ continue;
40
+ }
41
+ let o;
42
+ try {
43
+ o = JSON.parse(line);
44
+ } catch {
45
+ skipped++;
46
+ continue;
47
+ }
48
+ parsed++;
49
+ const t = norm(o.text);
50
+ if (t) {
51
+ const arr = rec.get(o.page);
52
+ if (arr) arr.push(t);
53
+ else rec.set(o.page, [
54
+ t
55
+ ]);
56
+ }
57
+ }
58
+ const md = readFileSync(mdPath, 'utf8');
59
+ const out = bodyByPage(md);
60
+ const lines = [];
61
+ let total = 0;
62
+ let strict = 0;
63
+ const perPage = [];
64
+ const pages = [
65
+ ...rec.keys()
66
+ ].sort((a, b)=>a - b);
67
+ for (const page of pages){
68
+ var _ref, _rec_get;
69
+ var _out_get;
70
+ const toks = ((_ref = (_out_get = out.get(page)) === null || _out_get === void 0 ? void 0 : _out_get.match(/\S+/g)) !== null && _ref !== void 0 ? _ref : []).map(norm).filter((t)=>t.length > 0);
71
+ const ms = counter(toks);
72
+ const joined = toks.join('');
73
+ const missing = [];
74
+ for (const [tok, n] of counter((_rec_get = rec.get(page)) !== null && _rec_get !== void 0 ? _rec_get : [])){
75
+ var _ms_get;
76
+ const d = n - ((_ms_get = ms.get(tok)) !== null && _ms_get !== void 0 ? _ms_get : 0);
77
+ if (d <= 0) continue;
78
+ strict += d;
79
+ if (joined.includes(tok)) continue; // hyphen-join / split repair
80
+ for(let i = 0; i < d; i++)missing.push(tok);
81
+ }
82
+ if (missing.length > 0) {
83
+ total += missing.length;
84
+ perPage.push({
85
+ page,
86
+ missing
87
+ });
88
+ lines.push(`p${page}: MISSING ${missing.length}: ${pyRepr(missing.slice(0, 12))}`);
89
+ }
90
+ }
91
+ const words = pages.reduce((sum, p)=>{
92
+ var _ref;
93
+ var _rec_get;
94
+ return sum + ((_ref = (_rec_get = rec.get(p)) === null || _rec_get === void 0 ? void 0 : _rec_get.length) !== null && _ref !== void 0 ? _ref : 0);
95
+ }, 0);
96
+ const summaryLine = `parsed=${parsed} non-json-skipped=${skipped} pages=${rec.size} words=${words} strict_deficits=${strict} MISSING=${total}`;
97
+ lines.push(summaryLine);
98
+ return {
99
+ parsed,
100
+ nonJsonSkipped: skipped,
101
+ pages: rec.size,
102
+ words,
103
+ strictDeficits: strict,
104
+ missing: total,
105
+ perPage,
106
+ summaryLine,
107
+ lines
108
+ };
109
+ }
110
+ // Renders a string list as ['a', 'b'] (single-quoted) to match frozen baseline comparisons byte-for-byte.
111
+ function pyRepr(items) {
112
+ return `[${items.map((s)=>`'${s.replace(/\\/g, '\\\\').replace(/'/g, "\\'")}'`).join(', ')}]`;
113
+ }
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Projects/pdf-to-md/src/audit.ts"],"sourcesContent":["// audit.ts — the severity-1 audit instrument (README's \"Design contract\"),\n// shipped as part of the product, not just a path into the source tree.\n//\n// Every recognized word (a --debug-words dump, or an Analysis page's\n// `words`) must appear in its page's markdown — normalized, compared as a per-page multiset, with substring repair for hyphen joins.\n//\n// NOTE: no producer guarantees JSON key order in a --debug-words dump —\n// parse, never grep.\nimport { readFileSync } from 'node:fs';\nimport { bodyByPage } from './page-marker.ts';\n\nfunction norm(s: string): string {\n return s\n .normalize('NFKD')\n .replace(/\\p{M}/gu, '')\n .toLowerCase()\n .replace(/[^a-z0-9]/g, '');\n}\n\nfunction counter(items: Iterable<string>): Map<string, number> {\n const m = new Map<string, number>();\n for (const item of items) m.set(item, (m.get(item) ?? 0) + 1);\n return m;\n}\n\n/** auditWords()'s result — the severity-1 no-silent-loss check. */\nexport interface AuditResult {\n /** JSON lines successfully parsed from the word dump */\n parsed: number;\n /** lines in the dump that were not parseable JSON, skipped rather than\n * failing the whole audit (same policy `--words-json` itself uses) */\n nonJsonSkipped: number;\n /** distinct page numbers present in the dump */\n pages: number;\n /** total recognized words across all pages in the dump */\n words: number;\n /** words present in the dump but not found verbatim on their page, BEFORE\n * hyphen-join/split repair is applied; always `>= missing` */\n strictDeficits: number;\n /** words genuinely absent from the markdown after repair-checking; `0` is\n * the only passing value (`pdf-to-md audit` exits 1 otherwise) */\n missing: number;\n /** per-page detail for every page with at least one missing word (capped\n * sample of the missing tokens per page) */\n perPage: { page: number; missing: string[] }[];\n /** the one-line `parsed=… pages=… words=… strict_deficits=… MISSING=…`\n * summary, also the last entry of `lines` */\n summaryLine: string;\n /** every line `pdf-to-md audit` prints, in order: one per page with\n * missing words, then `summaryLine` */\n lines: string[];\n}\n\n/**\n * Audit a --debug-words dump against a markdown file: every recognized word\n * must land somewhere in its page's markdown. `result.missing === 0` is the pass/fail signal.\n *\n * @param wordsPath path to a --debug-words JSON-lines dump (`{page, text, ...}` per line)\n * @param mdPath path to the markdown file to check the dump against\n */\nexport function auditWords(wordsPath: string, mdPath: string): AuditResult {\n const rec = new Map<number, string[]>();\n let parsed = 0;\n let skipped = 0;\n\n const wordsRaw = readFileSync(wordsPath, 'utf8');\n // A trailing \"\\n\" at EOF must not yield a spurious final empty line; a\n // genuine blank line mid-file does count, as skipped.\n const rawLines = wordsRaw.split('\\n');\n if (rawLines.length > 0 && rawLines[rawLines.length - 1] === '') rawLines.pop();\n for (let line of rawLines) {\n line = line.trim();\n if (!line.startsWith('{')) {\n skipped++;\n continue;\n }\n let o: { page: number; text: string };\n try {\n o = JSON.parse(line);\n } catch {\n skipped++;\n continue;\n }\n parsed++;\n const t = norm(o.text);\n if (t) {\n const arr = rec.get(o.page);\n if (arr) arr.push(t);\n else rec.set(o.page, [t]);\n }\n }\n\n const md = readFileSync(mdPath, 'utf8');\n const out = bodyByPage(md);\n\n const lines: string[] = [];\n let total = 0;\n let strict = 0;\n const perPage: { page: number; missing: string[] }[] = [];\n const pages = [...rec.keys()].sort((a, b) => a - b);\n\n for (const page of pages) {\n const toks = (out.get(page)?.match(/\\S+/g) ?? []).map(norm).filter((t) => t.length > 0);\n const ms = counter(toks);\n const joined = toks.join('');\n const missing: string[] = [];\n for (const [tok, n] of counter(rec.get(page) ?? [])) {\n const d = n - (ms.get(tok) ?? 0);\n if (d <= 0) continue;\n strict += d;\n if (joined.includes(tok)) continue; // hyphen-join / split repair\n for (let i = 0; i < d; i++) missing.push(tok);\n }\n if (missing.length > 0) {\n total += missing.length;\n perPage.push({ page, missing });\n lines.push(`p${page}: MISSING ${missing.length}: ${pyRepr(missing.slice(0, 12))}`);\n }\n }\n\n const words = pages.reduce((sum, p) => sum + (rec.get(p)?.length ?? 0), 0);\n const summaryLine = `parsed=${parsed} non-json-skipped=${skipped} pages=${rec.size} words=${words} strict_deficits=${strict} MISSING=${total}`;\n lines.push(summaryLine);\n\n return {\n parsed,\n nonJsonSkipped: skipped,\n pages: rec.size,\n words,\n strictDeficits: strict,\n missing: total,\n perPage,\n summaryLine,\n lines,\n };\n}\n\n// Renders a string list as ['a', 'b'] (single-quoted) to match frozen baseline comparisons byte-for-byte.\nfunction pyRepr(items: string[]): string {\n return `[${items.map((s) => `'${s.replace(/\\\\/g, '\\\\\\\\').replace(/'/g, \"\\\\'\")}'`).join(', ')}]`;\n}\n"],"names":["readFileSync","bodyByPage","norm","s","normalize","replace","toLowerCase","counter","items","m","Map","item","set","get","auditWords","wordsPath","mdPath","rec","parsed","skipped","wordsRaw","rawLines","split","length","pop","line","trim","startsWith","o","JSON","parse","t","text","arr","page","push","md","out","lines","total","strict","perPage","pages","keys","sort","a","b","toks","match","map","filter","ms","joined","join","missing","tok","n","d","includes","i","pyRepr","slice","words","reduce","sum","p","summaryLine","size","nonJsonSkipped","strictDeficits"],"mappings":"AAAA,2EAA2E;AAC3E,wEAAwE;AACxE,EAAE;AACF,qEAAqE;AACrE,qIAAqI;AACrI,EAAE;AACF,wEAAwE;AACxE,qBAAqB;AACrB,SAASA,YAAY,QAAQ,UAAU;AACvC,SAASC,UAAU,QAAQ,mBAAmB;AAE9C,SAASC,KAAKC,CAAS;IACrB,OAAOA,EACJC,SAAS,CAAC,QACVC,OAAO,CAAC,WAAW,IACnBC,WAAW,GACXD,OAAO,CAAC,cAAc;AAC3B;AAEA,SAASE,QAAQC,KAAuB;QAECC;IADvC,MAAMA,IAAI,IAAIC;IACd,KAAK,MAAMC,QAAQH,MAAOC,EAAEG,GAAG,CAACD,MAAM,EAACF,SAAAA,EAAEI,GAAG,CAACF,mBAANF,oBAAAA,SAAe,KAAK;IAC3D,OAAOA;AACT;AA8BA;;;;;;CAMC,GACD,OAAO,SAASK,WAAWC,SAAiB,EAAEC,MAAc;IAC1D,MAAMC,MAAM,IAAIP;IAChB,IAAIQ,SAAS;IACb,IAAIC,UAAU;IAEd,MAAMC,WAAWpB,aAAae,WAAW;IACzC,uEAAuE;IACvE,sDAAsD;IACtD,MAAMM,WAAWD,SAASE,KAAK,CAAC;IAChC,IAAID,SAASE,MAAM,GAAG,KAAKF,QAAQ,CAACA,SAASE,MAAM,GAAG,EAAE,KAAK,IAAIF,SAASG,GAAG;IAC7E,KAAK,IAAIC,QAAQJ,SAAU;QACzBI,OAAOA,KAAKC,IAAI;QAChB,IAAI,CAACD,KAAKE,UAAU,CAAC,MAAM;YACzBR;YACA;QACF;QACA,IAAIS;QACJ,IAAI;YACFA,IAAIC,KAAKC,KAAK,CAACL;QACjB,EAAE,OAAM;YACNN;YACA;QACF;QACAD;QACA,MAAMa,IAAI7B,KAAK0B,EAAEI,IAAI;QACrB,IAAID,GAAG;YACL,MAAME,MAAMhB,IAAIJ,GAAG,CAACe,EAAEM,IAAI;YAC1B,IAAID,KAAKA,IAAIE,IAAI,CAACJ;iBACbd,IAAIL,GAAG,CAACgB,EAAEM,IAAI,EAAE;gBAACH;aAAE;QAC1B;IACF;IAEA,MAAMK,KAAKpC,aAAagB,QAAQ;IAChC,MAAMqB,MAAMpC,WAAWmC;IAEvB,MAAME,QAAkB,EAAE;IAC1B,IAAIC,QAAQ;IACZ,IAAIC,SAAS;IACb,MAAMC,UAAiD,EAAE;IACzD,MAAMC,QAAQ;WAAIzB,IAAI0B,IAAI;KAAG,CAACC,IAAI,CAAC,CAACC,GAAGC,IAAMD,IAAIC;IAEjD,KAAK,MAAMZ,QAAQQ,MAAO;kBAKOzB;YAJjBoB;QAAd,MAAMU,OAAO,UAACV,WAAAA,IAAIxB,GAAG,CAACqB,mBAARG,+BAAAA,SAAeW,KAAK,CAAC,8CAAW,EAAE,EAAEC,GAAG,CAAC/C,MAAMgD,MAAM,CAAC,CAACnB,IAAMA,EAAER,MAAM,GAAG;QACrF,MAAM4B,KAAK5C,QAAQwC;QACnB,MAAMK,SAASL,KAAKM,IAAI,CAAC;QACzB,MAAMC,UAAoB,EAAE;QAC5B,KAAK,MAAM,CAACC,KAAKC,EAAE,IAAIjD,SAAQU,WAAAA,IAAIJ,GAAG,CAACqB,mBAARjB,sBAAAA,WAAiB,EAAE,EAAG;gBACpCkC;YAAf,MAAMM,IAAID,MAAKL,UAAAA,GAAGtC,GAAG,CAAC0C,kBAAPJ,qBAAAA,UAAe;YAC9B,IAAIM,KAAK,GAAG;YACZjB,UAAUiB;YACV,IAAIL,OAAOM,QAAQ,CAACH,MAAM,UAAU,6BAA6B;YACjE,IAAK,IAAII,IAAI,GAAGA,IAAIF,GAAGE,IAAKL,QAAQnB,IAAI,CAACoB;QAC3C;QACA,IAAID,QAAQ/B,MAAM,GAAG,GAAG;YACtBgB,SAASe,QAAQ/B,MAAM;YACvBkB,QAAQN,IAAI,CAAC;gBAAED;gBAAMoB;YAAQ;YAC7BhB,MAAMH,IAAI,CAAC,CAAC,CAAC,EAAED,KAAK,UAAU,EAAEoB,QAAQ/B,MAAM,CAAC,EAAE,EAAEqC,OAAON,QAAQO,KAAK,CAAC,GAAG,MAAM;QACnF;IACF;IAEA,MAAMC,QAAQpB,MAAMqB,MAAM,CAAC,CAACC,KAAKC;;YAAahD;eAAP+C,gBAAO/C,WAAAA,IAAIJ,GAAG,CAACoD,gBAARhD,+BAAAA,SAAYM,MAAM,uCAAI;OAAI;IACxE,MAAM2C,cAAc,CAAC,OAAO,EAAEhD,OAAO,kBAAkB,EAAEC,QAAQ,OAAO,EAAEF,IAAIkD,IAAI,CAAC,OAAO,EAAEL,MAAM,iBAAiB,EAAEtB,OAAO,SAAS,EAAED,OAAO;IAC9ID,MAAMH,IAAI,CAAC+B;IAEX,OAAO;QACLhD;QACAkD,gBAAgBjD;QAChBuB,OAAOzB,IAAIkD,IAAI;QACfL;QACAO,gBAAgB7B;QAChBc,SAASf;QACTE;QACAyB;QACA5B;IACF;AACF;AAEA,0GAA0G;AAC1G,SAASsB,OAAOpD,KAAe;IAC7B,OAAO,CAAC,CAAC,EAAEA,MAAMyC,GAAG,CAAC,CAAC9C,IAAM,CAAC,CAAC,EAAEA,EAAEE,OAAO,CAAC,OAAO,QAAQA,OAAO,CAAC,MAAM,OAAO,CAAC,CAAC,EAAEgD,IAAI,CAAC,MAAM,CAAC,CAAC;AACjG"}
@@ -0,0 +1,3 @@
1
+ import type { Command } from './types.js';
2
+ declare const audit: Command;
3
+ export default audit;
@@ -0,0 +1,20 @@
1
+ // audit command — the severity-1 audit instrument (README's "Design
2
+ // contract") as a first-class subcommand. Thin wrapper over src/audit.ts's auditWords (the library returns data, only the CLI prints and exits).
3
+ const USAGE = 'usage: pdf-to-md audit <words.jsonl> <file.md>';
4
+ const audit = async (ctx)=>{
5
+ if (ctx.rest.includes('-h') || ctx.rest.includes('--help')) {
6
+ console.log(USAGE);
7
+ return;
8
+ }
9
+ const [wordsPath, mdPath] = ctx.rest;
10
+ if (!wordsPath || !mdPath) ctx.usageError(USAGE);
11
+ // Imported here, not at module top, same lazy-load rule as every other
12
+ // command — `pdf-to-md audit` should never pay for what the other commands need.
13
+ const { auditWords } = await import('../audit.js');
14
+ const result = auditWords(wordsPath, mdPath);
15
+ for (const line of result.lines)console.log(line);
16
+ // MISSING>0 is the severity-1 fail signal: a recognized word this markdown
17
+ // lost. Exit 1 so the subcommand is usable as a CI gate; strict_deficits from hyphen-join repairs are reported but not failures.
18
+ if (result.missing > 0) process.exitCode = 1;
19
+ };
20
+ export default audit;
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Projects/pdf-to-md/src/cli/audit.ts"],"sourcesContent":["// audit command — the severity-1 audit instrument (README's \"Design\n// contract\") as a first-class subcommand. Thin wrapper over src/audit.ts's auditWords (the library returns data, only the CLI prints and exits).\nimport type { Command } from './types.ts';\n\nconst USAGE = 'usage: pdf-to-md audit <words.jsonl> <file.md>';\n\nconst audit: Command = async (ctx) => {\n if (ctx.rest.includes('-h') || ctx.rest.includes('--help')) {\n console.log(USAGE);\n return;\n }\n const [wordsPath, mdPath] = ctx.rest;\n if (!wordsPath || !mdPath) ctx.usageError(USAGE);\n\n // Imported here, not at module top, same lazy-load rule as every other\n // command — `pdf-to-md audit` should never pay for what the other commands need.\n const { auditWords } = await import('../audit.ts');\n const result = auditWords(wordsPath, mdPath);\n for (const line of result.lines) console.log(line);\n\n // MISSING>0 is the severity-1 fail signal: a recognized word this markdown\n // lost. Exit 1 so the subcommand is usable as a CI gate; strict_deficits from hyphen-join repairs are reported but not failures.\n if (result.missing > 0) process.exitCode = 1;\n};\n\nexport default audit;\n"],"names":["USAGE","audit","ctx","rest","includes","console","log","wordsPath","mdPath","usageError","auditWords","result","line","lines","missing","process","exitCode"],"mappings":"AAAA,oEAAoE;AACpE,iJAAiJ;AAGjJ,MAAMA,QAAQ;AAEd,MAAMC,QAAiB,OAAOC;IAC5B,IAAIA,IAAIC,IAAI,CAACC,QAAQ,CAAC,SAASF,IAAIC,IAAI,CAACC,QAAQ,CAAC,WAAW;QAC1DC,QAAQC,GAAG,CAACN;QACZ;IACF;IACA,MAAM,CAACO,WAAWC,OAAO,GAAGN,IAAIC,IAAI;IACpC,IAAI,CAACI,aAAa,CAACC,QAAQN,IAAIO,UAAU,CAACT;IAE1C,uEAAuE;IACvE,iFAAiF;IACjF,MAAM,EAAEU,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC;IACpC,MAAMC,SAASD,WAAWH,WAAWC;IACrC,KAAK,MAAMI,QAAQD,OAAOE,KAAK,CAAER,QAAQC,GAAG,CAACM;IAE7C,2EAA2E;IAC3E,iIAAiI;IACjI,IAAID,OAAOG,OAAO,GAAG,GAAGC,QAAQC,QAAQ,GAAG;AAC7C;AAEA,eAAef,MAAM"}
@@ -0,0 +1,3 @@
1
+ import type { Command } from './types.js';
2
+ declare const extract: Command;
3
+ export default extract;
@@ -0,0 +1,313 @@
1
+ // extract command — parse this command's flags, call src/extract.ts, print
2
+ // and write the result. No business logic here — the library returns data and throws; only the CLI prints and exits.
3
+ import { readFileSync, writeFileSync } from 'node:fs';
4
+ import path from 'node:path';
5
+ import { parseArgs } from 'node:util';
6
+ const USAGE = 'usage: pdf-to-md [extract] <file.pdf> [out.md] [--json] [--page-markers] [--ocr] [--no-ocr] ' + '[--pages N[-M]] [--stdout] [--words-json=FILE] [--debug-words=FILE] ' + '[--format md|txt|raw (default md; raw = Analysis JSON incl. the report; txt = reading-order plain text)] ' + '[--dpi=N (OCR render resolution, default 288; raise for tiny print or chart/label pages, ' + '>~400 measured worse on ordinary prose pages)] ' + '[--float-max-words=N] [--float-max-width=F] [--float-margin=F] [--para-gap=F] [--heading-scale=F] ' + '[--col-max-width=F] [--col-split=F] [--line-y-tol=F] [--line-height-ratio=F]';
7
+ const FORMATS = [
8
+ 'md',
9
+ 'txt',
10
+ 'raw'
11
+ ];
12
+ // A bad tuning-flag value (non-numeric, or <= 0 unless the flag allows zero) is a usage error, not a silent no-op.
13
+ const TUNING_FLAGS = [
14
+ {
15
+ name: 'float-max-words',
16
+ key: 'floatMaxWords',
17
+ allowZero: true,
18
+ isInt: true
19
+ },
20
+ {
21
+ name: 'float-max-width',
22
+ key: 'floatMaxWidth'
23
+ },
24
+ {
25
+ name: 'float-margin',
26
+ key: 'floatMargin'
27
+ },
28
+ {
29
+ name: 'para-gap',
30
+ key: 'paraGap'
31
+ },
32
+ {
33
+ name: 'heading-scale',
34
+ key: 'headingScale'
35
+ },
36
+ {
37
+ name: 'col-max-width',
38
+ key: 'colMaxWidth'
39
+ },
40
+ {
41
+ name: 'col-split',
42
+ key: 'colSplit'
43
+ },
44
+ {
45
+ name: 'line-y-tol',
46
+ key: 'lineYTol'
47
+ },
48
+ {
49
+ name: 'line-height-ratio',
50
+ key: 'lineHeightRatio'
51
+ }
52
+ ];
53
+ const extract = async (ctx)=>{
54
+ // --debug-words takes a required FILE value; a bare --debug-words is a
55
+ // usage error, checked before parseArgs so it can't silently swallow the next token as its value.
56
+ if (ctx.rest.includes('--debug-words')) {
57
+ ctx.usageError(`${USAGE}\n--debug-words requires a value: --debug-words=FILE`);
58
+ }
59
+ const parsed = (()=>{
60
+ try {
61
+ return parseArgs({
62
+ args: ctx.rest,
63
+ options: {
64
+ json: {
65
+ type: 'boolean',
66
+ default: false
67
+ },
68
+ 'page-markers': {
69
+ type: 'boolean',
70
+ default: false
71
+ },
72
+ ocr: {
73
+ type: 'boolean',
74
+ default: false
75
+ },
76
+ 'no-ocr': {
77
+ type: 'boolean',
78
+ default: false
79
+ },
80
+ stdout: {
81
+ type: 'boolean',
82
+ default: false
83
+ },
84
+ 'debug-words': {
85
+ type: 'string'
86
+ },
87
+ help: {
88
+ type: 'boolean',
89
+ default: false,
90
+ short: 'h'
91
+ },
92
+ pages: {
93
+ type: 'string'
94
+ },
95
+ dpi: {
96
+ type: 'string'
97
+ },
98
+ format: {
99
+ type: 'string',
100
+ default: 'md'
101
+ },
102
+ 'words-json': {
103
+ type: 'string'
104
+ },
105
+ ...Object.fromEntries(TUNING_FLAGS.map((f)=>[
106
+ f.name,
107
+ {
108
+ type: 'string'
109
+ }
110
+ ]))
111
+ },
112
+ allowPositionals: true
113
+ });
114
+ } catch (err) {
115
+ // usageError exits; the throw is dead code that tells the compiler so.
116
+ throw ctx.usageError(`${err.message}\n${USAGE}`);
117
+ }
118
+ })();
119
+ const values = parsed.values;
120
+ const positionals = parsed.positionals;
121
+ if (values.help) {
122
+ console.log(USAGE);
123
+ return;
124
+ }
125
+ const [src, outArg] = positionals;
126
+ const wordsPath = values['words-json'];
127
+ if (!src && !wordsPath) ctx.usageError(USAGE);
128
+ const formatRaw = values.format;
129
+ if (!FORMATS.includes(formatRaw)) {
130
+ ctx.usageError(`${USAGE}\n--format must be one of ${FORMATS.join('|')} (got ${JSON.stringify(formatRaw)})`);
131
+ }
132
+ const format = formatRaw;
133
+ // --pages: a malformed spec is a usage error; out-of-range page NUMBERS
134
+ // clamp instead (collect.ts / the engine apply the same clamping policy).
135
+ let pages;
136
+ if (values.pages !== undefined) {
137
+ const m = /^(\d+)(?:-(\d+))?$/.exec(values.pages);
138
+ const first = m ? Number(m[1]) : Number.NaN;
139
+ const last = m ? m[2] !== undefined ? Number(m[2]) : first : Number.NaN;
140
+ if (!m || first < 1 || last < first) {
141
+ ctx.usageError(`${USAGE}\n--pages requires N or N-M, N >= 1, M >= N (got ${JSON.stringify(values.pages)})`);
142
+ }
143
+ pages = {
144
+ first,
145
+ last
146
+ };
147
+ }
148
+ const tuning = {};
149
+ for (const spec of TUNING_FLAGS){
150
+ const raw = values[spec.name];
151
+ if (raw === undefined) continue;
152
+ const value = Number(raw);
153
+ const ok = raw !== '' && Number.isFinite(value) && (spec.allowZero ? value >= 0 : value > 0);
154
+ if (!ok) {
155
+ ctx.usageError(`${USAGE}\n--${spec.name} requires a ${spec.allowZero ? 'non-negative' : 'positive'} number (got ${JSON.stringify(raw)})`);
156
+ }
157
+ tuning[spec.key] = spec.isInt ? Math.trunc(value) : value;
158
+ }
159
+ // --dpi: page-render resolution, validated like the tuning flags. Default
160
+ // 288 is the measured operating point; >~400 measured WORSE on ordinary pages — not a "bigger is safer" knob.
161
+ let dpi;
162
+ if (values.dpi !== undefined) {
163
+ const raw = values.dpi;
164
+ const value = Number(raw);
165
+ if (raw === '' || !Number.isFinite(value) || value <= 0) {
166
+ ctx.usageError(`${USAGE}\n--dpi requires a positive number (got ${JSON.stringify(raw)})`);
167
+ }
168
+ dpi = value;
169
+ }
170
+ // In --words-json mode there may be no PDF positional; the words file
171
+ // stands in for naming, extension stripped so the default dest can never overwrite the input. Default extension follows --format.
172
+ const source = src !== null && src !== void 0 ? src : wordsPath;
173
+ const defaultExt = format === 'raw' ? 'json' : format === 'txt' ? 'txt' : 'md';
174
+ const dest = outArg || `${source.replace(/\.(pdf|jsonl)$/i, '')}.${defaultExt}`;
175
+ const onWarning = (m)=>console.error(m);
176
+ const showProgress = process.stderr.isTTY;
177
+ // onProgress carries per-page timing/count only; human-readable
178
+ // diagnostics (e.g. the sparse-retry notice) arrive via onWarning instead, so they print regardless of TTY-ness.
179
+ const onProgress = (e)=>{
180
+ if (showProgress) console.error(`p${e.page}: ${e.words} words, ${e.ms}ms`);
181
+ };
182
+ const debugWordsPath = values['debug-words'];
183
+ if (debugWordsPath !== undefined && debugWordsPath.trim() === '') {
184
+ ctx.usageError(`${USAGE}\n--debug-words requires a value: --debug-words=FILE`);
185
+ }
186
+ // The severity-1 audit instrument: every recognized word as clean JSON
187
+ // lines to its own file. stderr stays a human channel (mixed prose), never a dump target — this file is guaranteed byte-for-byte parseable.
188
+ const onWords = debugWordsPath ? (words)=>{
189
+ const lines = words.map((w)=>{
190
+ const entry = {
191
+ page: w.page,
192
+ text: w.text,
193
+ x: w.x,
194
+ y: w.y,
195
+ w: w.w,
196
+ h: w.h
197
+ };
198
+ if (w.confidence !== undefined) entry.confidence = w.confidence;
199
+ return JSON.stringify(entry);
200
+ });
201
+ writeFileSync(debugWordsPath, lines.length ? `${lines.join('\n')}\n` : '');
202
+ } : undefined;
203
+ const common = {
204
+ pages,
205
+ onWarning
206
+ };
207
+ const ocrCommon = {
208
+ ...common,
209
+ tuning,
210
+ dpi,
211
+ onProgress,
212
+ onWords
213
+ };
214
+ let words;
215
+ let skipped = 0;
216
+ if (wordsPath !== undefined) {
217
+ ({ words, skipped } = parseWordLines(readFileSync(wordsPath, 'utf8')));
218
+ if (skipped > 0) console.error(`--words-json: skipped ${skipped} non-JSON line(s)`);
219
+ }
220
+ let content;
221
+ let report;
222
+ if (format === 'md') {
223
+ // Unchanged path: extractOcr/pdfToMarkdown directly, exactly as before
224
+ // --format existed.
225
+ const { extractOcr, pdfToMarkdown } = await import('../extract.js');
226
+ let markdown;
227
+ if (words !== undefined) {
228
+ ({ markdown, report } = await extractOcr({
229
+ words
230
+ }, ocrCommon));
231
+ } else {
232
+ ({ markdown, report } = await pdfToMarkdown(src, {
233
+ ...ocrCommon,
234
+ ocr: values.ocr,
235
+ noOcr: values['no-ocr'],
236
+ pageMarkers: values['page-markers']
237
+ }));
238
+ }
239
+ content = markdown;
240
+ } else {
241
+ // raw/txt: both render from an Analysis — raw IS the Analysis as JSON;
242
+ // txt is toText(analysis). Mirrors pdfToMarkdown's own auto-OCR fallback decision so a different --format doesn't silently pick a different path.
243
+ const { analyze } = await import('../analyze.js');
244
+ const { toText } = await import('../render-analysis.js');
245
+ const { needsOcrFallback, ocrFallbackNotice } = await import('../extract.js');
246
+ let analysis;
247
+ if (words !== undefined) {
248
+ analysis = await analyze({
249
+ words
250
+ }, {
251
+ ...ocrCommon,
252
+ path: 'ocr'
253
+ });
254
+ } else if (values.ocr) {
255
+ analysis = await analyze(src, {
256
+ ...ocrCommon,
257
+ path: 'ocr'
258
+ });
259
+ } else {
260
+ const text = await analyze(src, {
261
+ ...common,
262
+ path: 'text',
263
+ pageMarkers: values['page-markers']
264
+ });
265
+ if (needsOcrFallback(text.report.charsPerPage) && !values['no-ocr']) {
266
+ onWarning(ocrFallbackNotice(text.report.charsPerPage));
267
+ analysis = await analyze(src, {
268
+ ...ocrCommon,
269
+ path: 'ocr',
270
+ fallback: true
271
+ });
272
+ } else {
273
+ analysis = text;
274
+ }
275
+ }
276
+ report = analysis.report;
277
+ content = format === 'raw' ? JSON.stringify(analysis, null, 1) : toText(analysis);
278
+ }
279
+ if (values.stdout) process.stdout.write(content);
280
+ else writeFileSync(dest, content);
281
+ // The library's Report stays clean of source/out; --json consumers keep
282
+ // report.json files around and need to know which input produced each one, so the CLI (only) decorates the printed JSON with source/out here.
283
+ if (values.json) console.log(JSON.stringify({
284
+ source,
285
+ out: dest,
286
+ ...report
287
+ }, null, 1));
288
+ else if (report.path === 'ocr') {
289
+ var _report_floats;
290
+ console.error(`${path.basename(dest).padEnd(32)} OCR path heads=${report.headings}(junk ${report.junkHeadings}) ` + `paras=${report.paragraphs} callouts=${report.callouts} floats=${(_report_floats = report.floats) !== null && _report_floats !== void 0 ? _report_floats : 0} dangling=${report.danglingLong} (long) ` + `~${report.charsPerPage} chars/page`);
291
+ } else {
292
+ console.error(`${path.basename(dest).padEnd(32)} body=${report.bodyFontHeight}pt heads=${report.headings}(junk ${report.junkHeadings}) ` + `lists=${report.listItems} paras=${report.paragraphs} callouts=${report.callouts} pairs=${report.joinedPairs} ` + `~${report.charsPerPage} chars/page` + (report.sparse ? ' ⚠️ implausibly sparse — pages are likely images; re-run with --ocr' : ''));
293
+ }
294
+ };
295
+ // Parse one word-dump JSON line per word — key order is not guaranteed, parse
296
+ // never grep. Non-JSON lines are skipped with a count, not fatal (the auditor applies the same policy).
297
+ function parseWordLines(raw) {
298
+ const words = [];
299
+ let skipped = 0;
300
+ for (const line of raw.split('\n')){
301
+ if (!line.trim()) continue;
302
+ try {
303
+ words.push(JSON.parse(line));
304
+ } catch {
305
+ skipped++;
306
+ }
307
+ }
308
+ return {
309
+ words,
310
+ skipped
311
+ };
312
+ }
313
+ export default extract;
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Projects/pdf-to-md/src/cli/extract.ts"],"sourcesContent":["// extract command — parse this command's flags, call src/extract.ts, print\n// and write the result. No business logic here — the library returns data and throws; only the CLI prints and exits.\nimport { readFileSync, writeFileSync } from 'node:fs';\nimport path from 'node:path';\nimport { parseArgs } from 'node:util';\nimport type { OcrWordInput } from '../extract.ts';\nimport type { Tuning } from '../geometry.ts';\nimport type { Analysis, Report } from '../types.ts';\nimport type { Command } from './types.ts';\n\nconst USAGE =\n 'usage: pdf-to-md [extract] <file.pdf> [out.md] [--json] [--page-markers] [--ocr] [--no-ocr] ' +\n '[--pages N[-M]] [--stdout] [--words-json=FILE] [--debug-words=FILE] ' +\n '[--format md|txt|raw (default md; raw = Analysis JSON incl. the report; txt = reading-order plain text)] ' +\n '[--dpi=N (OCR render resolution, default 288; raise for tiny print or chart/label pages, ' +\n '>~400 measured worse on ordinary prose pages)] ' +\n '[--float-max-words=N] [--float-max-width=F] [--float-margin=F] [--para-gap=F] [--heading-scale=F] ' +\n '[--col-max-width=F] [--col-split=F] [--line-y-tol=F] [--line-height-ratio=F]';\n\nconst FORMATS = ['md', 'txt', 'raw'] as const;\ntype Format = (typeof FORMATS)[number];\n\n// A bad tuning-flag value (non-numeric, or <= 0 unless the flag allows zero) is a usage error, not a silent no-op.\nconst TUNING_FLAGS: {\n name: string;\n key: keyof Tuning;\n allowZero?: boolean;\n isInt?: boolean;\n}[] = [\n {\n name: 'float-max-words',\n key: 'floatMaxWords',\n allowZero: true,\n isInt: true,\n },\n { name: 'float-max-width', key: 'floatMaxWidth' },\n { name: 'float-margin', key: 'floatMargin' },\n { name: 'para-gap', key: 'paraGap' },\n { name: 'heading-scale', key: 'headingScale' },\n { name: 'col-max-width', key: 'colMaxWidth' },\n { name: 'col-split', key: 'colSplit' },\n { name: 'line-y-tol', key: 'lineYTol' },\n { name: 'line-height-ratio', key: 'lineHeightRatio' },\n];\n\nconst extract: Command = async (ctx) => {\n // --debug-words takes a required FILE value; a bare --debug-words is a\n // usage error, checked before parseArgs so it can't silently swallow the next token as its value.\n if (ctx.rest.includes('--debug-words')) {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n\n const parsed = (() => {\n try {\n return parseArgs({\n args: ctx.rest,\n options: {\n json: { type: 'boolean', default: false },\n 'page-markers': { type: 'boolean', default: false },\n ocr: { type: 'boolean', default: false },\n 'no-ocr': { type: 'boolean', default: false },\n stdout: { type: 'boolean', default: false },\n 'debug-words': { type: 'string' },\n help: { type: 'boolean', default: false, short: 'h' },\n pages: { type: 'string' },\n dpi: { type: 'string' },\n format: { type: 'string', default: 'md' },\n 'words-json': { type: 'string' },\n ...Object.fromEntries(TUNING_FLAGS.map((f) => [f.name, { type: 'string' as const }])),\n },\n allowPositionals: true,\n });\n } catch (err) {\n // usageError exits; the throw is dead code that tells the compiler so.\n throw ctx.usageError(`${(err as Error).message}\\n${USAGE}`);\n }\n })();\n const values = parsed.values as Record<string, string | boolean | undefined>;\n const positionals = parsed.positionals;\n\n if (values.help) {\n console.log(USAGE);\n return;\n }\n\n const [src, outArg] = positionals;\n const wordsPath = values['words-json'] as string | undefined;\n if (!src && !wordsPath) ctx.usageError(USAGE);\n\n const formatRaw = values.format as string;\n if (!(FORMATS as readonly string[]).includes(formatRaw)) {\n ctx.usageError(`${USAGE}\\n--format must be one of ${FORMATS.join('|')} (got ${JSON.stringify(formatRaw)})`);\n }\n const format = formatRaw as Format;\n\n // --pages: a malformed spec is a usage error; out-of-range page NUMBERS\n // clamp instead (collect.ts / the engine apply the same clamping policy).\n let pages: { first: number; last: number } | undefined;\n if (values.pages !== undefined) {\n const m = /^(\\d+)(?:-(\\d+))?$/.exec(values.pages as string);\n const first = m ? Number(m[1]) : Number.NaN;\n const last = m ? (m[2] !== undefined ? Number(m[2]) : first) : Number.NaN;\n if (!m || first < 1 || last < first) {\n ctx.usageError(`${USAGE}\\n--pages requires N or N-M, N >= 1, M >= N (got ${JSON.stringify(values.pages)})`);\n }\n pages = { first, last };\n }\n\n const tuning: Partial<Tuning> = {};\n for (const spec of TUNING_FLAGS) {\n const raw = values[spec.name] as string | undefined;\n if (raw === undefined) continue;\n const value = Number(raw);\n const ok = raw !== '' && Number.isFinite(value) && (spec.allowZero ? value >= 0 : value > 0);\n if (!ok) {\n ctx.usageError(`${USAGE}\\n--${spec.name} requires a ${spec.allowZero ? 'non-negative' : 'positive'} number (got ${JSON.stringify(raw)})`);\n }\n (tuning[spec.key] as number) = spec.isInt ? Math.trunc(value) : value;\n }\n\n // --dpi: page-render resolution, validated like the tuning flags. Default\n // 288 is the measured operating point; >~400 measured WORSE on ordinary pages — not a \"bigger is safer\" knob.\n let dpi: number | undefined;\n if (values.dpi !== undefined) {\n const raw = values.dpi as string;\n const value = Number(raw);\n if (raw === '' || !Number.isFinite(value) || value <= 0) {\n ctx.usageError(`${USAGE}\\n--dpi requires a positive number (got ${JSON.stringify(raw)})`);\n }\n dpi = value;\n }\n\n // In --words-json mode there may be no PDF positional; the words file\n // stands in for naming, extension stripped so the default dest can never overwrite the input. Default extension follows --format.\n const source = src ?? (wordsPath as string);\n const defaultExt = format === 'raw' ? 'json' : format === 'txt' ? 'txt' : 'md';\n const dest = outArg || `${source.replace(/\\.(pdf|jsonl)$/i, '')}.${defaultExt}`;\n\n const onWarning = (m: string) => console.error(m);\n const showProgress = process.stderr.isTTY;\n // onProgress carries per-page timing/count only; human-readable\n // diagnostics (e.g. the sparse-retry notice) arrive via onWarning instead, so they print regardless of TTY-ness.\n const onProgress = (e: { type: 'page'; page: number; words: number; ms: number }) => {\n if (showProgress) console.error(`p${e.page}: ${e.words} words, ${e.ms}ms`);\n };\n const debugWordsPath = values['debug-words'] as string | undefined;\n if (debugWordsPath !== undefined && debugWordsPath.trim() === '') {\n ctx.usageError(`${USAGE}\\n--debug-words requires a value: --debug-words=FILE`);\n }\n // The severity-1 audit instrument: every recognized word as clean JSON\n // lines to its own file. stderr stays a human channel (mixed prose), never a dump target — this file is guaranteed byte-for-byte parseable.\n const onWords = debugWordsPath\n ? (words: OcrWordInput[]) => {\n const lines = words.map((w) => {\n const entry: OcrWordInput = { page: w.page, text: w.text, x: w.x, y: w.y, w: w.w, h: w.h };\n if (w.confidence !== undefined) entry.confidence = w.confidence;\n return JSON.stringify(entry);\n });\n writeFileSync(debugWordsPath, lines.length ? `${lines.join('\\n')}\\n` : '');\n }\n : undefined;\n\n const common = { pages, onWarning };\n const ocrCommon = {\n ...common,\n tuning,\n dpi,\n onProgress,\n onWords,\n };\n\n let words: OcrWordInput[] | undefined;\n let skipped = 0;\n if (wordsPath !== undefined) {\n ({ words, skipped } = parseWordLines(readFileSync(wordsPath, 'utf8')));\n if (skipped > 0) console.error(`--words-json: skipped ${skipped} non-JSON line(s)`);\n }\n\n let content: string;\n let report: Report;\n if (format === 'md') {\n // Unchanged path: extractOcr/pdfToMarkdown directly, exactly as before\n // --format existed.\n const { extractOcr, pdfToMarkdown } = await import('../extract.ts');\n let markdown: string;\n if (words !== undefined) {\n ({ markdown, report } = await extractOcr({ words }, ocrCommon));\n } else {\n ({ markdown, report } = await pdfToMarkdown(src, {\n ...ocrCommon,\n ocr: values.ocr as boolean,\n noOcr: values['no-ocr'] as boolean,\n pageMarkers: values['page-markers'] as boolean,\n }));\n }\n content = markdown;\n } else {\n // raw/txt: both render from an Analysis — raw IS the Analysis as JSON;\n // txt is toText(analysis). Mirrors pdfToMarkdown's own auto-OCR fallback decision so a different --format doesn't silently pick a different path.\n const { analyze } = await import('../analyze.ts');\n const { toText } = await import('../render-analysis.ts');\n const { needsOcrFallback, ocrFallbackNotice } = await import('../extract.ts');\n let analysis: Analysis;\n if (words !== undefined) {\n analysis = await analyze({ words }, { ...ocrCommon, path: 'ocr' });\n } else if (values.ocr) {\n analysis = await analyze(src, { ...ocrCommon, path: 'ocr' });\n } else {\n const text = await analyze(src, { ...common, path: 'text', pageMarkers: values['page-markers'] as boolean });\n if (needsOcrFallback(text.report.charsPerPage) && !values['no-ocr']) {\n onWarning(ocrFallbackNotice(text.report.charsPerPage));\n analysis = await analyze(src, { ...ocrCommon, path: 'ocr', fallback: true });\n } else {\n analysis = text;\n }\n }\n report = analysis.report;\n content = format === 'raw' ? JSON.stringify(analysis, null, 1) : toText(analysis);\n }\n\n if (values.stdout) process.stdout.write(content);\n else writeFileSync(dest, content);\n\n // The library's Report stays clean of source/out; --json consumers keep\n // report.json files around and need to know which input produced each one, so the CLI (only) decorates the printed JSON with source/out here.\n if (values.json) console.log(JSON.stringify({ source, out: dest, ...report }, null, 1));\n else if (report.path === 'ocr') {\n console.error(`${path.basename(dest).padEnd(32)} OCR path heads=${report.headings}(junk ${report.junkHeadings}) ` + `paras=${report.paragraphs} callouts=${report.callouts} floats=${report.floats ?? 0} dangling=${report.danglingLong} (long) ` + `~${report.charsPerPage} chars/page`);\n } else {\n console.error(\n `${path.basename(dest).padEnd(32)} body=${report.bodyFontHeight}pt heads=${report.headings}(junk ${report.junkHeadings}) ` +\n `lists=${report.listItems} paras=${report.paragraphs} callouts=${report.callouts} pairs=${report.joinedPairs} ` +\n `~${report.charsPerPage} chars/page` +\n (report.sparse ? ' ⚠️ implausibly sparse — pages are likely images; re-run with --ocr' : '')\n );\n }\n};\n\n// Parse one word-dump JSON line per word — key order is not guaranteed, parse\n// never grep. Non-JSON lines are skipped with a count, not fatal (the auditor applies the same policy).\nfunction parseWordLines(raw: string): { words: OcrWordInput[]; skipped: number } {\n const words: OcrWordInput[] = [];\n let skipped = 0;\n for (const line of raw.split('\\n')) {\n if (!line.trim()) continue;\n try {\n words.push(JSON.parse(line));\n } catch {\n skipped++;\n }\n }\n return { words, skipped };\n}\n\nexport default extract;\n"],"names":["readFileSync","writeFileSync","path","parseArgs","USAGE","FORMATS","TUNING_FLAGS","name","key","allowZero","isInt","extract","ctx","rest","includes","usageError","parsed","args","options","json","type","default","ocr","stdout","help","short","pages","dpi","format","Object","fromEntries","map","f","allowPositionals","err","message","values","positionals","console","log","src","outArg","wordsPath","formatRaw","join","JSON","stringify","undefined","m","exec","first","Number","NaN","last","tuning","spec","raw","value","ok","isFinite","Math","trunc","source","defaultExt","dest","replace","onWarning","error","showProgress","process","stderr","isTTY","onProgress","e","page","words","ms","debugWordsPath","trim","onWords","lines","w","entry","text","x","y","h","confidence","length","common","ocrCommon","skipped","parseWordLines","content","report","extractOcr","pdfToMarkdown","markdown","noOcr","pageMarkers","analyze","toText","needsOcrFallback","ocrFallbackNotice","analysis","charsPerPage","fallback","write","out","basename","padEnd","headings","junkHeadings","paragraphs","callouts","floats","danglingLong","bodyFontHeight","listItems","joinedPairs","sparse","line","split","push","parse"],"mappings":"AAAA,2EAA2E;AAC3E,qHAAqH;AACrH,SAASA,YAAY,EAAEC,aAAa,QAAQ,UAAU;AACtD,OAAOC,UAAU,YAAY;AAC7B,SAASC,SAAS,QAAQ,YAAY;AAMtC,MAAMC,QACJ,iGACA,yEACA,8GACA,8FACA,oDACA,uGACA;AAEF,MAAMC,UAAU;IAAC;IAAM;IAAO;CAAM;AAGpC,mHAAmH;AACnH,MAAMC,eAKA;IACJ;QACEC,MAAM;QACNC,KAAK;QACLC,WAAW;QACXC,OAAO;IACT;IACA;QAAEH,MAAM;QAAmBC,KAAK;IAAgB;IAChD;QAAED,MAAM;QAAgBC,KAAK;IAAc;IAC3C;QAAED,MAAM;QAAYC,KAAK;IAAU;IACnC;QAAED,MAAM;QAAiBC,KAAK;IAAe;IAC7C;QAAED,MAAM;QAAiBC,KAAK;IAAc;IAC5C;QAAED,MAAM;QAAaC,KAAK;IAAW;IACrC;QAAED,MAAM;QAAcC,KAAK;IAAW;IACtC;QAAED,MAAM;QAAqBC,KAAK;IAAkB;CACrD;AAED,MAAMG,UAAmB,OAAOC;IAC9B,uEAAuE;IACvE,kGAAkG;IAClG,IAAIA,IAAIC,IAAI,CAACC,QAAQ,CAAC,kBAAkB;QACtCF,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IAEA,MAAMY,SAAS,AAAC,CAAA;QACd,IAAI;YACF,OAAOb,UAAU;gBACfc,MAAML,IAAIC,IAAI;gBACdK,SAAS;oBACPC,MAAM;wBAAEC,MAAM;wBAAWC,SAAS;oBAAM;oBACxC,gBAAgB;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAClDC,KAAK;wBAAEF,MAAM;wBAAWC,SAAS;oBAAM;oBACvC,UAAU;wBAAED,MAAM;wBAAWC,SAAS;oBAAM;oBAC5CE,QAAQ;wBAAEH,MAAM;wBAAWC,SAAS;oBAAM;oBAC1C,eAAe;wBAAED,MAAM;oBAAS;oBAChCI,MAAM;wBAAEJ,MAAM;wBAAWC,SAAS;wBAAOI,OAAO;oBAAI;oBACpDC,OAAO;wBAAEN,MAAM;oBAAS;oBACxBO,KAAK;wBAAEP,MAAM;oBAAS;oBACtBQ,QAAQ;wBAAER,MAAM;wBAAUC,SAAS;oBAAK;oBACxC,cAAc;wBAAED,MAAM;oBAAS;oBAC/B,GAAGS,OAAOC,WAAW,CAACxB,aAAayB,GAAG,CAAC,CAACC,IAAM;4BAACA,EAAEzB,IAAI;4BAAE;gCAAEa,MAAM;4BAAkB;yBAAE,EAAE;gBACvF;gBACAa,kBAAkB;YACpB;QACF,EAAE,OAAOC,KAAK;YACZ,uEAAuE;YACvE,MAAMtB,IAAIG,UAAU,CAAC,GAAG,AAACmB,IAAcC,OAAO,CAAC,EAAE,EAAE/B,OAAO;QAC5D;IACF,CAAA;IACA,MAAMgC,SAASpB,OAAOoB,MAAM;IAC5B,MAAMC,cAAcrB,OAAOqB,WAAW;IAEtC,IAAID,OAAOZ,IAAI,EAAE;QACfc,QAAQC,GAAG,CAACnC;QACZ;IACF;IAEA,MAAM,CAACoC,KAAKC,OAAO,GAAGJ;IACtB,MAAMK,YAAYN,MAAM,CAAC,aAAa;IACtC,IAAI,CAACI,OAAO,CAACE,WAAW9B,IAAIG,UAAU,CAACX;IAEvC,MAAMuC,YAAYP,OAAOR,MAAM;IAC/B,IAAI,CAAC,AAACvB,QAA8BS,QAAQ,CAAC6B,YAAY;QACvD/B,IAAIG,UAAU,CAAC,GAAGX,MAAM,0BAA0B,EAAEC,QAAQuC,IAAI,CAAC,KAAK,MAAM,EAAEC,KAAKC,SAAS,CAACH,WAAW,CAAC,CAAC;IAC5G;IACA,MAAMf,SAASe;IAEf,wEAAwE;IACxE,0EAA0E;IAC1E,IAAIjB;IACJ,IAAIU,OAAOV,KAAK,KAAKqB,WAAW;QAC9B,MAAMC,IAAI,qBAAqBC,IAAI,CAACb,OAAOV,KAAK;QAChD,MAAMwB,QAAQF,IAAIG,OAAOH,CAAC,CAAC,EAAE,IAAIG,OAAOC,GAAG;QAC3C,MAAMC,OAAOL,IAAKA,CAAC,CAAC,EAAE,KAAKD,YAAYI,OAAOH,CAAC,CAAC,EAAE,IAAIE,QAASC,OAAOC,GAAG;QACzE,IAAI,CAACJ,KAAKE,QAAQ,KAAKG,OAAOH,OAAO;YACnCtC,IAAIG,UAAU,CAAC,GAAGX,MAAM,iDAAiD,EAAEyC,KAAKC,SAAS,CAACV,OAAOV,KAAK,EAAE,CAAC,CAAC;QAC5G;QACAA,QAAQ;YAAEwB;YAAOG;QAAK;IACxB;IAEA,MAAMC,SAA0B,CAAC;IACjC,KAAK,MAAMC,QAAQjD,aAAc;QAC/B,MAAMkD,MAAMpB,MAAM,CAACmB,KAAKhD,IAAI,CAAC;QAC7B,IAAIiD,QAAQT,WAAW;QACvB,MAAMU,QAAQN,OAAOK;QACrB,MAAME,KAAKF,QAAQ,MAAML,OAAOQ,QAAQ,CAACF,UAAWF,CAAAA,KAAK9C,SAAS,GAAGgD,SAAS,IAAIA,QAAQ,CAAA;QAC1F,IAAI,CAACC,IAAI;YACP9C,IAAIG,UAAU,CAAC,GAAGX,MAAM,IAAI,EAAEmD,KAAKhD,IAAI,CAAC,YAAY,EAAEgD,KAAK9C,SAAS,GAAG,iBAAiB,WAAW,aAAa,EAAEoC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1I;QACCF,MAAM,CAACC,KAAK/C,GAAG,CAAC,GAAc+C,KAAK7C,KAAK,GAAGkD,KAAKC,KAAK,CAACJ,SAASA;IAClE;IAEA,0EAA0E;IAC1E,8GAA8G;IAC9G,IAAI9B;IACJ,IAAIS,OAAOT,GAAG,KAAKoB,WAAW;QAC5B,MAAMS,MAAMpB,OAAOT,GAAG;QACtB,MAAM8B,QAAQN,OAAOK;QACrB,IAAIA,QAAQ,MAAM,CAACL,OAAOQ,QAAQ,CAACF,UAAUA,SAAS,GAAG;YACvD7C,IAAIG,UAAU,CAAC,GAAGX,MAAM,wCAAwC,EAAEyC,KAAKC,SAAS,CAACU,KAAK,CAAC,CAAC;QAC1F;QACA7B,MAAM8B;IACR;IAEA,sEAAsE;IACtE,kIAAkI;IAClI,MAAMK,SAAStB,gBAAAA,iBAAAA,MAAQE;IACvB,MAAMqB,aAAanC,WAAW,QAAQ,SAASA,WAAW,QAAQ,QAAQ;IAC1E,MAAMoC,OAAOvB,UAAU,GAAGqB,OAAOG,OAAO,CAAC,mBAAmB,IAAI,CAAC,EAAEF,YAAY;IAE/E,MAAMG,YAAY,CAAClB,IAAcV,QAAQ6B,KAAK,CAACnB;IAC/C,MAAMoB,eAAeC,QAAQC,MAAM,CAACC,KAAK;IACzC,gEAAgE;IAChE,iHAAiH;IACjH,MAAMC,aAAa,CAACC;QAClB,IAAIL,cAAc9B,QAAQ6B,KAAK,CAAC,CAAC,CAAC,EAAEM,EAAEC,IAAI,CAAC,EAAE,EAAED,EAAEE,KAAK,CAAC,QAAQ,EAAEF,EAAEG,EAAE,CAAC,EAAE,CAAC;IAC3E;IACA,MAAMC,iBAAiBzC,MAAM,CAAC,cAAc;IAC5C,IAAIyC,mBAAmB9B,aAAa8B,eAAeC,IAAI,OAAO,IAAI;QAChElE,IAAIG,UAAU,CAAC,GAAGX,MAAM,oDAAoD,CAAC;IAC/E;IACA,uEAAuE;IACvE,4IAA4I;IAC5I,MAAM2E,UAAUF,iBACZ,CAACF;QACC,MAAMK,QAAQL,MAAM5C,GAAG,CAAC,CAACkD;YACvB,MAAMC,QAAsB;gBAAER,MAAMO,EAAEP,IAAI;gBAAES,MAAMF,EAAEE,IAAI;gBAAEC,GAAGH,EAAEG,CAAC;gBAAEC,GAAGJ,EAAEI,CAAC;gBAAEJ,GAAGA,EAAEA,CAAC;gBAAEK,GAAGL,EAAEK,CAAC;YAAC;YACzF,IAAIL,EAAEM,UAAU,KAAKxC,WAAWmC,MAAMK,UAAU,GAAGN,EAAEM,UAAU;YAC/D,OAAO1C,KAAKC,SAAS,CAACoC;QACxB;QACAjF,cAAc4E,gBAAgBG,MAAMQ,MAAM,GAAG,GAAGR,MAAMpC,IAAI,CAAC,MAAM,EAAE,CAAC,GAAG;IACzE,IACAG;IAEJ,MAAM0C,SAAS;QAAE/D;QAAOwC;IAAU;IAClC,MAAMwB,YAAY;QAChB,GAAGD,MAAM;QACTnC;QACA3B;QACA6C;QACAO;IACF;IAEA,IAAIJ;IACJ,IAAIgB,UAAU;IACd,IAAIjD,cAAcK,WAAW;QAC1B,CAAA,EAAE4B,KAAK,EAAEgB,OAAO,EAAE,GAAGC,eAAe5F,aAAa0C,WAAW,QAAO;QACpE,IAAIiD,UAAU,GAAGrD,QAAQ6B,KAAK,CAAC,CAAC,sBAAsB,EAAEwB,QAAQ,iBAAiB,CAAC;IACpF;IAEA,IAAIE;IACJ,IAAIC;IACJ,IAAIlE,WAAW,MAAM;QACnB,uEAAuE;QACvE,oBAAoB;QACpB,MAAM,EAAEmE,UAAU,EAAEC,aAAa,EAAE,GAAG,MAAM,MAAM,CAAC;QACnD,IAAIC;QACJ,IAAItB,UAAU5B,WAAW;YACtB,CAAA,EAAEkD,QAAQ,EAAEH,MAAM,EAAE,GAAG,MAAMC,WAAW;gBAAEpB;YAAM,GAAGe,UAAS;QAC/D,OAAO;YACJ,CAAA,EAAEO,QAAQ,EAAEH,MAAM,EAAE,GAAG,MAAME,cAAcxD,KAAK;gBAC/C,GAAGkD,SAAS;gBACZpE,KAAKc,OAAOd,GAAG;gBACf4E,OAAO9D,MAAM,CAAC,SAAS;gBACvB+D,aAAa/D,MAAM,CAAC,eAAe;YACrC,EAAC;QACH;QACAyD,UAAUI;IACZ,OAAO;QACL,uEAAuE;QACvE,kJAAkJ;QAClJ,MAAM,EAAEG,OAAO,EAAE,GAAG,MAAM,MAAM,CAAC;QACjC,MAAM,EAAEC,MAAM,EAAE,GAAG,MAAM,MAAM,CAAC;QAChC,MAAM,EAAEC,gBAAgB,EAAEC,iBAAiB,EAAE,GAAG,MAAM,MAAM,CAAC;QAC7D,IAAIC;QACJ,IAAI7B,UAAU5B,WAAW;YACvByD,WAAW,MAAMJ,QAAQ;gBAAEzB;YAAM,GAAG;gBAAE,GAAGe,SAAS;gBAAExF,MAAM;YAAM;QAClE,OAAO,IAAIkC,OAAOd,GAAG,EAAE;YACrBkF,WAAW,MAAMJ,QAAQ5D,KAAK;gBAAE,GAAGkD,SAAS;gBAAExF,MAAM;YAAM;QAC5D,OAAO;YACL,MAAMiF,OAAO,MAAMiB,QAAQ5D,KAAK;gBAAE,GAAGiD,MAAM;gBAAEvF,MAAM;gBAAQiG,aAAa/D,MAAM,CAAC,eAAe;YAAY;YAC1G,IAAIkE,iBAAiBnB,KAAKW,MAAM,CAACW,YAAY,KAAK,CAACrE,MAAM,CAAC,SAAS,EAAE;gBACnE8B,UAAUqC,kBAAkBpB,KAAKW,MAAM,CAACW,YAAY;gBACpDD,WAAW,MAAMJ,QAAQ5D,KAAK;oBAAE,GAAGkD,SAAS;oBAAExF,MAAM;oBAAOwG,UAAU;gBAAK;YAC5E,OAAO;gBACLF,WAAWrB;YACb;QACF;QACAW,SAASU,SAASV,MAAM;QACxBD,UAAUjE,WAAW,QAAQiB,KAAKC,SAAS,CAAC0D,UAAU,MAAM,KAAKH,OAAOG;IAC1E;IAEA,IAAIpE,OAAOb,MAAM,EAAE8C,QAAQ9C,MAAM,CAACoF,KAAK,CAACd;SACnC5F,cAAc+D,MAAM6B;IAEzB,wEAAwE;IACxE,8IAA8I;IAC9I,IAAIzD,OAAOjB,IAAI,EAAEmB,QAAQC,GAAG,CAACM,KAAKC,SAAS,CAAC;QAAEgB;QAAQ8C,KAAK5C;QAAM,GAAG8B,MAAM;IAAC,GAAG,MAAM;SAC/E,IAAIA,OAAO5F,IAAI,KAAK,OAAO;YACuJ4F;QAArLxD,QAAQ6B,KAAK,CAAC,GAAGjE,KAAK2G,QAAQ,CAAC7C,MAAM8C,MAAM,CAAC,IAAI,gBAAgB,EAAEhB,OAAOiB,QAAQ,CAAC,MAAM,EAAEjB,OAAOkB,YAAY,CAAC,EAAE,CAAC,GAAG,CAAC,MAAM,EAAElB,OAAOmB,UAAU,CAAC,UAAU,EAAEnB,OAAOoB,QAAQ,CAAC,QAAQ,GAAEpB,iBAAAA,OAAOqB,MAAM,cAAbrB,4BAAAA,iBAAiB,EAAE,UAAU,EAAEA,OAAOsB,YAAY,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,EAAEtB,OAAOW,YAAY,CAAC,WAAW,CAAC;IAC1R,OAAO;QACLnE,QAAQ6B,KAAK,CACX,GAAGjE,KAAK2G,QAAQ,CAAC7C,MAAM8C,MAAM,CAAC,IAAI,MAAM,EAAEhB,OAAOuB,cAAc,CAAC,SAAS,EAAEvB,OAAOiB,QAAQ,CAAC,MAAM,EAAEjB,OAAOkB,YAAY,CAAC,EAAE,CAAC,GACxH,CAAC,MAAM,EAAElB,OAAOwB,SAAS,CAAC,OAAO,EAAExB,OAAOmB,UAAU,CAAC,UAAU,EAAEnB,OAAOoB,QAAQ,CAAC,OAAO,EAAEpB,OAAOyB,WAAW,CAAC,CAAC,CAAC,GAC/G,CAAC,CAAC,EAAEzB,OAAOW,YAAY,CAAC,WAAW,CAAC,GACnCX,CAAAA,OAAO0B,MAAM,GAAG,0EAA0E,EAAC;IAElG;AACF;AAEA,8EAA8E;AAC9E,wGAAwG;AACxG,SAAS5B,eAAepC,GAAW;IACjC,MAAMmB,QAAwB,EAAE;IAChC,IAAIgB,UAAU;IACd,KAAK,MAAM8B,QAAQjE,IAAIkE,KAAK,CAAC,MAAO;QAClC,IAAI,CAACD,KAAK3C,IAAI,IAAI;QAClB,IAAI;YACFH,MAAMgD,IAAI,CAAC9E,KAAK+E,KAAK,CAACH;QACxB,EAAE,OAAM;YACN9B;QACF;IACF;IACA,OAAO;QAAEhB;QAAOgB;IAAQ;AAC1B;AAEA,eAAehF,QAAQ"}
@@ -0,0 +1,5 @@
1
+ import type { Command } from './types.js';
2
+ export declare const COMMANDS: Record<string, () => Promise<{
3
+ default: Command;
4
+ }>>;
5
+ export type { Command, Ctx } from './types.js';
@@ -0,0 +1,7 @@
1
+ // Lazy registry: a command's imports load only when it runs. These names are
2
+ // reserved — a PDF literally named `extract`/`render` (no extension) needs a ./ prefix.
3
+ export const COMMANDS = {
4
+ extract: ()=>import('./extract.js'),
5
+ render: ()=>import('./render.js'),
6
+ audit: ()=>import('./audit.js')
7
+ };
@@ -0,0 +1 @@
1
+ {"version":3,"sources":["/Users/kevin/Projects/pdf-to-md/src/cli/index.ts"],"sourcesContent":["import type { Command } from './types.ts';\n\n// Lazy registry: a command's imports load only when it runs. These names are\n// reserved — a PDF literally named `extract`/`render` (no extension) needs a ./ prefix.\nexport const COMMANDS: Record<string, () => Promise<{ default: Command }>> = {\n extract: () => import('./extract.ts'),\n render: () => import('./render.ts'),\n audit: () => import('./audit.ts'),\n};\n\nexport type { Command, Ctx } from './types.ts';\n"],"names":["COMMANDS","extract","render","audit"],"mappings":"AAEA,6EAA6E;AAC7E,wFAAwF;AACxF,OAAO,MAAMA,WAAgE;IAC3EC,SAAS,IAAM,MAAM,CAAC;IACtBC,QAAQ,IAAM,MAAM,CAAC;IACrBC,OAAO,IAAM,MAAM,CAAC;AACtB,EAAE"}