sensemaking 0.22.1 → 0.23.0

package/dist/cjs/chunk/group.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/group.ts"],"sourcesContent":["import { extractText } from './extract.ts';\nimport { parse } from './parse.ts';\nimport { DEFAULT_TARGET_TOKENS, estimateTokens } from './tokens.ts';\nimport type { Block, BlockType, Chunk, ChunkOptions } from './types.ts';\n\nconst PGC_GROUP_SIZE = 2;\nconst OVERSIZE_TRIGGER_MULTIPLE = 2;\n\ninterface ResolvedOptions {\n  targetTokens: number;\n  text: 'extracted' | 'raw';\n}\n\nfunction resolveOptions(opts?: ChunkOptions): ResolvedOptions {\n  return {\n    targetTokens: opts?.targetTokens ?? DEFAULT_TARGET_TOKENS,\n    text: opts?.text ?? 'raw',\n  };\n}\n\n// A heading of any depth ends the current scope and starts a new one (D1); the heading block\n// itself is carried into the new scope, where it joins that scope's first group (F7/F10).\nfunction splitScopes(blocks: Block[]): Block[][] {\n  const scopes: Block[][] = [];\n  let current: Block[] = [];\n  for (const block of blocks) {\n    if (block.type === 'heading' && current.length > 0) {\n      scopes.push(current);\n      current = [];\n    }\n    current.push(block);\n  }\n  if (current.length > 0) scopes.push(current);\n  return scopes;\n}\n\ninterface Part {\n  startLine: number;\n  endLine: number;\n  text: string;\n  // True for a sub-line split piece: its text is already the final slice, not the whole line --\n  // group()'s extent-based raw re-slice must not touch it (siblings share startLine === endLine).\n  final?: boolean;\n}\n\nfunction finalize(parts: Part[]): (Chunk & { final?: boolean }) | undefined {\n  if (parts.length === 0) return undefined;\n  const first = parts[0];\n  const last = parts[parts.length - 1];\n  return { startLine: first.startLine, endLine: last.endLine, text: parts.map((p) => p.text).join('\\n'), final: parts.some((p) => p.final) };\n}\n\nconst NEWLINE_TOKENS = estimateTokens('\\n');\n// Built on first use and kept: each construction is ~3.5 ms, and only an oversize block is ever\n// split, so no command pays for a segmenter it never reaches.\nconst SEGMENTERS = new Map<string, Intl.Segmenter>();\n\nfunction segmentsOf(text: string, granularity: 'sentence' | 'word'): string[] {\n  let segmenter = SEGMENTERS.get(granularity);\n  if (!segmenter) {\n    segmenter = new Intl.Segmenter(undefined, { granularity });\n    SEGMENTERS.set(granularity, segmenter);\n  }\n  return Array.from(segmenter.segment(text), (s) => s.segment);\n}\n\n// Greedily packs segments (already contiguous, tiling the source text with no gaps) into groups\n// of at most `working` estimated tokens; a lone segment over `working` still stands alone.\nfunction pack(segments: string[], working: number): string[] {\n  const groups: string[] = [];\n  let current = '';\n  let tokens = 0;\n  for (const segment of segments) {\n    const segmentTokens = estimateTokens(segment);\n    if (current.length > 0 && tokens + segmentTokens > working) {\n      groups.push(current);\n      current = '';\n      tokens = 0;\n    }\n    current += segment;\n    tokens += segmentTokens;\n  }\n  if (current.length > 0) groups.push(current);\n  return groups;\n}\n\n// Line-split alone can't shrink a lone dense line (the CJK case): falls back to sentence then\n// word boundaries (Intl.Segmenter, the same grapheme-safe engine as segment.ts), mode-agnostic on `text`.\nfunction splitLineText(text: string, working: number): string[] {\n  const sentences = segmentsOf(text, 'sentence');\n  const out: string[] = [];\n  let current = '';\n  let tokens = 0;\n  const flush = () => {\n    if (current.length > 0) {\n      out.push(current);\n      current = '';\n      tokens = 0;\n    }\n  };\n  for (const sentence of sentences) {\n    const sentenceTokens = estimateTokens(sentence);\n    if (sentenceTokens > working) {\n      flush();\n      out.push(...pack(segmentsOf(sentence, 'word'), working));\n      continue;\n    }\n    if (current.length > 0 && tokens + sentenceTokens > working) flush();\n    current += sentence;\n    tokens += sentenceTokens;\n  }\n  flush();\n  return out;\n}\n\nconst ATOMIC_TYPES: ReadonlySet<BlockType> = new Set(['code', 'table', 'list']);\n\n// Atomic (code/table/list) pieces are always a raw line slice: re-parsing a table's later pieces\n// without their header/delimiter rows would demote them to paragraph text.\nfunction piece(pieceLines: string[], startLine: number, endLine: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part {\n  const text =\n    ATOMIC_TYPES.has(blockType) || textMode === 'raw'\n      ? pieceLines.join('\\n')\n      : parse(pieceLines.join('\\n'))\n          .map((b) => extractText(b.node))\n          .join('\\n');\n  return { startLine, endLine, text };\n}\n\n// A one-line piece over working can't shrink via another line-boundary pass (rule 5's gap), so it\n// splits at sentence/word boundaries instead; `final` stops group() re-deriving its text by extent (F5).\nfunction finalizePiece(pieceLines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part[] {\n  const p = piece(pieceLines, startLine, endLine, blockType, textMode);\n  if (pieceLines.length === 1 && estimateTokens(p.text) > working) {\n    return splitLineText(p.text, working).map((text) => ({ startLine, endLine, text, final: true }));\n  }\n  return [p];\n}\n\n// A block over 2x working size splits at line boundaries into pieces each <= working size, never\n// mid-line. `seed`: pending tokens (e.g. a heading) the first piece must join, checked against the limit.\nfunction splitOversizeBlock(lines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw', seed = 0): Part[] {\n  const pieces: Part[] = [];\n  let pieceLines: string[] = [];\n  let pieceStart = startLine;\n  let tokens = seed;\n  for (let line = startLine; line <= endLine; line++) {\n    const lineText = lines[line - 1];\n    const sep = pieceLines.length > 0 || tokens > 0 ? NEWLINE_TOKENS : 0;\n    const lineTokens = estimateTokens(lineText);\n    if (pieceLines.length > 0 && tokens + sep + lineTokens > working) {\n      pieces.push(...finalizePiece(pieceLines, pieceStart, line - 1, working, blockType, textMode));\n      pieceLines = [];\n      tokens = 0;\n      pieceStart = line;\n      pieceLines.push(lineText);\n      tokens += lineTokens;\n      continue;\n    }\n    pieceLines.push(lineText);\n    tokens += sep + lineTokens;\n  }\n  if (pieceLines.length > 0) pieces.push(...finalizePiece(pieceLines, pieceStart, endLine, working, blockType, textMode));\n  return pieces;\n}\n\n// One heading scope's groups (D1): a heading opens the first group, and an oversize block\n// (rule 5, including an oversize heading) splits into pieces that each close their own group.\nfunction groupScope(scopeBlocks: Block[], lines: string[], resolved: ResolvedOptions): (Chunk & { final?: boolean })[] {\n  const working = resolved.targetTokens;\n  const trigger = working * OVERSIZE_TRIGGER_MULTIPLE;\n  const finished: (Chunk & { final?: boolean })[] = [];\n  let parts: Part[] = [];\n  let paragraphCount = 0;\n  let tokens = 0;\n\n  function close(): void {\n    const group = finalize(parts);\n    if (group) finished.push(group);\n    parts = [];\n    paragraphCount = 0;\n    tokens = 0;\n  }\n\n  // tokens tracks the active text mode's own estimate (a newline between parts costs\n  // NEWLINE_TOKENS too), so packing decisions size the text the chunk will actually ship as.\n  function addPart(text: string, sizeText: string, startLine: number, endLine: number): void {\n    tokens += (parts.length > 0 ? NEWLINE_TOKENS : 0) + estimateTokens(sizeText);\n    parts.push({ startLine, endLine, text });\n  }\n\n  for (const block of scopeBlocks) {\n    const raw = lines.slice(block.startLine - 1, block.endLine).join('\\n');\n    const blockTokens = estimateTokens(raw);\n\n    if (blockTokens > trigger) {\n      const seed = parts.length > 0 ? tokens : 0;\n      for (const p of splitOversizeBlock(lines, block.startLine, block.endLine, working, block.type, resolved.text, seed)) {\n        parts.push(p);\n        close();\n      }\n      continue;\n    }\n\n    const extracted = extractText(block.node);\n    const sizeText = resolved.text === 'raw' ? raw : extracted;\n\n    if (block.type === 'heading') {\n      addPart(extracted, sizeText, block.startLine, block.endLine);\n      continue;\n    }\n\n    // The 2x-working invariant holds even under pgc's paper-faithful 2-paragraph pairing --\n    // close first if the pair about to form would cross it.\n    const pairOversize = parts.length > 0 && tokens + NEWLINE_TOKENS + blockTokens > trigger;\n    if (pairOversize) close();\n\n    addPart(extracted, sizeText, block.startLine, block.endLine);\n    paragraphCount++;\n\n    if (paragraphCount >= PGC_GROUP_SIZE) close();\n  }\n  close();\n\n  return finished;\n}\n\n// Groups already-parsed blocks per opts (D1/D3), against the same body the blocks were parsed\n// from (line lookups for oversize splitting).\nexport function group(blocks: Block[], body: string, opts?: ChunkOptions): Chunk[] {\n  const resolved = resolveOptions(opts);\n  const lines = body.split('\\n');\n  const chunks: (Chunk & { final?: boolean })[] = [];\n  for (const scope of splitScopes(blocks)) chunks.push(...groupScope(scope, lines, resolved));\n  // 'raw': the chunk's own source lines verbatim, replacing the flavor-resolved join above (D9).\n  // A `final` chunk already carries its own slice's raw text; re-slicing by extent would return the whole shared line.\n  const texted =\n    resolved.text === 'raw'\n      ? chunks.map((c) =>\n          c.final\n            ? c\n            : {\n                ...c,\n                text: lines\n                  .slice(c.startLine - 1, c.endLine)\n                  .join('\\n')\n                  .trim(),\n              }\n        )\n      : chunks;\n  // A group can be all-blank (flavor-stripped to nothing, or a raw slice of pure syntax); it never produces a chunk.\n  return texted.filter((c) => c.text.trim().length > 0).map((c) => ({ startLine: c.startLine, endLine: c.endLine, text: c.text }));\n}\n"],"names":["group","PGC_GROUP_SIZE","OVERSIZE_TRIGGER_MULTIPLE","resolveOptions","opts","targetTokens","DEFAULT_TARGET_TOKENS","text","splitScopes","blocks","scopes","current","block","type","length","push","finalize","parts","undefined","first","last","startLine","endLine","map","p","join","final","some","NEWLINE_TOKENS","estimateTokens","SEGMENTERS","Map","segmentsOf","granularity","segmenter","get","Intl","Segmenter","set","Array","from","segment","s","pack","segments","working","groups","tokens","segmentTokens","splitLineText","sentences","out","flush","sentence","sentenceTokens","ATOMIC_TYPES","Set","piece","pieceLines","blockType","textMode","has","parse","b","extractText","node","finalizePiece","splitOversizeBlock","lines","seed","pieces","pieceStart","line","lineText","sep","lineTokens","groupScope","scopeBlocks","resolved","trigger","finished","paragraphCount","close","addPart","sizeText","raw","slice","blockTokens","extracted","pairOversize","body","chunks","split","scope","texted","c","trim","filter"],"mappings":";;;;+BAqOgBA;;;eAAAA;;;yBArOY;uBACN;wBACgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAGtD,IAAMC,iBAAiB;AACvB,IAAMC,4BAA4B;AAOlC,SAASC,eAAeC,IAAmB;;IACzC,OAAO;QACLC,YAAY,UAAED,iBAAAA,2BAAAA,KAAMC,YAAY,uCAAIC,+BAAqB;QACzDC,IAAI,WAAEH,iBAAAA,2BAAAA,KAAMG,IAAI,yCAAI;IACtB;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASC,YAAYC,MAAe;IAClC,IAAMC,SAAoB,EAAE;IAC5B,IAAIC,UAAmB,EAAE;QACpB,kCAAA,2BAAA;;QAAL,QAAK,YAAeF,2BAAf,SAAA,6BAAA,QAAA,yBAAA,iCAAuB;YAAvB,IAAMG,QAAN;YACH,IAAIA,MAAMC,IAAI,KAAK,aAAaF,QAAQG,MAAM,GAAG,GAAG;gBAClDJ,OAAOK,IAAI,CAACJ;gBACZA,UAAU,EAAE;YACd;YACAA,QAAQI,IAAI,CAACH;QACf;;QANK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAOL,IAAID,QAAQG,MAAM,GAAG,GAAGJ,OAAOK,IAAI,CAACJ;IACpC,OAAOD;AACT;AAWA,SAASM,SAASC,KAAa;IAC7B,IAAIA,MAAMH,MAAM,KAAK,GAAG,OAAOI;IAC/B,IAAMC,QAAQF,KAAK,CAAC,EAAE;IACtB,IAAMG,OAAOH,KAAK,CAACA,MAAMH,MAAM,GAAG,EAAE;IACpC,OAAO;QAAEO,WAAWF,MAAME,SAAS;QAAEC,SAASF,KAAKE,OAAO;QAAEf,MAAMU,MAAMM,GAAG,CAAC,SAACC;mBAAMA,EAAEjB,IAAI;WAAEkB,IAAI,CAAC;QAAOC,OAAOT,MAAMU,IAAI,CAAC,SAACH;mBAAMA,EAAEE,KAAK;;IAAE;AAC3I;AAEA,IAAME,iBAAiBC,IAAAA,wBAAc,EAAC;AACtC,gGAAgG;AAChG,8DAA8D;AAC9D,IAAMC,aAAa,IAAIC;AAEvB,SAASC,WAAWzB,IAAY,EAAE0B,WAAgC;IAChE,IAAIC,YAAYJ,WAAWK,GAAG,CAACF;IAC/B,IAAI,CAACC,WAAW;QACdA,YAAY,IAAIE,KAAKC,SAAS,CAACnB,WAAW;YAAEe,aAAAA;QAAY;QACxDH,WAAWQ,GAAG,CAACL,aAAaC;IAC9B;IACA,OAAOK,MAAMC,IAAI,CAACN,UAAUO,OAAO,CAAClC,OAAO,SAACmC;eAAMA,EAAED,OAAO;;AAC7D;AAEA,gGAAgG;AAChG,2FAA2F;AAC3F,SAASE,KAAKC,QAAkB,EAAEC,OAAe;IAC/C,IAAMC,SAAmB,EAAE;IAC3B,IAAInC,UAAU;IACd,IAAIoC,SAAS;QACR,kCAAA,2BAAA;;QAAL,QAAK,YAAiBH,6BAAjB,SAAA,6BAAA,QAAA,yBAAA,iCAA2B;YAA3B,IAAMH,UAAN;YACH,IAAMO,gBAAgBnB,IAAAA,wBAAc,EAACY;YACrC,IAAI9B,QAAQG,MAAM,GAAG,KAAKiC,SAASC,gBAAgBH,SAAS;gBAC1DC,OAAO/B,IAAI,CAACJ;gBACZA,UAAU;gBACVoC,SAAS;YACX;YACApC,WAAW8B;YACXM,UAAUC;QACZ;;QATK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,IAAIrC,QAAQG,MAAM,GAAG,GAAGgC,OAAO/B,IAAI,CAACJ;IACpC,OAAOmC;AACT;AAEA,8FAA8F;AAC9F,0GAA0G;AAC1G,SAASG,cAAc1C,IAAY,EAAEsC,OAAe;IAClD,IAAMK,YAAYlB,WAAWzB,MAAM;IACnC,IAAM4C,MAAgB,EAAE;IACxB,IAAIxC,UAAU;IACd,IAAIoC,SAAS;IACb,IAAMK,QAAQ;QACZ,IAAIzC,QAAQG,MAAM,GAAG,GAAG;YACtBqC,IAAIpC,IAAI,CAACJ;YACTA,UAAU;YACVoC,SAAS;QACX;IACF;QACK,kCAAA,2BAAA;;QAAL,QAAK,YAAkBG,8BAAlB,SAAA,6BAAA,QAAA,yBAAA,iCAA6B;YAA7B,IAAMG,WAAN;YACH,IAAMC,iBAAiBzB,IAAAA,wBAAc,EAACwB;YACtC,IAAIC,iBAAiBT,SAAS;oBAE5BM;gBADAC;gBACAD,CAAAA,OAAAA,KAAIpC,IAAI,OAARoC,MAAS,qBAAGR,KAAKX,WAAWqB,UAAU,SAASR;gBAC/C;YACF;YACA,IAAIlC,QAAQG,MAAM,GAAG,KAAKiC,SAASO,iBAAiBT,SAASO;YAC7DzC,WAAW0C;YACXN,UAAUO;QACZ;;QAVK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAWLF;IACA,OAAOD;AACT;AAEA,IAAMI,eAAuC,IAAIC,IAAI;IAAC;IAAQ;IAAS;CAAO;AAE9E,iGAAiG;AACjG,2EAA2E;AAC3E,SAASC,MAAMC,UAAoB,EAAErC,SAAiB,EAAEC,OAAe,EAAEqC,SAAoB,EAAEC,QAA6B;IAC1H,IAAMrD,OACJgD,aAAaM,GAAG,CAACF,cAAcC,aAAa,QACxCF,WAAWjC,IAAI,CAAC,QAChBqC,IAAAA,cAAK,EAACJ,WAAWjC,IAAI,CAAC,OACnBF,GAAG,CAAC,SAACwC;eAAMC,IAAAA,sBAAW,EAACD,EAAEE,IAAI;OAC7BxC,IAAI,CAAC;IACd,OAAO;QAAEJ,WAAAA;QAAWC,SAAAA;QAASf,MAAAA;IAAK;AACpC;AAEA,kGAAkG;AAClG,yGAAyG;AACzG,SAAS2D,cAAcR,UAAoB,EAAErC,SAAiB,EAAEC,OAAe,EAAEuB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;IACnJ,IAAMpC,IAAIiC,MAAMC,YAAYrC,WAAWC,SAASqC,WAAWC;IAC3D,IAAIF,WAAW5C,MAAM,KAAK,KAAKe,IAAAA,wBAAc,EAACL,EAAEjB,IAAI,IAAIsC,SAAS;QAC/D,OAAOI,cAAczB,EAAEjB,IAAI,EAAEsC,SAAStB,GAAG,CAAC,SAAChB;mBAAU;gBAAEc,WAAAA;gBAAWC,SAAAA;gBAASf,MAAAA;gBAAMmB,OAAO;YAAK;;IAC/F;IACA,OAAO;QAACF;KAAE;AACZ;AAEA,iGAAiG;AACjG,0GAA0G;AAC1G,SAAS2C,mBAAmBC,KAAe,EAAE/C,SAAiB,EAAEC,OAAe,EAAEuB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;QAAES,OAAAA,iEAAO;QAqBjIC;IApB3B,IAAMA,SAAiB,EAAE;IACzB,IAAIZ,aAAuB,EAAE;IAC7B,IAAIa,aAAalD;IACjB,IAAI0B,SAASsB;IACb,IAAK,IAAIG,OAAOnD,WAAWmD,QAAQlD,SAASkD,OAAQ;QAClD,IAAMC,WAAWL,KAAK,CAACI,OAAO,EAAE;QAChC,IAAME,MAAMhB,WAAW5C,MAAM,GAAG,KAAKiC,SAAS,IAAInB,iBAAiB;QACnE,IAAM+C,aAAa9C,IAAAA,wBAAc,EAAC4C;QAClC,IAAIf,WAAW5C,MAAM,GAAG,KAAKiC,SAAS2B,MAAMC,aAAa9B,SAAS;gBAChEyB;YAAAA,CAAAA,WAAAA,QAAOvD,IAAI,OAAXuD,UAAY,qBAAGJ,cAAcR,YAAYa,YAAYC,OAAO,GAAG3B,SAASc,WAAWC;YACnFF,aAAa,EAAE;YACfX,SAAS;YACTwB,aAAaC;YACbd,WAAW3C,IAAI,CAAC0D;YAChB1B,UAAU4B;YACV;QACF;QACAjB,WAAW3C,IAAI,CAAC0D;QAChB1B,UAAU2B,MAAMC;IAClB;IACA,IAAIjB,WAAW5C,MAAM,GAAG,GAAGwD,CAAAA,UAAAA,QAAOvD,IAAI,OAAXuD,SAAY,qBAAGJ,cAAcR,YAAYa,YAAYjD,SAASuB,SAASc,WAAWC;IAC7G,OAAOU;AACT;AAEA,0FAA0F;AAC1F,8FAA8F;AAC9F,SAASM,WAAWC,WAAoB,EAAET,KAAe,EAAEU,QAAyB;IAClF,IAAMjC,UAAUiC,SAASzE,YAAY;IACrC,IAAM0E,UAAUlC,UAAU3C;IAC1B,IAAM8E,WAA4C,EAAE;IACpD,IAAI/D,QAAgB,EAAE;IACtB,IAAIgE,iBAAiB;IACrB,IAAIlC,SAAS;IAEb,SAASmC;QACP,IAAMlF,QAAQgB,SAASC;QACvB,IAAIjB,OAAOgF,SAASjE,IAAI,CAACf;QACzBiB,QAAQ,EAAE;QACVgE,iBAAiB;QACjBlC,SAAS;IACX;IAEA,mFAAmF;IACnF,2FAA2F;IAC3F,SAASoC,QAAQ5E,IAAY,EAAE6E,QAAgB,EAAE/D,SAAiB,EAAEC,OAAe;QACjFyB,UAAU,AAAC9B,CAAAA,MAAMH,MAAM,GAAG,IAAIc,iBAAiB,CAAA,IAAKC,IAAAA,wBAAc,EAACuD;QACnEnE,MAAMF,IAAI,CAAC;YAAEM,WAAAA;YAAWC,SAAAA;YAASf,MAAAA;QAAK;IACxC;QAEK,kCAAA,2BAAA;;QAAL,QAAK,YAAesE,gCAAf,SAAA,6BAAA,QAAA,yBAAA,iCAA4B;YAA5B,IAAMjE,QAAN;YACH,IAAMyE,MAAMjB,MAAMkB,KAAK,CAAC1E,MAAMS,SAAS,GAAG,GAAGT,MAAMU,OAAO,EAAEG,IAAI,CAAC;YACjE,IAAM8D,cAAc1D,IAAAA,wBAAc,EAACwD;YAEnC,IAAIE,cAAcR,SAAS;gBACzB,IAAMV,OAAOpD,MAAMH,MAAM,GAAG,IAAIiC,SAAS;oBACpC,mCAAA,4BAAA;;oBAAL,QAAK,aAAWoB,mBAAmBC,OAAOxD,MAAMS,SAAS,EAAET,MAAMU,OAAO,EAAEuB,SAASjC,MAAMC,IAAI,EAAEiE,SAASvE,IAAI,EAAE8D,0BAAzG,UAAA,8BAAA,SAAA,0BAAA,kCAAgH;wBAAhH,IAAM7C,IAAN;wBACHP,MAAMF,IAAI,CAACS;wBACX0D;oBACF;;oBAHK;oBAAA;;;6BAAA,8BAAA;4BAAA;;;4BAAA;kCAAA;;;;gBAIL;YACF;YAEA,IAAMM,YAAYxB,IAAAA,sBAAW,EAACpD,MAAMqD,IAAI;YACxC,IAAMmB,WAAWN,SAASvE,IAAI,KAAK,QAAQ8E,MAAMG;YAEjD,IAAI5E,MAAMC,IAAI,KAAK,WAAW;gBAC5BsE,QAAQK,WAAWJ,UAAUxE,MAAMS,SAAS,EAAET,MAAMU,OAAO;gBAC3D;YACF;YAEA,wFAAwF;YACxF,wDAAwD;YACxD,IAAMmE,eAAexE,MAAMH,MAAM,GAAG,KAAKiC,SAASnB,iBAAiB2D,cAAcR;YACjF,IAAIU,cAAcP;YAElBC,QAAQK,WAAWJ,UAAUxE,MAAMS,SAAS,EAAET,MAAMU,OAAO;YAC3D2D;YAEA,IAAIA,kBAAkBhF,gBAAgBiF;QACxC;;QA9BK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IA+BLA;IAEA,OAAOF;AACT;AAIO,SAAShF,MAAMS,MAAe,EAAEiF,IAAY,EAAEtF,IAAmB;QAI7BuF;IAHzC,IAAMb,WAAW3E,eAAeC;IAChC,IAAMgE,QAAQsB,KAAKE,KAAK,CAAC;IACzB,IAAMD,SAA0C,EAAE;QAC7C,kCAAA,2BAAA;;QAAL,QAAK,YAAenF,YAAYC,4BAA3B,SAAA,6BAAA,QAAA,yBAAA;YAAA,IAAMoF,QAAN;YAAoCF,CAAAA,UAAAA,QAAO5E,IAAI,OAAX4E,SAAY,qBAAGf,WAAWiB,OAAOzB,OAAOU;;;QAA5E;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IACL,+FAA+F;IAC/F,qHAAqH;IACrH,IAAMgB,SACJhB,SAASvE,IAAI,KAAK,QACdoF,OAAOpE,GAAG,CAAC,SAACwE;eACVA,EAAErE,KAAK,GACHqE,IACA,wCACKA;YACHxF,MAAM6D,MACHkB,KAAK,CAACS,EAAE1E,SAAS,GAAG,GAAG0E,EAAEzE,OAAO,EAChCG,IAAI,CAAC,MACLuE,IAAI;;SAGfL;IACN,mHAAmH;IACnH,OAAOG,OAAOG,MAAM,CAAC,SAACF;eAAMA,EAAExF,IAAI,CAACyF,IAAI,GAAGlF,MAAM,GAAG;OAAGS,GAAG,CAAC,SAACwE;eAAO;YAAE1E,WAAW0E,EAAE1E,SAAS;YAAEC,SAASyE,EAAEzE,OAAO;YAAEf,MAAMwF,EAAExF,IAAI;QAAC;;AAC/H"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/group.ts"],"sourcesContent":["import { extractText } from './extract.ts';\nimport { parse } from './parse.ts';\nimport { DEFAULT_TARGET_TOKENS, estimateTokens } from './tokens.ts';\nimport type { Block, BlockType, Chunk, ChunkOptions } from './types.ts';\n\nconst PGC_GROUP_SIZE = 2;\nconst OVERSIZE_TRIGGER_MULTIPLE = 2;\n\ninterface ResolvedOptions {\n  targetTokens: number;\n  text: 'extracted' | 'raw';\n}\n\nfunction resolveOptions(opts?: ChunkOptions): ResolvedOptions {\n  return {\n    targetTokens: opts?.targetTokens ?? DEFAULT_TARGET_TOKENS,\n    text: opts?.text ?? 'raw',\n  };\n}\n\n// A heading of any depth ends the current scope and starts a new one (D1); the heading block\n// itself is carried into the new scope, where it joins that scope's first group (F7/F10).\nfunction splitScopes(blocks: Block[]): Block[][] {\n  const scopes: Block[][] = [];\n  let current: Block[] = [];\n  for (const block of blocks) {\n    if (block.type === 'heading' && current.length > 0) {\n      scopes.push(current);\n      current = [];\n    }\n    current.push(block);\n  }\n  if (current.length > 0) scopes.push(current);\n  return scopes;\n}\n\ninterface Part {\n  startLine: number;\n  endLine: number;\n  text: string;\n  // True for a sub-line split piece: its text is already the final slice, not the whole line --\n  // group()'s extent-based raw re-slice must not touch it (siblings share startLine === endLine).\n  final?: boolean;\n}\n\nfunction finalize(parts: Part[]): (Chunk & { final?: boolean }) | undefined {\n  if (parts.length === 0) return undefined;\n  const first = parts[0];\n  const last = parts[parts.length - 1];\n  return { startLine: first.startLine, endLine: last.endLine, text: parts.map((p) => p.text).join('\\n'), final: parts.some((p) => p.final) };\n}\n\nconst NEWLINE_TOKENS = estimateTokens('\\n');\n// Built on first use and kept: each construction is ~3.5 ms, and only an oversize block is ever\n// split, so no command pays for a segmenter it never reaches.\nconst SEGMENTERS = new Map<string, Intl.Segmenter>();\n\nfunction segmentsOf(text: string, granularity: 'sentence' | 'word'): string[] {\n  let segmenter = SEGMENTERS.get(granularity);\n  if (!segmenter) {\n    segmenter = new Intl.Segmenter(undefined, { granularity });\n    SEGMENTERS.set(granularity, segmenter);\n  }\n  return Array.from(segmenter.segment(text), (s) => s.segment);\n}\n\n// Greedily packs segments (already contiguous, tiling the source text with no gaps) into groups\n// of at most `working` estimated tokens; a lone segment over `working` still stands alone.\nfunction pack(segments: string[], working: number): string[] {\n  const groups: string[] = [];\n  let current = '';\n  let tokens = 0;\n  for (const segment of segments) {\n    const segmentTokens = estimateTokens(segment);\n    if (current.length > 0 && tokens + segmentTokens > working) {\n      groups.push(current);\n      current = '';\n      tokens = 0;\n    }\n    current += segment;\n    tokens += segmentTokens;\n  }\n  if (current.length > 0) groups.push(current);\n  return groups;\n}\n\n// Line-split alone can't shrink a lone dense line (the CJK case): falls back to sentence then\n// word boundaries (Intl.Segmenter, the same grapheme-safe engine as segment.ts), mode-agnostic on `text`.\nfunction splitLineText(text: string, working: number): string[] {\n  const sentences = segmentsOf(text, 'sentence');\n  const out: string[] = [];\n  let current = '';\n  let tokens = 0;\n  const flush = () => {\n    if (current.length > 0) {\n      out.push(current);\n      current = '';\n      tokens = 0;\n    }\n  };\n  for (const sentence of sentences) {\n    const sentenceTokens = estimateTokens(sentence);\n    if (sentenceTokens > working) {\n      flush();\n      out.push(...pack(segmentsOf(sentence, 'word'), working));\n      continue;\n    }\n    if (current.length > 0 && tokens + sentenceTokens > working) flush();\n    current += sentence;\n    tokens += sentenceTokens;\n  }\n  flush();\n  return out;\n}\n\nconst ATOMIC_TYPES: ReadonlySet<BlockType> = new Set(['code', 'table', 'list']);\n\n// Atomic (code/table/list) pieces are always a raw line slice: re-parsing a table's later pieces\n// without their header/delimiter rows would demote them to paragraph text.\nfunction piece(pieceLines: string[], startLine: number, endLine: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part {\n  const text =\n    ATOMIC_TYPES.has(blockType) || textMode === 'raw'\n      ? pieceLines.join('\\n')\n      : parse(pieceLines.join('\\n'))\n          .map((b) => extractText(b))\n          .join('\\n');\n  return { startLine, endLine, text };\n}\n\n// A one-line piece over working can't shrink via another line-boundary pass (rule 5's gap), so it\n// splits at sentence/word boundaries instead; `final` stops group() re-deriving its text by extent (F5).\nfunction finalizePiece(pieceLines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw'): Part[] {\n  const p = piece(pieceLines, startLine, endLine, blockType, textMode);\n  if (pieceLines.length === 1 && estimateTokens(p.text) > working) {\n    return splitLineText(p.text, working).map((text) => ({ startLine, endLine, text, final: true }));\n  }\n  return [p];\n}\n\n// A block over 2x working size splits at line boundaries into pieces each <= working size, never\n// mid-line. `seed`: pending tokens (e.g. a heading) the first piece must join, checked against the limit.\nfunction splitOversizeBlock(lines: string[], startLine: number, endLine: number, working: number, blockType: BlockType, textMode: 'extracted' | 'raw', seed = 0): Part[] {\n  const pieces: Part[] = [];\n  let pieceLines: string[] = [];\n  let pieceStart = startLine;\n  let tokens = seed;\n  for (let line = startLine; line <= endLine; line++) {\n    const lineText = lines[line - 1];\n    const sep = pieceLines.length > 0 || tokens > 0 ? NEWLINE_TOKENS : 0;\n    const lineTokens = estimateTokens(lineText);\n    if (pieceLines.length > 0 && tokens + sep + lineTokens > working) {\n      pieces.push(...finalizePiece(pieceLines, pieceStart, line - 1, working, blockType, textMode));\n      pieceLines = [];\n      tokens = 0;\n      pieceStart = line;\n      pieceLines.push(lineText);\n      tokens += lineTokens;\n      continue;\n    }\n    pieceLines.push(lineText);\n    tokens += sep + lineTokens;\n  }\n  if (pieceLines.length > 0) pieces.push(...finalizePiece(pieceLines, pieceStart, endLine, working, blockType, textMode));\n  return pieces;\n}\n\n// One heading scope's groups (D1): a heading opens the first group, and an oversize block\n// (rule 5, including an oversize heading) splits into pieces that each close their own group.\nfunction groupScope(scopeBlocks: Block[], lines: string[], resolved: ResolvedOptions): (Chunk & { final?: boolean })[] {\n  const working = resolved.targetTokens;\n  const trigger = working * OVERSIZE_TRIGGER_MULTIPLE;\n  const finished: (Chunk & { final?: boolean })[] = [];\n  let parts: Part[] = [];\n  let paragraphCount = 0;\n  let tokens = 0;\n\n  function close(): void {\n    const group = finalize(parts);\n    if (group) finished.push(group);\n    parts = [];\n    paragraphCount = 0;\n    tokens = 0;\n  }\n\n  // tokens tracks the active text mode's own estimate (a newline between parts costs\n  // NEWLINE_TOKENS too), so packing decisions size the text the chunk will actually ship as.\n  function addPart(text: string, sizeText: string, startLine: number, endLine: number): void {\n    tokens += (parts.length > 0 ? NEWLINE_TOKENS : 0) + estimateTokens(sizeText);\n    parts.push({ startLine, endLine, text });\n  }\n\n  for (const block of scopeBlocks) {\n    const raw = lines.slice(block.startLine - 1, block.endLine).join('\\n');\n    const blockTokens = estimateTokens(raw);\n\n    if (blockTokens > trigger) {\n      const seed = parts.length > 0 ? tokens : 0;\n      for (const p of splitOversizeBlock(lines, block.startLine, block.endLine, working, block.type, resolved.text, seed)) {\n        parts.push(p);\n        close();\n      }\n      continue;\n    }\n\n    const extracted = extractText(block);\n    const sizeText = resolved.text === 'raw' ? raw : extracted;\n\n    if (block.type === 'heading') {\n      addPart(extracted, sizeText, block.startLine, block.endLine);\n      continue;\n    }\n\n    // The 2x-working invariant holds even under pgc's paper-faithful 2-paragraph pairing --\n    // close first if the pair about to form would cross it.\n    const pairOversize = parts.length > 0 && tokens + NEWLINE_TOKENS + blockTokens > trigger;\n    if (pairOversize) close();\n\n    addPart(extracted, sizeText, block.startLine, block.endLine);\n    paragraphCount++;\n\n    if (paragraphCount >= PGC_GROUP_SIZE) close();\n  }\n  close();\n\n  return finished;\n}\n\n// Groups already-parsed blocks per opts (D1/D3), against the same body the blocks were parsed\n// from (line lookups for oversize splitting).\nexport function group(blocks: Block[], body: string, opts?: ChunkOptions): Chunk[] {\n  const resolved = resolveOptions(opts);\n  const lines = body.split('\\n');\n  const chunks: (Chunk & { final?: boolean })[] = [];\n  for (const scope of splitScopes(blocks)) chunks.push(...groupScope(scope, lines, resolved));\n  // 'raw': the chunk's own source lines verbatim, replacing the flavor-resolved join above (D9).\n  // A `final` chunk already carries its own slice's raw text; re-slicing by extent would return the whole shared line.\n  const texted =\n    resolved.text === 'raw'\n      ? chunks.map((c) =>\n          c.final\n            ? c\n            : {\n                ...c,\n                text: lines\n                  .slice(c.startLine - 1, c.endLine)\n                  .join('\\n')\n                  .trim(),\n              }\n        )\n      : chunks;\n  // A group can be all-blank (flavor-stripped to nothing, or a raw slice of pure syntax); it never produces a chunk.\n  return texted.filter((c) => c.text.trim().length > 0).map((c) => ({ startLine: c.startLine, endLine: c.endLine, text: c.text }));\n}\n"],"names":["group","PGC_GROUP_SIZE","OVERSIZE_TRIGGER_MULTIPLE","resolveOptions","opts","targetTokens","DEFAULT_TARGET_TOKENS","text","splitScopes","blocks","scopes","current","block","type","length","push","finalize","parts","undefined","first","last","startLine","endLine","map","p","join","final","some","NEWLINE_TOKENS","estimateTokens","SEGMENTERS","Map","segmentsOf","granularity","segmenter","get","Intl","Segmenter","set","Array","from","segment","s","pack","segments","working","groups","tokens","segmentTokens","splitLineText","sentences","out","flush","sentence","sentenceTokens","ATOMIC_TYPES","Set","piece","pieceLines","blockType","textMode","has","parse","b","extractText","finalizePiece","splitOversizeBlock","lines","seed","pieces","pieceStart","line","lineText","sep","lineTokens","groupScope","scopeBlocks","resolved","trigger","finished","paragraphCount","close","addPart","sizeText","raw","slice","blockTokens","extracted","pairOversize","body","chunks","split","scope","texted","c","trim","filter"],"mappings":";;;;+BAqOgBA;;;eAAAA;;;yBArOY;uBACN;wBACgC;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;AAGtD,IAAMC,iBAAiB;AACvB,IAAMC,4BAA4B;AAOlC,SAASC,eAAeC,IAAmB;;IACzC,OAAO;QACLC,YAAY,UAAED,iBAAAA,2BAAAA,KAAMC,YAAY,uCAAIC,+BAAqB;QACzDC,IAAI,WAAEH,iBAAAA,2BAAAA,KAAMG,IAAI,yCAAI;IACtB;AACF;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,SAASC,YAAYC,MAAe;IAClC,IAAMC,SAAoB,EAAE;IAC5B,IAAIC,UAAmB,EAAE;QACpB,kCAAA,2BAAA;;QAAL,QAAK,YAAeF,2BAAf,SAAA,6BAAA,QAAA,yBAAA,iCAAuB;YAAvB,IAAMG,QAAN;YACH,IAAIA,MAAMC,IAAI,KAAK,aAAaF,QAAQG,MAAM,GAAG,GAAG;gBAClDJ,OAAOK,IAAI,CAACJ;gBACZA,UAAU,EAAE;YACd;YACAA,QAAQI,IAAI,CAACH;QACf;;QANK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAOL,IAAID,QAAQG,MAAM,GAAG,GAAGJ,OAAOK,IAAI,CAACJ;IACpC,OAAOD;AACT;AAWA,SAASM,SAASC,KAAa;IAC7B,IAAIA,MAAMH,MAAM,KAAK,GAAG,OAAOI;IAC/B,IAAMC,QAAQF,KAAK,CAAC,EAAE;IACtB,IAAMG,OAAOH,KAAK,CAACA,MAAMH,MAAM,GAAG,EAAE;IACpC,OAAO;QAAEO,WAAWF,MAAME,SAAS;QAAEC,SAASF,KAAKE,OAAO;QAAEf,MAAMU,MAAMM,GAAG,CAAC,SAACC;mBAAMA,EAAEjB,IAAI;WAAEkB,IAAI,CAAC;QAAOC,OAAOT,MAAMU,IAAI,CAAC,SAACH;mBAAMA,EAAEE,KAAK;;IAAE;AAC3I;AAEA,IAAME,iBAAiBC,IAAAA,wBAAc,EAAC;AACtC,gGAAgG;AAChG,8DAA8D;AAC9D,IAAMC,aAAa,IAAIC;AAEvB,SAASC,WAAWzB,IAAY,EAAE0B,WAAgC;IAChE,IAAIC,YAAYJ,WAAWK,GAAG,CAACF;IAC/B,IAAI,CAACC,WAAW;QACdA,YAAY,IAAIE,KAAKC,SAAS,CAACnB,WAAW;YAAEe,aAAAA;QAAY;QACxDH,WAAWQ,GAAG,CAACL,aAAaC;IAC9B;IACA,OAAOK,MAAMC,IAAI,CAACN,UAAUO,OAAO,CAAClC,OAAO,SAACmC;eAAMA,EAAED,OAAO;;AAC7D;AAEA,gGAAgG;AAChG,2FAA2F;AAC3F,SAASE,KAAKC,QAAkB,EAAEC,OAAe;IAC/C,IAAMC,SAAmB,EAAE;IAC3B,IAAInC,UAAU;IACd,IAAIoC,SAAS;QACR,kCAAA,2BAAA;;QAAL,QAAK,YAAiBH,6BAAjB,SAAA,6BAAA,QAAA,yBAAA,iCAA2B;YAA3B,IAAMH,UAAN;YACH,IAAMO,gBAAgBnB,IAAAA,wBAAc,EAACY;YACrC,IAAI9B,QAAQG,MAAM,GAAG,KAAKiC,SAASC,gBAAgBH,SAAS;gBAC1DC,OAAO/B,IAAI,CAACJ;gBACZA,UAAU;gBACVoC,SAAS;YACX;YACApC,WAAW8B;YACXM,UAAUC;QACZ;;QATK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAUL,IAAIrC,QAAQG,MAAM,GAAG,GAAGgC,OAAO/B,IAAI,CAACJ;IACpC,OAAOmC;AACT;AAEA,8FAA8F;AAC9F,0GAA0G;AAC1G,SAASG,cAAc1C,IAAY,EAAEsC,OAAe;IAClD,IAAMK,YAAYlB,WAAWzB,MAAM;IACnC,IAAM4C,MAAgB,EAAE;IACxB,IAAIxC,UAAU;IACd,IAAIoC,SAAS;IACb,IAAMK,QAAQ;QACZ,IAAIzC,QAAQG,MAAM,GAAG,GAAG;YACtBqC,IAAIpC,IAAI,CAACJ;YACTA,UAAU;YACVoC,SAAS;QACX;IACF;QACK,kCAAA,2BAAA;;QAAL,QAAK,YAAkBG,8BAAlB,SAAA,6BAAA,QAAA,yBAAA,iCAA6B;YAA7B,IAAMG,WAAN;YACH,IAAMC,iBAAiBzB,IAAAA,wBAAc,EAACwB;YACtC,IAAIC,iBAAiBT,SAAS;oBAE5BM;gBADAC;gBACAD,CAAAA,OAAAA,KAAIpC,IAAI,OAARoC,MAAS,qBAAGR,KAAKX,WAAWqB,UAAU,SAASR;gBAC/C;YACF;YACA,IAAIlC,QAAQG,MAAM,GAAG,KAAKiC,SAASO,iBAAiBT,SAASO;YAC7DzC,WAAW0C;YACXN,UAAUO;QACZ;;QAVK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IAWLF;IACA,OAAOD;AACT;AAEA,IAAMI,eAAuC,IAAIC,IAAI;IAAC;IAAQ;IAAS;CAAO;AAE9E,iGAAiG;AACjG,2EAA2E;AAC3E,SAASC,MAAMC,UAAoB,EAAErC,SAAiB,EAAEC,OAAe,EAAEqC,SAAoB,EAAEC,QAA6B;IAC1H,IAAMrD,OACJgD,aAAaM,GAAG,CAACF,cAAcC,aAAa,QACxCF,WAAWjC,IAAI,CAAC,QAChBqC,IAAAA,cAAK,EAACJ,WAAWjC,IAAI,CAAC,OACnBF,GAAG,CAAC,SAACwC;eAAMC,IAAAA,sBAAW,EAACD;OACvBtC,IAAI,CAAC;IACd,OAAO;QAAEJ,WAAAA;QAAWC,SAAAA;QAASf,MAAAA;IAAK;AACpC;AAEA,kGAAkG;AAClG,yGAAyG;AACzG,SAAS0D,cAAcP,UAAoB,EAAErC,SAAiB,EAAEC,OAAe,EAAEuB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;IACnJ,IAAMpC,IAAIiC,MAAMC,YAAYrC,WAAWC,SAASqC,WAAWC;IAC3D,IAAIF,WAAW5C,MAAM,KAAK,KAAKe,IAAAA,wBAAc,EAACL,EAAEjB,IAAI,IAAIsC,SAAS;QAC/D,OAAOI,cAAczB,EAAEjB,IAAI,EAAEsC,SAAStB,GAAG,CAAC,SAAChB;mBAAU;gBAAEc,WAAAA;gBAAWC,SAAAA;gBAASf,MAAAA;gBAAMmB,OAAO;YAAK;;IAC/F;IACA,OAAO;QAACF;KAAE;AACZ;AAEA,iGAAiG;AACjG,0GAA0G;AAC1G,SAAS0C,mBAAmBC,KAAe,EAAE9C,SAAiB,EAAEC,OAAe,EAAEuB,OAAe,EAAEc,SAAoB,EAAEC,QAA6B;QAAEQ,OAAAA,iEAAO;QAqBjIC;IApB3B,IAAMA,SAAiB,EAAE;IACzB,IAAIX,aAAuB,EAAE;IAC7B,IAAIY,aAAajD;IACjB,IAAI0B,SAASqB;IACb,IAAK,IAAIG,OAAOlD,WAAWkD,QAAQjD,SAASiD,OAAQ;QAClD,IAAMC,WAAWL,KAAK,CAACI,OAAO,EAAE;QAChC,IAAME,MAAMf,WAAW5C,MAAM,GAAG,KAAKiC,SAAS,IAAInB,iBAAiB;QACnE,IAAM8C,aAAa7C,IAAAA,wBAAc,EAAC2C;QAClC,IAAId,WAAW5C,MAAM,GAAG,KAAKiC,SAAS0B,MAAMC,aAAa7B,SAAS;gBAChEwB;YAAAA,CAAAA,WAAAA,QAAOtD,IAAI,OAAXsD,UAAY,qBAAGJ,cAAcP,YAAYY,YAAYC,OAAO,GAAG1B,SAASc,WAAWC;YACnFF,aAAa,EAAE;YACfX,SAAS;YACTuB,aAAaC;YACbb,WAAW3C,IAAI,CAACyD;YAChBzB,UAAU2B;YACV;QACF;QACAhB,WAAW3C,IAAI,CAACyD;QAChBzB,UAAU0B,MAAMC;IAClB;IACA,IAAIhB,WAAW5C,MAAM,GAAG,GAAGuD,CAAAA,UAAAA,QAAOtD,IAAI,OAAXsD,SAAY,qBAAGJ,cAAcP,YAAYY,YAAYhD,SAASuB,SAASc,WAAWC;IAC7G,OAAOS;AACT;AAEA,0FAA0F;AAC1F,8FAA8F;AAC9F,SAASM,WAAWC,WAAoB,EAAET,KAAe,EAAEU,QAAyB;IAClF,IAAMhC,UAAUgC,SAASxE,YAAY;IACrC,IAAMyE,UAAUjC,UAAU3C;IAC1B,IAAM6E,WAA4C,EAAE;IACpD,IAAI9D,QAAgB,EAAE;IACtB,IAAI+D,iBAAiB;IACrB,IAAIjC,SAAS;IAEb,SAASkC;QACP,IAAMjF,QAAQgB,SAASC;QACvB,IAAIjB,OAAO+E,SAAShE,IAAI,CAACf;QACzBiB,QAAQ,EAAE;QACV+D,iBAAiB;QACjBjC,SAAS;IACX;IAEA,mFAAmF;IACnF,2FAA2F;IAC3F,SAASmC,QAAQ3E,IAAY,EAAE4E,QAAgB,EAAE9D,SAAiB,EAAEC,OAAe;QACjFyB,UAAU,AAAC9B,CAAAA,MAAMH,MAAM,GAAG,IAAIc,iBAAiB,CAAA,IAAKC,IAAAA,wBAAc,EAACsD;QACnElE,MAAMF,IAAI,CAAC;YAAEM,WAAAA;YAAWC,SAAAA;YAASf,MAAAA;QAAK;IACxC;QAEK,kCAAA,2BAAA;;QAAL,QAAK,YAAeqE,gCAAf,SAAA,6BAAA,QAAA,yBAAA,iCAA4B;YAA5B,IAAMhE,QAAN;YACH,IAAMwE,MAAMjB,MAAMkB,KAAK,CAACzE,MAAMS,SAAS,GAAG,GAAGT,MAAMU,OAAO,EAAEG,IAAI,CAAC;YACjE,IAAM6D,cAAczD,IAAAA,wBAAc,EAACuD;YAEnC,IAAIE,cAAcR,SAAS;gBACzB,IAAMV,OAAOnD,MAAMH,MAAM,GAAG,IAAIiC,SAAS;oBACpC,mCAAA,4BAAA;;oBAAL,QAAK,aAAWmB,mBAAmBC,OAAOvD,MAAMS,SAAS,EAAET,MAAMU,OAAO,EAAEuB,SAASjC,MAAMC,IAAI,EAAEgE,SAAStE,IAAI,EAAE6D,0BAAzG,UAAA,8BAAA,SAAA,0BAAA,kCAAgH;wBAAhH,IAAM5C,IAAN;wBACHP,MAAMF,IAAI,CAACS;wBACXyD;oBACF;;oBAHK;oBAAA;;;6BAAA,8BAAA;4BAAA;;;4BAAA;kCAAA;;;;gBAIL;YACF;YAEA,IAAMM,YAAYvB,IAAAA,sBAAW,EAACpD;YAC9B,IAAMuE,WAAWN,SAAStE,IAAI,KAAK,QAAQ6E,MAAMG;YAEjD,IAAI3E,MAAMC,IAAI,KAAK,WAAW;gBAC5BqE,QAAQK,WAAWJ,UAAUvE,MAAMS,SAAS,EAAET,MAAMU,OAAO;gBAC3D;YACF;YAEA,wFAAwF;YACxF,wDAAwD;YACxD,IAAMkE,eAAevE,MAAMH,MAAM,GAAG,KAAKiC,SAASnB,iBAAiB0D,cAAcR;YACjF,IAAIU,cAAcP;YAElBC,QAAQK,WAAWJ,UAAUvE,MAAMS,SAAS,EAAET,MAAMU,OAAO;YAC3D0D;YAEA,IAAIA,kBAAkB/E,gBAAgBgF;QACxC;;QA9BK;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IA+BLA;IAEA,OAAOF;AACT;AAIO,SAAS/E,MAAMS,MAAe,EAAEgF,IAAY,EAAErF,IAAmB;QAI7BsF;IAHzC,IAAMb,WAAW1E,eAAeC;IAChC,IAAM+D,QAAQsB,KAAKE,KAAK,CAAC;IACzB,IAAMD,SAA0C,EAAE;QAC7C,kCAAA,2BAAA;;QAAL,QAAK,YAAelF,YAAYC,4BAA3B,SAAA,6BAAA,QAAA,yBAAA;YAAA,IAAMmF,QAAN;YAAoCF,CAAAA,UAAAA,QAAO3E,IAAI,OAAX2E,SAAY,qBAAGf,WAAWiB,OAAOzB,OAAOU;;;QAA5E;QAAA;;;iBAAA,6BAAA;gBAAA;;;gBAAA;sBAAA;;;;IACL,+FAA+F;IAC/F,qHAAqH;IACrH,IAAMgB,SACJhB,SAAStE,IAAI,KAAK,QACdmF,OAAOnE,GAAG,CAAC,SAACuE;eACVA,EAAEpE,KAAK,GACHoE,IACA,wCACKA;YACHvF,MAAM4D,MACHkB,KAAK,CAACS,EAAEzE,SAAS,GAAG,GAAGyE,EAAExE,OAAO,EAChCG,IAAI,CAAC,MACLsE,IAAI;;SAGfL;IACN,mHAAmH;IACnH,OAAOG,OAAOG,MAAM,CAAC,SAACF;eAAMA,EAAEvF,IAAI,CAACwF,IAAI,GAAGjF,MAAM,GAAG;OAAGS,GAAG,CAAC,SAACuE;eAAO;YAAEzE,WAAWyE,EAAEzE,SAAS;YAAEC,SAASwE,EAAExE,OAAO;YAAEf,MAAMuF,EAAEvF,IAAI;QAAC;;AAC/H"}

package/dist/cjs/chunk/parse.js

@@ -8,78 +8,74 @@ Object.defineProperty(exports, "parse", {
         return parse;
     }
 });
-var _nodemodule = /*#__PURE__*/ _interop_require_default(require("node:module"));
 var _extractts = require("./extract.js");
-function _interop_require_default(obj) {
-    return obj && obj.__esModule ? obj : {
-        default: obj
-    };
-}
-// Tier-2, as embed/static.ts: the parser's packages cost ~19 ms to load and a warm tree never
-// parses, so every store-opening command paid for them until a file actually changed.
-var _require = typeof require === 'undefined' ? _nodemodule.default.createRequire(require("url").pathToFileURL(__filename).toString()) : require;
+var _parserts = require("./parser.js");
+// Opening token type to Block type. Everything else (rules, raw html, footnote definitions) is
+// 'other', as mdast's BLOCK_TYPES mapped no entry for those nodes.
 var BLOCK_TYPES = {
-    heading: 'heading',
-    paragraph: 'paragraph',
-    code: 'code',
-    table: 'table',
-    list: 'list',
-    blockquote: 'blockquote'
+    heading_open: 'heading',
+    paragraph_open: 'paragraph',
+    fence: 'code',
+    code_block: 'code',
+    table_open: 'table',
+    ordered_list_open: 'list',
+    bullet_list_open: 'list',
+    blockquote_open: 'blockquote'
 };
-var cached;
-// Imported individually, not via micromark-extension-gfm/mdast-util-gfm: those bundles also pull
-// in gfm-tagfilter, an HTML sanitizer this library never uses (no htmlExtensions call anywhere).
-function parser() {
-    if (cached) return cached;
-    var fromMarkdown = _require('mdast-util-from-markdown').fromMarkdown;
-    var gfmAutolinkLiteralFromMarkdown = _require('mdast-util-gfm-autolink-literal').gfmAutolinkLiteralFromMarkdown;
-    var gfmFootnoteFromMarkdown = _require('mdast-util-gfm-footnote').gfmFootnoteFromMarkdown;
-    var gfmStrikethroughFromMarkdown = _require('mdast-util-gfm-strikethrough').gfmStrikethroughFromMarkdown;
-    var gfmTableFromMarkdown = _require('mdast-util-gfm-table').gfmTableFromMarkdown;
-    var gfmTaskListItemFromMarkdown = _require('mdast-util-gfm-task-list-item').gfmTaskListItemFromMarkdown;
-    var gfmAutolinkLiteral = _require('micromark-extension-gfm-autolink-literal').gfmAutolinkLiteral;
-    var gfmFootnote = _require('micromark-extension-gfm-footnote').gfmFootnote;
-    var gfmStrikethrough = _require('micromark-extension-gfm-strikethrough').gfmStrikethrough;
-    var gfmTable = _require('micromark-extension-gfm-table').gfmTable;
-    var gfmTaskListItem = _require('micromark-extension-gfm-task-list-item').gfmTaskListItem;
-    cached = {
-        fromMarkdown: fromMarkdown,
-        options: {
-            extensions: [
-                gfmAutolinkLiteral(),
-                gfmFootnote(),
-                gfmStrikethrough(),
-                gfmTable(),
-                gfmTaskListItem()
-            ],
-            mdastExtensions: [
-                gfmAutolinkLiteralFromMarkdown(),
-                gfmFootnoteFromMarkdown(),
-                gfmStrikethroughFromMarkdown(),
-                gfmTableFromMarkdown(),
-                gfmTaskListItemFromMarkdown()
-            ]
+function parse(body) {
+    var tokens = (0, _parserts.parser)().parse(body, {});
+    var lines = body.split('\n');
+    var blocks = [];
+    var i = 0;
+    while(i < tokens.length){
+        var token = tokens[i];
+        if (token.nesting === 0) {
+            var _BLOCK_TYPES_token_type;
+            blocks.push(makeBlock((_BLOCK_TYPES_token_type = BLOCK_TYPES[token.type]) !== null && _BLOCK_TYPES_token_type !== void 0 ? _BLOCK_TYPES_token_type : 'other', tokens, i, i + 1, lines));
+            i += 1;
+        } else if (token.nesting === 1) {
+            var _BLOCK_TYPES_token_type1;
+            var depth = 1;
+            var j = i + 1;
+            while(j < tokens.length && depth > 0){
+                depth += tokens[j].nesting === 1 ? 1 : tokens[j].nesting === -1 ? -1 : 0;
+                j += 1;
+            }
+            blocks.push(makeBlock((_BLOCK_TYPES_token_type1 = BLOCK_TYPES[token.type]) !== null && _BLOCK_TYPES_token_type1 !== void 0 ? _BLOCK_TYPES_token_type1 : 'other', tokens, i, j, lines));
+            i = j;
+        } else {
+            // Defensive: the balanced scan above consumes every close belonging to an open.
+            i += 1;
         }
-    };
-    return cached;
+    }
+    return blocks;
 }
-function parse(body) {
-    var _parser = parser(), fromMarkdown = _parser.fromMarkdown, options = _parser.options;
-    var tree = fromMarkdown(body, options);
-    return tree.children.map(function(node) {
-        var _BLOCK_TYPES_node_type;
-        var position = node.position;
-        var block = {
-            type: (_BLOCK_TYPES_node_type = BLOCK_TYPES[node.type]) !== null && _BLOCK_TYPES_node_type !== void 0 ? _BLOCK_TYPES_node_type : 'other',
-            startLine: position ? position.start.line : 1,
-            endLine: position ? position.end.line : 1,
-            node: node
-        };
-        if (node.type === 'heading') {
-            block.depth = node.depth;
-            block.text = (0, _extractts.extractText)(node);
+// 1-based inclusive extent from the min/max of the tokens' maps (0-based half-open); a block
+// whose tokens carry no map (an empty footnote definition) falls back to line 1. Trailing
+// blank lines trim so a list's endLine lands where mdast's position ended it.
+function makeBlock(type, tokens, i, j, lines) {
+    var start = Infinity;
+    var end = 0;
+    for(var k = i; k < j; k++){
+        var map = tokens[k].map;
+        if (map) {
+            start = Math.min(start, map[0]);
+            end = Math.max(end, map[1]);
         }
-        return block;
-    });
+    }
+    var startLine = Number.isFinite(start) ? start + 1 : 1;
+    var endLine = end > start ? end : startLine;
+    while(endLine > startLine && lines[endLine - 1].trim() === '')endLine--;
+    var block = {
+        type: type,
+        startLine: startLine,
+        endLine: endLine,
+        node: tokens.slice(i, j)
+    };
+    if (type === 'heading') {
+        block.depth = Number(tokens[i].tag.slice(1));
+        block.text = (0, _extractts.extractText)(block);
+    }
+    return block;
 }
 /* CJS INTEROP */ if (exports.__esModule && exports.default) { try { Object.defineProperty(exports.default, '__esModule', { value: true }); for (var key in exports) { exports.default[key] = exports[key]; } } catch (_) {}; module.exports = exports.default; }

package/dist/cjs/chunk/parse.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/parse.ts"],"sourcesContent":["import Module from 'node:module';\nimport type { RootContent } from 'mdast';\nimport { extractText } from './extract.ts';\nimport type { Block, BlockType } from './types.ts';\n\n// Tier-2, as embed/static.ts: the parser's packages cost ~19 ms to load and a warm tree never\n// parses, so every store-opening command paid for them until a file actually changed.\nconst _require = typeof require === 'undefined' ? Module.createRequire(import.meta.url) : require;\n\nconst BLOCK_TYPES: Partial<Record<RootContent['type'], BlockType>> = {\n  heading: 'heading',\n  paragraph: 'paragraph',\n  code: 'code',\n  table: 'table',\n  list: 'list',\n  blockquote: 'blockquote',\n};\n\ntype FromMarkdown = typeof import('mdast-util-from-markdown').fromMarkdown;\ntype Parser = { fromMarkdown: FromMarkdown; options: NonNullable<Parameters<FromMarkdown>[1]> };\nlet cached: Parser | undefined;\n\n// Imported individually, not via micromark-extension-gfm/mdast-util-gfm: those bundles also pull\n// in gfm-tagfilter, an HTML sanitizer this library never uses (no htmlExtensions call anywhere).\nfunction parser(): Parser {\n  if (cached) return cached;\n  const { fromMarkdown } = _require('mdast-util-from-markdown') as typeof import('mdast-util-from-markdown');\n  const { gfmAutolinkLiteralFromMarkdown } = _require('mdast-util-gfm-autolink-literal') as typeof import('mdast-util-gfm-autolink-literal');\n  const { gfmFootnoteFromMarkdown } = _require('mdast-util-gfm-footnote') as typeof import('mdast-util-gfm-footnote');\n  const { gfmStrikethroughFromMarkdown } = _require('mdast-util-gfm-strikethrough') as typeof import('mdast-util-gfm-strikethrough');\n  const { gfmTableFromMarkdown } = _require('mdast-util-gfm-table') as typeof import('mdast-util-gfm-table');\n  const { gfmTaskListItemFromMarkdown } = _require('mdast-util-gfm-task-list-item') as typeof import('mdast-util-gfm-task-list-item');\n  const { gfmAutolinkLiteral } = _require('micromark-extension-gfm-autolink-literal') as typeof import('micromark-extension-gfm-autolink-literal');\n  const { gfmFootnote } = _require('micromark-extension-gfm-footnote') as typeof import('micromark-extension-gfm-footnote');\n  const { gfmStrikethrough } = _require('micromark-extension-gfm-strikethrough') as typeof import('micromark-extension-gfm-strikethrough');\n  const { gfmTable } = _require('micromark-extension-gfm-table') as typeof import('micromark-extension-gfm-table');\n  const { gfmTaskListItem } = _require('micromark-extension-gfm-task-list-item') as typeof import('micromark-extension-gfm-task-list-item');\n  cached = {\n    fromMarkdown,\n    options: {\n      extensions: [gfmAutolinkLiteral(), gfmFootnote(), gfmStrikethrough(), gfmTable(), gfmTaskListItem()],\n      mdastExtensions: [gfmAutolinkLiteralFromMarkdown(), gfmFootnoteFromMarkdown(), gfmStrikethroughFromMarkdown(), gfmTableFromMarkdown(), gfmTaskListItemFromMarkdown()],\n    },\n  };\n  return cached;\n}\n\n// Top-level blocks of a markdown body, typed and line-extent bounded from mdast's own\n// node.position (never a regex guess). GFM extensions add tables, task lists, footnotes, strikethrough.\nexport function parse(body: string): Block[] {\n  const { fromMarkdown, options } = parser();\n  const tree = fromMarkdown(body, options);\n  return tree.children.map((node) => {\n    const position = node.position;\n    const block: Block = {\n      type: BLOCK_TYPES[node.type] ?? 'other',\n      startLine: position ? position.start.line : 1,\n      endLine: position ? position.end.line : 1,\n      node,\n    };\n    if (node.type === 'heading') {\n      block.depth = node.depth;\n      block.text = extractText(node);\n    }\n    return block;\n  });\n}\n"],"names":["parse","_require","require","Module","createRequire","BLOCK_TYPES","heading","paragraph","code","table","list","blockquote","cached","parser","fromMarkdown","gfmAutolinkLiteralFromMarkdown","gfmFootnoteFromMarkdown","gfmStrikethroughFromMarkdown","gfmTableFromMarkdown","gfmTaskListItemFromMarkdown","gfmAutolinkLiteral","gfmFootnote","gfmStrikethrough","gfmTable","gfmTaskListItem","options","extensions","mdastExtensions","body","tree","children","map","node","position","block","type","startLine","start","line","endLine","end","depth","text","extractText"],"mappings":";;;;+BAiDgBA;;;eAAAA;;;iEAjDG;yBAES;;;;;;AAG5B,8FAA8F;AAC9F,sFAAsF;AACtF,IAAMC,WAAW,OAAOC,YAAY,cAAcC,mBAAM,CAACC,aAAa,CAAC,uDAAmBF;AAE1F,IAAMG,cAA+D;IACnEC,SAAS;IACTC,WAAW;IACXC,MAAM;IACNC,OAAO;IACPC,MAAM;IACNC,YAAY;AACd;AAIA,IAAIC;AAEJ,iGAAiG;AACjG,iGAAiG;AACjG,SAASC;IACP,IAAID,QAAQ,OAAOA;IACnB,IAAM,AAAEE,eAAiBb,SAAS,4BAA1Ba;IACR,IAAM,AAAEC,iCAAmCd,SAAS,mCAA5Cc;IACR,IAAM,AAAEC,0BAA4Bf,SAAS,2BAArCe;IACR,IAAM,AAAEC,+BAAiChB,SAAS,gCAA1CgB;IACR,IAAM,AAAEC,uBAAyBjB,SAAS,wBAAlCiB;IACR,IAAM,AAAEC,8BAAgClB,SAAS,iCAAzCkB;IACR,IAAM,AAAEC,qBAAuBnB,SAAS,4CAAhCmB;IACR,IAAM,AAAEC,cAAgBpB,SAAS,oCAAzBoB;IACR,IAAM,AAAEC,mBAAqBrB,SAAS,yCAA9BqB;IACR,IAAM,AAAEC,WAAatB,SAAS,iCAAtBsB;IACR,IAAM,AAAEC,kBAAoBvB,SAAS,0CAA7BuB;IACRZ,SAAS;QACPE,cAAAA;QACAW,SAAS;YACPC,YAAY;gBAACN;gBAAsBC;gBAAeC;gBAAoBC;gBAAYC;aAAkB;YACpGG,iBAAiB;gBAACZ;gBAAkCC;gBAA2BC;gBAAgCC;gBAAwBC;aAA8B;QACvK;IACF;IACA,OAAOP;AACT;AAIO,SAASZ,MAAM4B,IAAY;IAChC,IAAkCf,UAAAA,UAA1BC,eAA0BD,QAA1BC,cAAcW,UAAYZ,QAAZY;IACtB,IAAMI,OAAOf,aAAac,MAAMH;IAChC,OAAOI,KAAKC,QAAQ,CAACC,GAAG,CAAC,SAACC;YAGhB3B;QAFR,IAAM4B,WAAWD,KAAKC,QAAQ;QAC9B,IAAMC,QAAe;YACnBC,IAAI,GAAE9B,yBAAAA,WAAW,CAAC2B,KAAKG,IAAI,CAAC,cAAtB9B,oCAAAA,yBAA0B;YAChC+B,WAAWH,WAAWA,SAASI,KAAK,CAACC,IAAI,GAAG;YAC5CC,SAASN,WAAWA,SAASO,GAAG,CAACF,IAAI,GAAG;YACxCN,MAAAA;QACF;QACA,IAAIA,KAAKG,IAAI,KAAK,WAAW;YAC3BD,MAAMO,KAAK,GAAGT,KAAKS,KAAK;YACxBP,MAAMQ,IAAI,GAAGC,IAAAA,sBAAW,EAACX;QAC3B;QACA,OAAOE;IACT;AACF"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/parse.ts"],"sourcesContent":["import type { Token } from 'markdown-it';\nimport { extractText } from './extract.ts';\nimport { parser } from './parser.ts';\nimport type { Block, BlockType } from './types.ts';\n\n// Opening token type to Block type. Everything else (rules, raw html, footnote definitions) is\n// 'other', as mdast's BLOCK_TYPES mapped no entry for those nodes.\nconst BLOCK_TYPES: Record<string, BlockType> = {\n  heading_open: 'heading',\n  paragraph_open: 'paragraph',\n  fence: 'code',\n  code_block: 'code',\n  table_open: 'table',\n  ordered_list_open: 'list',\n  bullet_list_open: 'list',\n  blockquote_open: 'blockquote',\n};\n\n// Top-level blocks of a markdown body, typed and line-extent bounded from markdown-it's own\n// token maps (never a regex guess). html and linkify on, with the footnote and task-list plugins.\nexport function parse(body: string): Block[] {\n  const tokens = parser().parse(body, {});\n  const lines = body.split('\\n');\n  const blocks: Block[] = [];\n  let i = 0;\n  while (i < tokens.length) {\n    const token = tokens[i];\n    if (token.nesting === 0) {\n      blocks.push(makeBlock(BLOCK_TYPES[token.type] ?? 'other', tokens, i, i + 1, lines));\n      i += 1;\n    } else if (token.nesting === 1) {\n      let depth = 1;\n      let j = i + 1;\n      while (j < tokens.length && depth > 0) {\n        depth += tokens[j].nesting === 1 ? 1 : tokens[j].nesting === -1 ? -1 : 0;\n        j += 1;\n      }\n      blocks.push(makeBlock(BLOCK_TYPES[token.type] ?? 'other', tokens, i, j, lines));\n      i = j;\n    } else {\n      // Defensive: the balanced scan above consumes every close belonging to an open.\n      i += 1;\n    }\n  }\n  return blocks;\n}\n\n// 1-based inclusive extent from the min/max of the tokens' maps (0-based half-open); a block\n// whose tokens carry no map (an empty footnote definition) falls back to line 1. Trailing\n// blank lines trim so a list's endLine lands where mdast's position ended it.\nfunction makeBlock(type: BlockType, tokens: Token[], i: number, j: number, lines: string[]): Block {\n  let start = Infinity;\n  let end = 0;\n  for (let k = i; k < j; k++) {\n    const map = tokens[k].map;\n    if (map) {\n      start = Math.min(start, map[0]);\n      end = Math.max(end, map[1]);\n    }\n  }\n  const startLine = Number.isFinite(start) ? start + 1 : 1;\n  let endLine = end > start ? end : startLine;\n  while (endLine > startLine && lines[endLine - 1].trim() === '') endLine--;\n  const block: Block = { type, startLine, endLine, node: tokens.slice(i, j) };\n  if (type === 'heading') {\n    block.depth = Number(tokens[i].tag.slice(1));\n    block.text = extractText(block);\n  }\n  return block;\n}\n"],"names":["parse","BLOCK_TYPES","heading_open","paragraph_open","fence","code_block","table_open","ordered_list_open","bullet_list_open","blockquote_open","body","tokens","parser","lines","split","blocks","i","length","token","nesting","push","makeBlock","type","depth","j","start","Infinity","end","k","map","Math","min","max","startLine","Number","isFinite","endLine","trim","block","node","slice","tag","text","extractText"],"mappings":";;;;+BAoBgBA;;;eAAAA;;;yBAnBY;wBACL;AAGvB,+FAA+F;AAC/F,mEAAmE;AACnE,IAAMC,cAAyC;IAC7CC,cAAc;IACdC,gBAAgB;IAChBC,OAAO;IACPC,YAAY;IACZC,YAAY;IACZC,mBAAmB;IACnBC,kBAAkB;IAClBC,iBAAiB;AACnB;AAIO,SAAST,MAAMU,IAAY;IAChC,IAAMC,SAASC,IAAAA,gBAAM,IAAGZ,KAAK,CAACU,MAAM,CAAC;IACrC,IAAMG,QAAQH,KAAKI,KAAK,CAAC;IACzB,IAAMC,SAAkB,EAAE;IAC1B,IAAIC,IAAI;IACR,MAAOA,IAAIL,OAAOM,MAAM,CAAE;QACxB,IAAMC,QAAQP,MAAM,CAACK,EAAE;QACvB,IAAIE,MAAMC,OAAO,KAAK,GAAG;gBACDlB;YAAtBc,OAAOK,IAAI,CAACC,WAAUpB,0BAAAA,WAAW,CAACiB,MAAMI,IAAI,CAAC,cAAvBrB,qCAAAA,0BAA2B,SAASU,QAAQK,GAAGA,IAAI,GAAGH;YAC5EG,KAAK;QACP,OAAO,IAAIE,MAAMC,OAAO,KAAK,GAAG;gBAORlB;YANtB,IAAIsB,QAAQ;YACZ,IAAIC,IAAIR,IAAI;YACZ,MAAOQ,IAAIb,OAAOM,MAAM,IAAIM,QAAQ,EAAG;gBACrCA,SAASZ,MAAM,CAACa,EAAE,CAACL,OAAO,KAAK,IAAI,IAAIR,MAAM,CAACa,EAAE,CAACL,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI;gBACvEK,KAAK;YACP;YACAT,OAAOK,IAAI,CAACC,WAAUpB,2BAAAA,WAAW,CAACiB,MAAMI,IAAI,CAAC,cAAvBrB,sCAAAA,2BAA2B,SAASU,QAAQK,GAAGQ,GAAGX;YACxEG,IAAIQ;QACN,OAAO;YACL,gFAAgF;YAChFR,KAAK;QACP;IACF;IACA,OAAOD;AACT;AAEA,6FAA6F;AAC7F,0FAA0F;AAC1F,8EAA8E;AAC9E,SAASM,UAAUC,IAAe,EAAEX,MAAe,EAAEK,CAAS,EAAEQ,CAAS,EAAEX,KAAe;IACxF,IAAIY,QAAQC;IACZ,IAAIC,MAAM;IACV,IAAK,IAAIC,IAAIZ,GAAGY,IAAIJ,GAAGI,IAAK;QAC1B,IAAMC,MAAMlB,MAAM,CAACiB,EAAE,CAACC,GAAG;QACzB,IAAIA,KAAK;YACPJ,QAAQK,KAAKC,GAAG,CAACN,OAAOI,GAAG,CAAC,EAAE;YAC9BF,MAAMG,KAAKE,GAAG,CAACL,KAAKE,GAAG,CAAC,EAAE;QAC5B;IACF;IACA,IAAMI,YAAYC,OAAOC,QAAQ,CAACV,SAASA,QAAQ,IAAI;IACvD,IAAIW,UAAUT,MAAMF,QAAQE,MAAMM;IAClC,MAAOG,UAAUH,aAAapB,KAAK,CAACuB,UAAU,EAAE,CAACC,IAAI,OAAO,GAAID;IAChE,IAAME,QAAe;QAAEhB,MAAAA;QAAMW,WAAAA;QAAWG,SAAAA;QAASG,MAAM5B,OAAO6B,KAAK,CAACxB,GAAGQ;IAAG;IAC1E,IAAIF,SAAS,WAAW;QACtBgB,MAAMf,KAAK,GAAGW,OAAOvB,MAAM,CAACK,EAAE,CAACyB,GAAG,CAACD,KAAK,CAAC;QACzCF,MAAMI,IAAI,GAAGC,IAAAA,sBAAW,EAACL;IAC3B;IACA,OAAOA;AACT"}

package/dist/cjs/chunk/parser.d.cts

@@ -0,0 +1,2 @@
+import type { MarkdownIt } from 'markdown-it';
+export declare function parser(): MarkdownIt;

package/dist/cjs/chunk/parser.d.ts

@@ -0,0 +1,2 @@
+import type { MarkdownIt } from 'markdown-it';
+export declare function parser(): MarkdownIt;

package/dist/cjs/chunk/parser.js

@@ -0,0 +1,40 @@
+"use strict";
+Object.defineProperty(exports, "__esModule", {
+    value: true
+});
+Object.defineProperty(exports, "parser", {
+    enumerable: true,
+    get: function() {
+        return parser;
+    }
+});
+var _nodemodule = /*#__PURE__*/ _interop_require_default(require("node:module"));
+function _interop_require_default(obj) {
+    return obj && obj.__esModule ? obj : {
+        default: obj
+    };
+}
+// Tier-2, as embed/static.ts: the parser's packages cost ~19 ms to load and a warm tree never
+// parses, so every store-opening command paid for them until a file actually changed.
+var _require = typeof require === 'undefined' ? _nodemodule.default.createRequire(require("url").pathToFileURL(__filename).toString()) : require;
+var cached;
+function parser() {
+    if (cached) return cached;
+    var Ctor = _require('markdown-it');
+    var footnote = _require('markdown-it-footnote');
+    var taskLists = _require('markdown-it-task-lists');
+    var md = new Ctor({
+        html: true,
+        linkify: true
+    }).use(footnote).use(taskLists);
+    // Fuzzy linking is what links www. domains and emails (the scheme matcher does neither); its
+    // bare-domain over-matching vs GFM is reined in at extraction, where the text is kept.
+    md.linkify.set({
+        fuzzyLink: true
+    });
+    md.core.ruler.disable('footnote_tail');
+    md.inline.ruler.disable('footnote_inline');
+    cached = md;
+    return cached;
+}
+/* CJS INTEROP */ if (exports.__esModule && exports.default) { try { Object.defineProperty(exports.default, '__esModule', { value: true }); for (var key in exports) { exports.default[key] = exports[key]; } } catch (_) {}; module.exports = exports.default; }

package/dist/cjs/chunk/parser.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/parser.ts"],"sourcesContent":["import Module from 'node:module';\nimport type { MarkdownIt } from 'markdown-it';\nimport type footnotePlugin from 'markdown-it-footnote';\nimport type taskListsPlugin from 'markdown-it-task-lists';\n\n// Tier-2, as embed/static.ts: the parser's packages cost ~19 ms to load and a warm tree never\n// parses, so every store-opening command paid for them until a file actually changed.\nconst _require = typeof require === 'undefined' ? Module.createRequire(import.meta.url) : require;\n\n// The CJS build resolves markdown-it's export= types, so the constructor is named locally.\ntype Ctor = new (options?: { html?: boolean; linkify?: boolean }) => MarkdownIt;\n\nlet cached: MarkdownIt | undefined;\n\n// The one shared parser: html and linkify on, fuzzy links, footnotes and task lists, with the\n// footnote rules disabled for GFM parity (definitions stay spans; ^[...] stays inert text).\nexport function parser(): MarkdownIt {\n  if (cached) return cached;\n  const Ctor = _require('markdown-it') as Ctor;\n  const footnote = _require('markdown-it-footnote') as typeof footnotePlugin;\n  const taskLists = _require('markdown-it-task-lists') as typeof taskListsPlugin;\n  const md = new Ctor({ html: true, linkify: true }).use(footnote).use(taskLists);\n  // Fuzzy linking is what links www. domains and emails (the scheme matcher does neither); its\n  // bare-domain over-matching vs GFM is reined in at extraction, where the text is kept.\n  md.linkify.set({ fuzzyLink: true });\n  md.core.ruler.disable('footnote_tail');\n  md.inline.ruler.disable('footnote_inline');\n  cached = md;\n  return cached;\n}\n"],"names":["parser","_require","require","Module","createRequire","cached","Ctor","footnote","taskLists","md","html","linkify","use","set","fuzzyLink","core","ruler","disable","inline"],"mappings":";;;;+BAgBgBA;;;eAAAA;;;iEAhBG;;;;;;AAKnB,8FAA8F;AAC9F,sFAAsF;AACtF,IAAMC,WAAW,OAAOC,YAAY,cAAcC,mBAAM,CAACC,aAAa,CAAC,uDAAmBF;AAK1F,IAAIG;AAIG,SAASL;IACd,IAAIK,QAAQ,OAAOA;IACnB,IAAMC,OAAOL,SAAS;IACtB,IAAMM,WAAWN,SAAS;IAC1B,IAAMO,YAAYP,SAAS;IAC3B,IAAMQ,KAAK,IAAIH,KAAK;QAAEI,MAAM;QAAMC,SAAS;IAAK,GAAGC,GAAG,CAACL,UAAUK,GAAG,CAACJ;IACrE,6FAA6F;IAC7F,uFAAuF;IACvFC,GAAGE,OAAO,CAACE,GAAG,CAAC;QAAEC,WAAW;IAAK;IACjCL,GAAGM,IAAI,CAACC,KAAK,CAACC,OAAO,CAAC;IACtBR,GAAGS,MAAM,CAACF,KAAK,CAACC,OAAO,CAAC;IACxBZ,SAASI;IACT,OAAOJ;AACT"}

package/dist/cjs/chunk/types.d.cts

@@ -1,4 +1,4 @@
-import type { RootContent } from 'mdast';
+import type { Token } from 'markdown-it';
 export type BlockType = 'heading' | 'paragraph' | 'code' | 'table' | 'list' | 'blockquote' | 'other';
 export interface Block {
     type: BlockType;
@@ -6,7 +6,7 @@ export interface Block {
     endLine: number;
     depth?: number;
     text?: string;
-    node: RootContent;
+    node: Token[];
 }
 export interface ChunkOptions {
     targetTokens?: number;

package/dist/cjs/chunk/types.d.ts

@@ -1,4 +1,4 @@
-import type { RootContent } from 'mdast';
+import type { Token } from 'markdown-it';
 export type BlockType = 'heading' | 'paragraph' | 'code' | 'table' | 'list' | 'blockquote' | 'other';
 export interface Block {
     type: BlockType;
@@ -6,7 +6,7 @@ export interface Block {
     endLine: number;
     depth?: number;
     text?: string;
-    node: RootContent;
+    node: Token[];
 }
 export interface ChunkOptions {
     targetTokens?: number;

package/dist/cjs/chunk/version.d.cts

@@ -1 +1 @@
-export declare const CHUNK_VERSION = "chunk:v5";
+export declare const CHUNK_VERSION = "chunk:v6";

package/dist/cjs/chunk/version.d.ts

@@ -1 +1 @@
-export declare const CHUNK_VERSION = "chunk:v5";
+export declare const CHUNK_VERSION = "chunk:v6";

package/dist/cjs/chunk/version.js

@@ -10,5 +10,5 @@ Object.defineProperty(exports, "CHUNK_VERSION", {
         return CHUNK_VERSION;
     }
 });
-var CHUNK_VERSION = 'chunk:v5';
+var CHUNK_VERSION = 'chunk:v6';
 /* CJS INTEROP */ if (exports.__esModule && exports.default) { try { Object.defineProperty(exports.default, '__esModule', { value: true }); for (var key in exports) { exports.default[key] = exports[key]; } } catch (_) {}; module.exports = exports.default; }

package/dist/cjs/chunk/version.js.map

@@ -1 +1 @@
-{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/version.ts"],"sourcesContent":["// D8: bump whenever chunk semantics change (grouping, sizing, splitting); the digest in\n// test/unit/chunk/version.test.ts tracks EXTRACTION changes only, so the two move independently.\nexport const CHUNK_VERSION = 'chunk:v5';\n"],"names":["CHUNK_VERSION"],"mappings":"AAAA,wFAAwF;AACxF,iGAAiG;;;;;+BACpFA;;;eAAAA;;;AAAN,IAAMA,gBAAgB"}
+{"version":3,"sources":["/Users/kevin/Dev/OpenSource/ai/sensemaking/src/chunk/version.ts"],"sourcesContent":["// D8: bump whenever chunk semantics change (grouping, sizing, splitting); the digest in\n// test/unit/chunk/version.test.ts tracks EXTRACTION changes only, so the two move independently.\nexport const CHUNK_VERSION = 'chunk:v6';\n"],"names":["CHUNK_VERSION"],"mappings":"AAAA,wFAAwF;AACxF,iGAAiG;;;;;+BACpFA;;;eAAAA;;;AAAN,IAAMA,gBAAgB"}

package/dist/cjs/features/links.js

@@ -508,45 +508,6 @@ function writeResolved(db, rows) {
         });
     })();
 }
-// A new or deleted file can change any note's resolution, so re-resolve the whole table.
-// Fallback for cold builds and large deltas -- see afterReconcile's threshold.
-function resolveAll(db, files) {
-    return _async_to_generator(function() {
-        var pathSet, byBase, stmt, rows, changed;
-        return _ts_generator(this, function(_state) {
-            switch(_state.label){
-                case 0:
-                    pathSet = new Set(files.map(function(f) {
-                        return f.relPath;
-                    }));
-                    byBase = buildByBase(files);
-                    return [
-                        4,
-                        db.prepare('SELECT src, target, dst, embed FROM links')
-                    ];
-                case 1:
-                    stmt = _state.sent();
-                    return [
-                        4,
-                        stmt.all()
-                    ];
-                case 2:
-                    rows = _state.sent();
-                    changed = resolveRows(rows, pathSet, byBase);
-                    return [
-                        4,
-                        writeResolved(db, changed)
-                    ];
-                case 3:
-                    _state.sent();
-                    return [
-                        2,
-                        changed.length > 0
-                    ];
-            }
-        });
-    })();
-}
 // Chunked so a large-but-under-threshold delta never exceeds SQLite's bound-variable limit
 // (SQLITE_MAX_VARIABLE_NUMBER, 32766); harmless headroom on DuckDB, which has no such cap.
 function selectIn(db, column, keys) {
@@ -757,8 +718,11 @@ function linkEdges(db) {
 // Records whether remove()/store() deleted any link row, since afterReconcile runs after both
 // and dstChanged alone misses the reparsed-to-zero-links case. Keyed on delta so state dies with it.
 var removedWithLinks = new WeakMap();
-// dst is written by resolution, not by store(), so it is absent from the column list and takes the
-// table's default on the append path.
+// Cold build only: whether store() resolved at least one row to a non-null dst, so afterReconcile
+// can report linksChanged without re-reading the table it just wrote.
+var coldResolved = new WeakMap();
+// dst stays NULL here on an incremental store (resolution runs afterward); a cold build resolves
+// it inline in store() instead, through the *_RESOLVED column list and SQL below.
 var LINK_COLUMNS = [
     'src',
     'target',
@@ -766,6 +730,14 @@ var LINK_COLUMNS = [
     'embed'
 ];
 var UPSERT_LINK_SQL = 'INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, NULL, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base';
+var LINK_COLUMNS_RESOLVED = [
+    'src',
+    'target',
+    'target_base',
+    'dst',
+    'embed'
+];
+var UPSERT_LINK_SQL_RESOLVED = 'INSERT INTO links (src, target, target_base, dst, embed) VALUES (?, ?, ?, ?, ?) ON CONFLICT(src, target, embed) DO UPDATE SET target_base = excluded.target_base';
 var links = {
     name: 'links',
     schema: function schema(db) {
@@ -840,7 +812,7 @@ var links = {
     },
     store: function store(db, docs, delta) {
         return _async_to_generator(function() {
-            var byPath, addedSet, reparsedExisting, existingRows, staleRows, appendable, upsertRows, _iteratorNormalCompletion, _didIteratorError, _iteratorError, _iterator, _step, _step_value, path, targets, _iteratorNormalCompletion1, _didIteratorError1, _iteratorError1, _iterator1, _step1, _step_value1, target, embed, row;
+            var byPath, addedSet, reparsedExisting, existingRows, staleRows, cold, resolveNew, pathSet, byBase, resolvedLink, appendable, upsertRows, _iteratorNormalCompletion, _didIteratorError, _iteratorError, _iterator, _step, _step_value, path, targets, _iteratorNormalCompletion1, _didIteratorError1, _iteratorError1, _iterator1, _step1, _step_value1, target, embed, targetBase, dst;
             return _ts_generator(this, function(_state) {
                 switch(_state.label){
                     case 0:
@@ -888,8 +860,21 @@ var links = {
                         removedWithLinks.set(delta, true);
                         _state.label = 3;
                     case 3:
-                        // Upsert preserves dst on surviving rows; only new rows start at NULL. A same-note anchor
-                        // gets no target_base, so SQL's `IN` excludes it from resolveIncremental's lookup.
+                        // Cold build: every path is in `added`, so pathSet/byBase cover the same complete tree
+                        // afterReconcile's large branch would otherwise re-derive; resolve dst here instead.
+                        cold = delta.files.length === 0 || delta.added.length === delta.files.length;
+                        if (cold) {
+                            pathSet = new Set(delta.files.map(function(f) {
+                                return f.relPath;
+                            }));
+                            byBase = buildByBase(delta.files);
+                            resolveNew = function resolveNew(src, target) {
+                                return resolveTarget(src, target, pathSet, byBase);
+                            };
+                        }
+                        resolvedLink = false;
+                        // Upsert preserves dst on surviving rows; a cold build resolves new rows, else they start NULL.
+                        // A same-note anchor gets no target_base, so SQL's `IN` excludes it from resolveIncremental's lookup.
                         // Split, unlike the other features: this hook has no remove() pass behind it (see above), so
                         // only a path in `added` is guaranteed to have no row yet. The rest genuinely conflict and the
                         // upsert is what preserves their `dst`.
@@ -903,14 +888,34 @@ var links = {
                                 try {
                                     for(_iterator1 = targets[Symbol.iterator](); !(_iteratorNormalCompletion1 = (_step1 = _iterator1.next()).done); _iteratorNormalCompletion1 = true){
                                         _step_value1 = _step1.value, target = _step_value1.target, embed = _step_value1.embed;
-                                        row = [
+                                        targetBase = target.startsWith('#') ? null : baseKey(cleanTarget(target));
+                                        if (!addedSet.has(path)) {
+                                            upsertRows.push([
+                                                path,
+                                                target,
+                                                targetBase,
+                                                embed ? 1 : 0
+                                            ]);
+                                            continue;
+                                        }
+                                        if (!resolveNew) {
+                                            appendable.push([
+                                                path,
+                                                target,
+                                                targetBase,
+                                                embed ? 1 : 0
+                                            ]);
+                                            continue;
+                                        }
+                                        dst = resolveNew(path, target);
+                                        if (dst !== null) resolvedLink = true;
+                                        appendable.push([
                                             path,
                                             target,
-                                            target.startsWith('#') ? null : baseKey(cleanTarget(target)),
+                                            targetBase,
+                                            dst,
                                             embed ? 1 : 0
-                                        ];
-                                        if (addedSet.has(path)) appendable.push(row);
-                                        else upsertRows.push(row);
+                                        ]);
                                     }
                                 } catch (err) {
                                     _didIteratorError1 = true;
@@ -943,7 +948,7 @@ var links = {
                         }
                         return [
                             4,
-                            (0, _sharedts.appendRows)(db, 'links', LINK_COLUMNS, UPSERT_LINK_SQL, appendable)
+                            (0, _sharedts.appendRows)(db, 'links', resolveNew ? LINK_COLUMNS_RESOLVED : LINK_COLUMNS, resolveNew ? UPSERT_LINK_SQL_RESOLVED : UPSERT_LINK_SQL, appendable)
                         ];
                     case 4:
                         _state.sent();
@@ -959,6 +964,7 @@ var links = {
                         _state.sent();
                         _state.label = 6;
                     case 6:
+                        if (cold) coldResolved.set(delta, resolvedLink);
                         return [
                             2
                         ];
@@ -968,7 +974,7 @@ var links = {
     },
     afterReconcile: function afterReconcile(db, delta) {
         return _async_to_generator(function() {
-            var _removedWithLinks_get, removeHadLinks, large, dstChanged, _tmp;
+            var _removedWithLinks_get, removeHadLinks, large, _coldResolved_get, dstChanged;
             return _ts_generator(this, function(_state) {
                 switch(_state.label){
                     case 0:
@@ -978,30 +984,20 @@ var links = {
                         // A cold build is the only case a full pass wins. Measured 2026-09-01 on 6,566 notes, resolve
                         // time alone: full is flat near 90ms at any churn, incremental 7ms at 10 files and 30ms at 2,000.
                         large = delta.files.length === 0 || delta.added.length === delta.files.length;
-                        if (!large) return [
-                            3,
-                            2
-                        ];
-                        return [
-                            4,
-                            resolveAll(db, delta.files)
-                        ];
-                    case 1:
-                        _tmp = _state.sent();
-                        return [
-                            3,
-                            4
-                        ];
-                    case 2:
+                        // store() already resolved dst inline on this cold build (same test), so nothing is left to update.
+                        if (large) {
+                            ;
+                            delta.linksChanged = ((_coldResolved_get = coldResolved.get(delta)) !== null && _coldResolved_get !== void 0 ? _coldResolved_get : false) || removeHadLinks;
+                            return [
+                                2
+                            ];
+                        }
                         return [
                             4,
                             resolveIncremental(db, delta)
                         ];
-                    case 3:
-                        _tmp = _state.sent();
-                        _state.label = 4;
-                    case 4:
-                        dstChanged = _tmp;
+                    case 1:
+                        dstChanged = _state.sent();
                         delta.linksChanged = dstChanged || removeHadLinks;
                         return [
                             2