wiki-graph-core 0.3.1 → 0.4.0

package/dist/graph/evidence-selection/anchor/resolver/matcher.js

@@ -0,0 +1,178 @@
+import { normalizeText } from "../../text.js";
+import {
+  charNgramScore,
+  createEmptyTextMatchScore,
+  levenshteinSimilarity,
+  lengthPenalty,
+  sequenceSimilarity
+} from "./scoring.js";
+import { MAX_BOUNDARY_BONUS, MIN_BOUNDARY_BONUS_LENGTH } from "./types.js";
+function scoreAnchor(anchor, candidateText, label) {
+  if (anchor.mode === "head_tail") {
+    return scoreHeadTail(anchor, candidateText);
+  }
+  return scoreTextQuery(anchor.text ?? "", candidateText, label);
+}
+function scoreHeadTail(anchor, candidateText) {
+  const head = anchor.head ?? "";
+  const tail = anchor.tail ?? "";
+  const candidateNormalized = normalizeText(candidateText);
+  const headNormalized = normalizeText(head);
+  const tailNormalized = normalizeText(tail);
+  if (headNormalized === "" || tailNormalized === "") {
+    return createEmptyTextMatchScore();
+  }
+  const headPosition = candidateNormalized.indexOf(headNormalized);
+  const tailPosition = candidateNormalized.lastIndexOf(tailNormalized);
+  if (headPosition !== -1 && tailPosition !== -1 && headPosition <= tailPosition) {
+    const coveredChars = headNormalized.length + tailNormalized.length;
+    const coverage = Math.min(
+      1,
+      coveredChars / Math.max(1, candidateNormalized.length)
+    );
+    return {
+      exactNormalized: false,
+      exactRaw: false,
+      exactSubstring: true,
+      matchEnd: tailPosition + tailNormalized.length,
+      matchStart: headPosition,
+      score: 0.82 + 0.18 * coverage
+    };
+  }
+  const headMatch = scoreTextQuery(head, candidateText, "start_anchor");
+  const tailMatch = scoreTextQuery(tail, candidateText, "end_anchor");
+  const ordered = headMatch.matchStart <= tailMatch.matchStart ? 1 : 0.85;
+  return {
+    exactNormalized: false,
+    exactRaw: false,
+    exactSubstring: false,
+    matchEnd: Math.max(headMatch.matchEnd, tailMatch.matchEnd),
+    matchStart: Math.min(headMatch.matchStart, tailMatch.matchStart),
+    score: (headMatch.score + tailMatch.score) / 2 * ordered
+  };
+}
+function scoreTextQuery(queryText, candidateText, label) {
+  const queryRaw = queryText.trim();
+  const candidateRaw = candidateText.trim();
+  const queryNormalized = normalizeText(queryRaw);
+  const candidateNormalized = normalizeText(candidateRaw);
+  if (queryNormalized === "" || candidateNormalized === "") {
+    return createEmptyTextMatchScore();
+  }
+  if (queryRaw === candidateRaw) {
+    return {
+      exactNormalized: true,
+      exactRaw: true,
+      exactSubstring: true,
+      matchEnd: candidateNormalized.length,
+      matchStart: 0,
+      score: 1
+    };
+  }
+  if (queryNormalized === candidateNormalized) {
+    return {
+      exactNormalized: true,
+      exactRaw: false,
+      exactSubstring: false,
+      matchEnd: candidateNormalized.length,
+      matchStart: 0,
+      score: 0.995
+    };
+  }
+  if (candidateNormalized.includes(queryNormalized)) {
+    const matchStart = candidateNormalized.indexOf(queryNormalized);
+    const coverage = queryNormalized.length / Math.max(1, candidateNormalized.length);
+    const boundaryBonus = calculateBoundaryBonus({
+      candidateNormalized,
+      label,
+      matchEnd: matchStart + queryNormalized.length,
+      matchStart,
+      queryNormalized
+    });
+    return {
+      exactNormalized: false,
+      exactRaw: false,
+      exactSubstring: true,
+      matchEnd: matchStart + queryNormalized.length,
+      matchStart,
+      score: Math.min(1, 0.75 + 0.25 * coverage + boundaryBonus)
+    };
+  }
+  if (queryNormalized.includes(candidateNormalized)) {
+    const coverage = candidateNormalized.length / Math.max(1, queryNormalized.length);
+    return {
+      exactNormalized: false,
+      exactRaw: false,
+      exactSubstring: false,
+      matchEnd: candidateNormalized.length,
+      matchStart: 0,
+      score: Math.min(0.92, 0.78 + 0.14 * coverage)
+    };
+  }
+  let bestScore = 0;
+  let bestStart = -1;
+  let bestEnd = -1;
+  const maxWindow = Math.min(
+    candidateNormalized.length,
+    Math.max(queryNormalized.length, Math.floor(queryNormalized.length * 1.2))
+  );
+  const minWindow = Math.min(
+    maxWindow,
+    Math.max(1, Math.floor(queryNormalized.length * 0.8))
+  );
+  const queryChars = new Set(queryNormalized);
+  for (let windowSize = minWindow; windowSize <= maxWindow; windowSize += 1) {
+    for (let start = 0; start <= candidateNormalized.length - windowSize; start += 1) {
+      const window = candidateNormalized.slice(start, start + windowSize);
+      const windowChars = new Set(window);
+      const unionSize = (/* @__PURE__ */ new Set([...queryChars, ...windowChars])).size;
+      const overlapSize = [...queryChars].filter(
+        (char) => windowChars.has(char)
+      ).length;
+      const jaccard = unionSize === 0 ? 0 : overlapSize / unionSize;
+      if (jaccard < 0.35) {
+        continue;
+      }
+      const totalScore = 0.35 * charNgramScore(queryNormalized, window) + 0.35 * sequenceSimilarity(queryNormalized, window) + 0.2 * levenshteinSimilarity(queryNormalized, window) + 0.1 * lengthPenalty(queryNormalized, window) + calculateBoundaryBonus({
+        candidateNormalized,
+        label,
+        matchEnd: start + windowSize,
+        matchStart: start,
+        queryNormalized
+      });
+      if (totalScore > bestScore) {
+        bestScore = Math.min(1, totalScore);
+        bestStart = start;
+        bestEnd = start + windowSize;
+      }
+    }
+  }
+  return {
+    exactNormalized: false,
+    exactRaw: false,
+    exactSubstring: false,
+    matchEnd: bestEnd,
+    matchStart: bestStart,
+    score: bestScore
+  };
+}
+function calculateBoundaryBonus(input) {
+  if (input.matchStart < 0 || input.matchEnd < 0) {
+    return 0;
+  }
+  if (input.queryNormalized.length < MIN_BOUNDARY_BONUS_LENGTH) {
+    return 0;
+  }
+  if (input.label === "start_anchor" && input.matchStart !== 0) {
+    return 0;
+  }
+  if (input.label === "end_anchor" && input.matchEnd !== input.candidateNormalized.length) {
+    return 0;
+  }
+  const coverage = input.queryNormalized.length / Math.max(1, input.candidateNormalized.length);
+  return MAX_BOUNDARY_BONUS * (0.5 + 0.5 * Math.min(1, coverage));
+}
+export {
+  scoreAnchor
+};
+//# sourceMappingURL=matcher.js.map

package/dist/graph/evidence-selection/anchor/resolver/matcher.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../../../src/graph/evidence-selection/anchor/resolver/matcher.ts"],"sourcesContent":["import { normalizeText } from \"../../text.js\";\n\nimport {\n  charNgramScore,\n  createEmptyTextMatchScore,\n  levenshteinSimilarity,\n  lengthPenalty,\n  sequenceSimilarity,\n  type TextMatchScore,\n} from \"./scoring.js\";\nimport { MAX_BOUNDARY_BONUS, MIN_BOUNDARY_BONUS_LENGTH } from \"./types.js\";\nimport type { AnchorSpec } from \"./types.js\";\n\nexport function scoreAnchor(\n  anchor: AnchorSpec,\n  candidateText: string,\n  label: string,\n): TextMatchScore {\n  if (anchor.mode === \"head_tail\") {\n    return scoreHeadTail(anchor, candidateText);\n  }\n\n  return scoreTextQuery(anchor.text ?? \"\", candidateText, label);\n}\n\nfunction scoreHeadTail(\n  anchor: AnchorSpec,\n  candidateText: string,\n): TextMatchScore {\n  const head = anchor.head ?? \"\";\n  const tail = anchor.tail ?? \"\";\n  const candidateNormalized = normalizeText(candidateText);\n  const headNormalized = normalizeText(head);\n  const tailNormalized = normalizeText(tail);\n\n  if (headNormalized === \"\" || tailNormalized === \"\") {\n    return createEmptyTextMatchScore();\n  }\n\n  const headPosition = candidateNormalized.indexOf(headNormalized);\n  const tailPosition = candidateNormalized.lastIndexOf(tailNormalized);\n\n  if (\n    headPosition !== -1 &&\n    tailPosition !== -1 &&\n    headPosition <= tailPosition\n  ) {\n    const coveredChars = headNormalized.length + tailNormalized.length;\n    const coverage = Math.min(\n      1,\n      coveredChars / Math.max(1, candidateNormalized.length),\n    );\n\n    return {\n      exactNormalized: false,\n      exactRaw: false,\n      exactSubstring: true,\n      matchEnd: tailPosition + tailNormalized.length,\n      matchStart: headPosition,\n      score: 0.82 + 0.18 * coverage,\n    };\n  }\n\n  const headMatch = scoreTextQuery(head, candidateText, \"start_anchor\");\n  const tailMatch = scoreTextQuery(tail, candidateText, \"end_anchor\");\n  const ordered = headMatch.matchStart <= tailMatch.matchStart ? 1 : 0.85;\n\n  return {\n    exactNormalized: false,\n    exactRaw: false,\n    exactSubstring: false,\n    matchEnd: Math.max(headMatch.matchEnd, tailMatch.matchEnd),\n    matchStart: Math.min(headMatch.matchStart, tailMatch.matchStart),\n    score: ((headMatch.score + tailMatch.score) / 2) * ordered,\n  };\n}\n\nfunction scoreTextQuery(\n  queryText: string,\n  candidateText: string,\n  label: string,\n): TextMatchScore {\n  const queryRaw = queryText.trim();\n  const candidateRaw = candidateText.trim();\n  const queryNormalized = normalizeText(queryRaw);\n  const candidateNormalized = normalizeText(candidateRaw);\n\n  if (queryNormalized === \"\" || candidateNormalized === \"\") {\n    return createEmptyTextMatchScore();\n  }\n\n  if (queryRaw === candidateRaw) {\n    return {\n      exactNormalized: true,\n      exactRaw: true,\n      exactSubstring: true,\n      matchEnd: candidateNormalized.length,\n      matchStart: 0,\n      score: 1,\n    };\n  }\n\n  if (queryNormalized === candidateNormalized) {\n    return {\n      exactNormalized: true,\n      exactRaw: false,\n      exactSubstring: false,\n      matchEnd: candidateNormalized.length,\n      matchStart: 0,\n      score: 0.995,\n    };\n  }\n\n  if (candidateNormalized.includes(queryNormalized)) {\n    const matchStart = candidateNormalized.indexOf(queryNormalized);\n    const coverage =\n      queryNormalized.length / Math.max(1, candidateNormalized.length);\n    const boundaryBonus = calculateBoundaryBonus({\n      candidateNormalized,\n      label,\n      matchEnd: matchStart + queryNormalized.length,\n      matchStart,\n      queryNormalized,\n    });\n\n    return {\n      exactNormalized: false,\n      exactRaw: false,\n      exactSubstring: true,\n      matchEnd: matchStart + queryNormalized.length,\n      matchStart,\n      score: Math.min(1, 0.75 + 0.25 * coverage + boundaryBonus),\n    };\n  }\n\n  if (queryNormalized.includes(candidateNormalized)) {\n    const coverage =\n      candidateNormalized.length / Math.max(1, queryNormalized.length);\n\n    return {\n      exactNormalized: false,\n      exactRaw: false,\n      exactSubstring: false,\n      matchEnd: candidateNormalized.length,\n      matchStart: 0,\n      score: Math.min(0.92, 0.78 + 0.14 * coverage),\n    };\n  }\n\n  let bestScore = 0;\n  let bestStart = -1;\n  let bestEnd = -1;\n  const maxWindow = Math.min(\n    candidateNormalized.length,\n    Math.max(queryNormalized.length, Math.floor(queryNormalized.length * 1.2)),\n  );\n  const minWindow = Math.min(\n    maxWindow,\n    Math.max(1, Math.floor(queryNormalized.length * 0.8)),\n  );\n  const queryChars = new Set(queryNormalized);\n\n  for (let windowSize = minWindow; windowSize <= maxWindow; windowSize += 1) {\n    for (\n      let start = 0;\n      start <= candidateNormalized.length - windowSize;\n      start += 1\n    ) {\n      const window = candidateNormalized.slice(start, start + windowSize);\n      const windowChars = new Set(window);\n      const unionSize = new Set([...queryChars, ...windowChars]).size;\n      const overlapSize = [...queryChars].filter((char) =>\n        windowChars.has(char),\n      ).length;\n      const jaccard = unionSize === 0 ? 0 : overlapSize / unionSize;\n\n      if (jaccard < 0.35) {\n        continue;\n      }\n\n      const totalScore =\n        0.35 * charNgramScore(queryNormalized, window) +\n        0.35 * sequenceSimilarity(queryNormalized, window) +\n        0.2 * levenshteinSimilarity(queryNormalized, window) +\n        0.1 * lengthPenalty(queryNormalized, window) +\n        calculateBoundaryBonus({\n          candidateNormalized,\n          label,\n          matchEnd: start + windowSize,\n          matchStart: start,\n          queryNormalized,\n        });\n\n      if (totalScore > bestScore) {\n        bestScore = Math.min(1, totalScore);\n        bestStart = start;\n        bestEnd = start + windowSize;\n      }\n    }\n  }\n\n  return {\n    exactNormalized: false,\n    exactRaw: false,\n    exactSubstring: false,\n    matchEnd: bestEnd,\n    matchStart: bestStart,\n    score: bestScore,\n  };\n}\n\nfunction calculateBoundaryBonus(input: {\n  label: string;\n  queryNormalized: string;\n  candidateNormalized: string;\n  matchStart: number;\n  matchEnd: number;\n}): number {\n  if (input.matchStart < 0 || input.matchEnd < 0) {\n    return 0;\n  }\n\n  if (input.queryNormalized.length < MIN_BOUNDARY_BONUS_LENGTH) {\n    return 0;\n  }\n\n  if (input.label === \"start_anchor\" && input.matchStart !== 0) {\n    return 0;\n  }\n\n  if (\n    input.label === \"end_anchor\" &&\n    input.matchEnd !== input.candidateNormalized.length\n  ) {\n    return 0;\n  }\n\n  const coverage =\n    input.queryNormalized.length /\n    Math.max(1, input.candidateNormalized.length);\n\n  return MAX_BOUNDARY_BONUS * (0.5 + 0.5 * Math.min(1, coverage));\n}\n"],"mappings":"AAAA,SAAS,qBAAqB;AAE9B;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,OAEK;AACP,SAAS,oBAAoB,iCAAiC;AAGvD,SAAS,YACd,QACA,eACA,OACgB;AAChB,MAAI,OAAO,SAAS,aAAa;AAC/B,WAAO,cAAc,QAAQ,aAAa;AAAA,EAC5C;AAEA,SAAO,eAAe,OAAO,QAAQ,IAAI,eAAe,KAAK;AAC/D;AAEA,SAAS,cACP,QACA,eACgB;AAChB,QAAM,OAAO,OAAO,QAAQ;AAC5B,QAAM,OAAO,OAAO,QAAQ;AAC5B,QAAM,sBAAsB,cAAc,aAAa;AACvD,QAAM,iBAAiB,cAAc,IAAI;AACzC,QAAM,iBAAiB,cAAc,IAAI;AAEzC,MAAI,mBAAmB,MAAM,mBAAmB,IAAI;AAClD,WAAO,0BAA0B;AAAA,EACnC;AAEA,QAAM,eAAe,oBAAoB,QAAQ,cAAc;AAC/D,QAAM,eAAe,oBAAoB,YAAY,cAAc;AAEnE,MACE,iBAAiB,MACjB,iBAAiB,MACjB,gBAAgB,cAChB;AACA,UAAM,eAAe,eAAe,SAAS,eAAe;AAC5D,UAAM,WAAW,KAAK;AAAA,MACpB;AAAA,MACA,eAAe,KAAK,IAAI,GAAG,oBAAoB,MAAM;AAAA,IACvD;AAEA,WAAO;AAAA,MACL,iBAAiB;AAAA,MACjB,UAAU;AAAA,MACV,gBAAgB;AAAA,MAChB,UAAU,eAAe,eAAe;AAAA,MACxC,YAAY;AAAA,MACZ,OAAO,OAAO,OAAO;AAAA,IACvB;AAAA,EACF;AAEA,QAAM,YAAY,eAAe,MAAM,eAAe,cAAc;AACpE,QAAM,YAAY,eAAe,MAAM,eAAe,YAAY;AAClE,QAAM,UAAU,UAAU,cAAc,UAAU,aAAa,IAAI;AAEnE,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB,UAAU;AAAA,IACV,gBAAgB;AAAA,IAChB,UAAU,KAAK,IAAI,UAAU,UAAU,UAAU,QAAQ;AAAA,IACzD,YAAY,KAAK,IAAI,UAAU,YAAY,UAAU,UAAU;AAAA,IAC/D,QAAS,UAAU,QAAQ,UAAU,SAAS,IAAK;AAAA,EACrD;AACF;AAEA,SAAS,eACP,WACA,eACA,OACgB;AAChB,QAAM,WAAW,UAAU,KAAK;AAChC,QAAM,eAAe,cAAc,KAAK;AACxC,QAAM,kBAAkB,cAAc,QAAQ;AAC9C,QAAM,sBAAsB,cAAc,YAAY;AAEtD,MAAI,oBAAoB,MAAM,wBAAwB,IAAI;AACxD,WAAO,0BAA0B;AAAA,EACnC;AAEA,MAAI,aAAa,cAAc;AAC7B,WAAO;AAAA,MACL,iBAAiB;AAAA,MACjB,UAAU;AAAA,MACV,gBAAgB;AAAA,MAChB,UAAU,oBAAoB;AAAA,MAC9B,YAAY;AAAA,MACZ,OAAO;AAAA,IACT;AAAA,EACF;AAEA,MAAI,oBAAoB,qBAAqB;AAC3C,WAAO;AAAA,MACL,iBAAiB;AAAA,MACjB,UAAU;AAAA,MACV,gBAAgB;AAAA,MAChB,UAAU,oBAAoB;AAAA,MAC9B,YAAY;AAAA,MACZ,OAAO;AAAA,IACT;AAAA,EACF;AAEA,MAAI,oBAAoB,SAAS,eAAe,GAAG;AACjD,UAAM,aAAa,oBAAoB,QAAQ,eAAe;AAC9D,UAAM,WACJ,gBAAgB,SAAS,KAAK,IAAI,GAAG,oBAAoB,MAAM;AACjE,UAAM,gBAAgB,uBAAuB;AAAA,MAC3C;AAAA,MACA;AAAA,MACA,UAAU,aAAa,gBAAgB;AAAA,MACvC;AAAA,MACA;AAAA,IACF,CAAC;AAED,WAAO;AAAA,MACL,iBAAiB;AAAA,MACjB,UAAU;AAAA,MACV,gBAAgB;AAAA,MAChB,UAAU,aAAa,gBAAgB;AAAA,MACvC;AAAA,MACA,OAAO,KAAK,IAAI,GAAG,OAAO,OAAO,WAAW,aAAa;AAAA,IAC3D;AAAA,EACF;AAEA,MAAI,gBAAgB,SAAS,mBAAmB,GAAG;AACjD,UAAM,WACJ,oBAAoB,SAAS,KAAK,IAAI,GAAG,gBAAgB,MAAM;AAEjE,WAAO;AAAA,MACL,iBAAiB;AAAA,MACjB,UAAU;AAAA,MACV,gBAAgB;AAAA,MAChB,UAAU,oBAAoB;AAAA,MAC9B,YAAY;AAAA,MACZ,OAAO,KAAK,IAAI,MAAM,OAAO,OAAO,QAAQ;AAAA,IAC9C;AAAA,EACF;AAEA,MAAI,YAAY;AAChB,MAAI,YAAY;AAChB,MAAI,UAAU;AACd,QAAM,YAAY,KAAK;AAAA,IACrB,oBAAoB;AAAA,IACpB,KAAK,IAAI,gBAAgB,QAAQ,KAAK,MAAM,gBAAgB,SAAS,GAAG,CAAC;AAAA,EAC3E;AACA,QAAM,YAAY,KAAK;AAAA,IACrB;AAAA,IACA,KAAK,IAAI,GAAG,KAAK,MAAM,gBAAgB,SAAS,GAAG,CAAC;AAAA,EACtD;AACA,QAAM,aAAa,IAAI,IAAI,eAAe;AAE1C,WAAS,aAAa,WAAW,cAAc,WAAW,cAAc,GAAG;AACzE,aACM,QAAQ,GACZ,SAAS,oBAAoB,SAAS,YACtC,SAAS,GACT;AACA,YAAM,SAAS,oBAAoB,MAAM,OAAO,QAAQ,UAAU;AAClE,YAAM,cAAc,IAAI,IAAI,MAAM;AAClC,YAAM,aAAY,oBAAI,IAAI,CAAC,GAAG,YAAY,GAAG,WAAW,CAAC,GAAE;AAC3D,YAAM,cAAc,CAAC,GAAG,UAAU,EAAE;AAAA,QAAO,CAAC,SAC1C,YAAY,IAAI,IAAI;AAAA,MACtB,EAAE;AACF,YAAM,UAAU,cAAc,IAAI,IAAI,cAAc;AAEpD,UAAI,UAAU,MAAM;AAClB;AAAA,MACF;AAEA,YAAM,aACJ,OAAO,eAAe,iBAAiB,MAAM,IAC7C,OAAO,mBAAmB,iBAAiB,MAAM,IACjD,MAAM,sBAAsB,iBAAiB,MAAM,IACnD,MAAM,cAAc,iBAAiB,MAAM,IAC3C,uBAAuB;AAAA,QACrB;AAAA,QACA;AAAA,QACA,UAAU,QAAQ;AAAA,QAClB,YAAY;AAAA,QACZ;AAAA,MACF,CAAC;AAEH,UAAI,aAAa,WAAW;AAC1B,oBAAY,KAAK,IAAI,GAAG,UAAU;AAClC,oBAAY;AACZ,kBAAU,QAAQ;AAAA,MACpB;AAAA,IACF;AAAA,EACF;AAEA,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB,UAAU;AAAA,IACV,gBAAgB;AAAA,IAChB,UAAU;AAAA,IACV,YAAY;AAAA,IACZ,OAAO;AAAA,EACT;AACF;AAEA,SAAS,uBAAuB,OAMrB;AACT,MAAI,MAAM,aAAa,KAAK,MAAM,WAAW,GAAG;AAC9C,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,gBAAgB,SAAS,2BAA2B;AAC5D,WAAO;AAAA,EACT;AAEA,MAAI,MAAM,UAAU,kBAAkB,MAAM,eAAe,GAAG;AAC5D,WAAO;AAAA,EACT;AAEA,MACE,MAAM,UAAU,gBAChB,MAAM,aAAa,MAAM,oBAAoB,QAC7C;AACA,WAAO;AAAA,EACT;AAEA,QAAM,WACJ,MAAM,gBAAgB,SACtB,KAAK,IAAI,GAAG,MAAM,oBAAoB,MAAM;AAE9C,SAAO,sBAAsB,MAAM,MAAM,KAAK,IAAI,GAAG,QAAQ;AAC/D;","names":[]}

package/dist/graph/evidence-selection/anchor/resolver/parser.d.ts

@@ -0,0 +1,7 @@
+import { AnchorSpec } from './types.js';
+
+declare function parseRawAnchor(value: unknown): readonly [anchor: AnchorSpec | undefined, error: string | undefined];
+declare function normalizeAnchor(anchor: AnchorSpec | undefined, fieldName: string): AnchorSpec | undefined;
+declare function anchorLength(anchor: AnchorSpec): number;
+
+export { anchorLength, normalizeAnchor, parseRawAnchor };

package/dist/graph/evidence-selection/anchor/resolver/parser.js

@@ -0,0 +1,91 @@
+import { normalizeText, splitTextIntoSentences } from "../../text.js";
+function parseRawAnchor(value) {
+  if (value === void 0 || value === null) {
+    return [void 0, "anchor is required"];
+  }
+  if (typeof value === "string") {
+    const stripped = value.trim();
+    if (stripped === "") {
+      return [void 0, "anchor string is empty"];
+    }
+    if (stripped.includes("...")) {
+      const parts = stripped.split("...").map((part) => part.trim()).filter((part) => part !== "");
+      if (parts.length >= 2) {
+        const head2 = parts[0];
+        const tail2 = parts[parts.length - 1];
+        if (head2 === void 0 || tail2 === void 0) {
+          return [void 0, "anchor string is empty"];
+        }
+        return [
+          {
+            head: head2,
+            mode: "head_tail",
+            tail: tail2
+          },
+          void 0
+        ];
+      }
+    }
+    return [{ mode: "full", text: stripped }, void 0];
+  }
+  if (typeof value !== "object") {
+    return [void 0, `expected string or object, got ${typeof value}`];
+  }
+  const rawValue = value;
+  const mode = rawValue.mode === "head_tail" || rawValue.head !== void 0 ? "head_tail" : "full";
+  if (mode === "full") {
+    const text = typeof rawValue.text === "string" ? rawValue.text.trim() : "";
+    if (text === "") {
+      return [void 0, "full anchor requires non-empty 'text'"];
+    }
+    return [{ mode: "full", text }, void 0];
+  }
+  const head = typeof rawValue.head === "string" ? rawValue.head.trim() : "";
+  const tail = typeof rawValue.tail === "string" ? rawValue.tail.trim() : "";
+  if (head !== "" && tail === "") {
+    return [{ mode: "full", text: head }, void 0];
+  }
+  if (head === "" && tail !== "") {
+    return [{ mode: "full", text: tail }, void 0];
+  }
+  if (head === "" || tail === "") {
+    return [void 0, "head_tail anchor requires non-empty 'head' and 'tail'"];
+  }
+  return [{ head, mode: "head_tail", tail }, void 0];
+}
+function normalizeAnchor(anchor, fieldName) {
+  if (anchor === void 0) {
+    return void 0;
+  }
+  if (anchor.mode === "head_tail") {
+    const boundaryText = fieldName === "end_anchor" ? selectBoundarySentence(anchor.tail ?? "", "end_anchor") : selectBoundarySentence(anchor.head ?? "", "start_anchor");
+    return boundaryText === "" ? void 0 : {
+      mode: "full",
+      text: boundaryText
+    };
+  }
+  const normalizedText = selectBoundarySentence(anchor.text ?? "", fieldName);
+  return {
+    mode: "full",
+    text: normalizedText
+  };
+}
+function anchorLength(anchor) {
+  if (anchor.mode === "head_tail") {
+    return normalizeText(anchor.head ?? "").length + normalizeText(anchor.tail ?? "").length;
+  }
+  return normalizeText(anchor.text ?? "").length;
+}
+function selectBoundarySentence(text, fieldName) {
+  const sentences = splitTextIntoSentences(text);
+  if (sentences.length <= 1) {
+    return text.trim();
+  }
+  return fieldName === "end_anchor" ? sentences[sentences.length - 1] ?? text.trim() : sentences[0] ?? text.trim();
+}
+export {
+  anchorLength,
+  normalizeAnchor,
+  parseRawAnchor
+};
+//# sourceMappingURL=parser.js.map

package/dist/graph/evidence-selection/anchor/resolver/parser.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../../../src/graph/evidence-selection/anchor/resolver/parser.ts"],"sourcesContent":["import { normalizeText, splitTextIntoSentences } from \"../../text.js\";\n\nimport type { AnchorSpec } from \"./types.js\";\n\nexport function parseRawAnchor(\n  value: unknown,\n): readonly [anchor: AnchorSpec | undefined, error: string | undefined] {\n  if (value === undefined || value === null) {\n    return [undefined, \"anchor is required\"];\n  }\n\n  if (typeof value === \"string\") {\n    const stripped = value.trim();\n\n    if (stripped === \"\") {\n      return [undefined, \"anchor string is empty\"];\n    }\n\n    if (stripped.includes(\"...\")) {\n      const parts = stripped\n        .split(\"...\")\n        .map((part) => part.trim())\n        .filter((part) => part !== \"\");\n\n      if (parts.length >= 2) {\n        const head = parts[0];\n        const tail = parts[parts.length - 1];\n\n        if (head === undefined || tail === undefined) {\n          return [undefined, \"anchor string is empty\"];\n        }\n\n        return [\n          {\n            head,\n            mode: \"head_tail\",\n            tail,\n          },\n          undefined,\n        ];\n      }\n    }\n\n    return [{ mode: \"full\", text: stripped }, undefined];\n  }\n\n  if (typeof value !== \"object\") {\n    return [undefined, `expected string or object, got ${typeof value}`];\n  }\n\n  const rawValue = value as Record<string, unknown>;\n  const mode =\n    rawValue.mode === \"head_tail\" || rawValue.head !== undefined\n      ? \"head_tail\"\n      : \"full\";\n\n  if (mode === \"full\") {\n    const text = typeof rawValue.text === \"string\" ? rawValue.text.trim() : \"\";\n\n    if (text === \"\") {\n      return [undefined, \"full anchor requires non-empty 'text'\"];\n    }\n\n    return [{ mode: \"full\", text }, undefined];\n  }\n\n  const head = typeof rawValue.head === \"string\" ? rawValue.head.trim() : \"\";\n  const tail = typeof rawValue.tail === \"string\" ? rawValue.tail.trim() : \"\";\n\n  if (head !== \"\" && tail === \"\") {\n    return [{ mode: \"full\", text: head }, undefined];\n  }\n\n  if (head === \"\" && tail !== \"\") {\n    return [{ mode: \"full\", text: tail }, undefined];\n  }\n\n  if (head === \"\" || tail === \"\") {\n    return [undefined, \"head_tail anchor requires non-empty 'head' and 'tail'\"];\n  }\n\n  return [{ head, mode: \"head_tail\", tail }, undefined];\n}\n\nexport function normalizeAnchor(\n  anchor: AnchorSpec | undefined,\n  fieldName: string,\n): AnchorSpec | undefined {\n  if (anchor === undefined) {\n    return undefined;\n  }\n\n  if (anchor.mode === \"head_tail\") {\n    const boundaryText =\n      fieldName === \"end_anchor\"\n        ? selectBoundarySentence(anchor.tail ?? \"\", \"end_anchor\")\n        : selectBoundarySentence(anchor.head ?? \"\", \"start_anchor\");\n\n    return boundaryText === \"\"\n      ? undefined\n      : {\n          mode: \"full\",\n          text: boundaryText,\n        };\n  }\n\n  const normalizedText = selectBoundarySentence(anchor.text ?? \"\", fieldName);\n\n  return {\n    mode: \"full\",\n    text: normalizedText,\n  };\n}\n\nexport function anchorLength(anchor: AnchorSpec): number {\n  if (anchor.mode === \"head_tail\") {\n    return (\n      normalizeText(anchor.head ?? \"\").length +\n      normalizeText(anchor.tail ?? \"\").length\n    );\n  }\n\n  return normalizeText(anchor.text ?? \"\").length;\n}\n\nfunction selectBoundarySentence(text: string, fieldName: string): string {\n  const sentences = splitTextIntoSentences(text);\n\n  if (sentences.length <= 1) {\n    return text.trim();\n  }\n\n  return fieldName === \"end_anchor\"\n    ? (sentences[sentences.length - 1] ?? text.trim())\n    : (sentences[0] ?? text.trim());\n}\n"],"mappings":"AAAA,SAAS,eAAe,8BAA8B;AAI/C,SAAS,eACd,OACsE;AACtE,MAAI,UAAU,UAAa,UAAU,MAAM;AACzC,WAAO,CAAC,QAAW,oBAAoB;AAAA,EACzC;AAEA,MAAI,OAAO,UAAU,UAAU;AAC7B,UAAM,WAAW,MAAM,KAAK;AAE5B,QAAI,aAAa,IAAI;AACnB,aAAO,CAAC,QAAW,wBAAwB;AAAA,IAC7C;AAEA,QAAI,SAAS,SAAS,KAAK,GAAG;AAC5B,YAAM,QAAQ,SACX,MAAM,KAAK,EACX,IAAI,CAAC,SAAS,KAAK,KAAK,CAAC,EACzB,OAAO,CAAC,SAAS,SAAS,EAAE;AAE/B,UAAI,MAAM,UAAU,GAAG;AACrB,cAAMA,QAAO,MAAM,CAAC;AACpB,cAAMC,QAAO,MAAM,MAAM,SAAS,CAAC;AAEnC,YAAID,UAAS,UAAaC,UAAS,QAAW;AAC5C,iBAAO,CAAC,QAAW,wBAAwB;AAAA,QAC7C;AAEA,eAAO;AAAA,UACL;AAAA,YACE,MAAAD;AAAA,YACA,MAAM;AAAA,YACN,MAAAC;AAAA,UACF;AAAA,UACA;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAEA,WAAO,CAAC,EAAE,MAAM,QAAQ,MAAM,SAAS,GAAG,MAAS;AAAA,EACrD;AAEA,MAAI,OAAO,UAAU,UAAU;AAC7B,WAAO,CAAC,QAAW,kCAAkC,OAAO,KAAK,EAAE;AAAA,EACrE;AAEA,QAAM,WAAW;AACjB,QAAM,OACJ,SAAS,SAAS,eAAe,SAAS,SAAS,SAC/C,cACA;AAEN,MAAI,SAAS,QAAQ;AACnB,UAAM,OAAO,OAAO,SAAS,SAAS,WAAW,SAAS,KAAK,KAAK,IAAI;AAExE,QAAI,SAAS,IAAI;AACf,aAAO,CAAC,QAAW,uCAAuC;AAAA,IAC5D;AAEA,WAAO,CAAC,EAAE,MAAM,QAAQ,KAAK,GAAG,MAAS;AAAA,EAC3C;AAEA,QAAM,OAAO,OAAO,SAAS,SAAS,WAAW,SAAS,KAAK,KAAK,IAAI;AACxE,QAAM,OAAO,OAAO,SAAS,SAAS,WAAW,SAAS,KAAK,KAAK,IAAI;AAExE,MAAI,SAAS,MAAM,SAAS,IAAI;AAC9B,WAAO,CAAC,EAAE,MAAM,QAAQ,MAAM,KAAK,GAAG,MAAS;AAAA,EACjD;AAEA,MAAI,SAAS,MAAM,SAAS,IAAI;AAC9B,WAAO,CAAC,EAAE,MAAM,QAAQ,MAAM,KAAK,GAAG,MAAS;AAAA,EACjD;AAEA,MAAI,SAAS,MAAM,SAAS,IAAI;AAC9B,WAAO,CAAC,QAAW,uDAAuD;AAAA,EAC5E;AAEA,SAAO,CAAC,EAAE,MAAM,MAAM,aAAa,KAAK,GAAG,MAAS;AACtD;AAEO,SAAS,gBACd,QACA,WACwB;AACxB,MAAI,WAAW,QAAW;AACxB,WAAO;AAAA,EACT;AAEA,MAAI,OAAO,SAAS,aAAa;AAC/B,UAAM,eACJ,cAAc,eACV,uBAAuB,OAAO,QAAQ,IAAI,YAAY,IACtD,uBAAuB,OAAO,QAAQ,IAAI,cAAc;AAE9D,WAAO,iBAAiB,KACpB,SACA;AAAA,MACE,MAAM;AAAA,MACN,MAAM;AAAA,IACR;AAAA,EACN;AAEA,QAAM,iBAAiB,uBAAuB,OAAO,QAAQ,IAAI,SAAS;AAE1E,SAAO;AAAA,IACL,MAAM;AAAA,IACN,MAAM;AAAA,EACR;AACF;AAEO,SAAS,aAAa,QAA4B;AACvD,MAAI,OAAO,SAAS,aAAa;AAC/B,WACE,cAAc,OAAO,QAAQ,EAAE,EAAE,SACjC,cAAc,OAAO,QAAQ,EAAE,EAAE;AAAA,EAErC;AAEA,SAAO,cAAc,OAAO,QAAQ,EAAE,EAAE;AAC1C;AAEA,SAAS,uBAAuB,MAAc,WAA2B;AACvE,QAAM,YAAY,uBAAuB,IAAI;AAE7C,MAAI,UAAU,UAAU,GAAG;AACzB,WAAO,KAAK,KAAK;AAAA,EACnB;AAEA,SAAO,cAAc,eAChB,UAAU,UAAU,SAAS,CAAC,KAAK,KAAK,KAAK,IAC7C,UAAU,CAAC,KAAK,KAAK,KAAK;AACjC;","names":["head","tail"]}

package/dist/graph/evidence-selection/anchor/resolver/scoring.d.ts

@@ -0,0 +1,20 @@
+import { RankedSentenceCandidate } from '../types.js';
+import '../../../../document/types.js';
+
+interface TextMatchScore {
+    readonly score: number;
+    readonly exactRaw: boolean;
+    readonly exactNormalized: boolean;
+    readonly exactSubstring: boolean;
+    readonly matchStart: number;
+    readonly matchEnd: number;
+}
+declare function createEmptyTextMatchScore(): TextMatchScore;
+declare function charNgramScore(left: string, right: string): number;
+declare function levenshteinSimilarity(left: string, right: string): number;
+declare function lengthPenalty(left: string, right: string): number;
+declare function sequenceSimilarity(left: string, right: string): number;
+declare function formatCandidate(candidate: RankedSentenceCandidate): string;
+declare function truncate(text: string, limit?: number): string;
+
+export { type TextMatchScore, charNgramScore, createEmptyTextMatchScore, formatCandidate, lengthPenalty, levenshteinSimilarity, sequenceSimilarity, truncate };

package/dist/graph/evidence-selection/anchor/resolver/scoring.js

@@ -0,0 +1,119 @@
+function createEmptyTextMatchScore() {
+  return {
+    exactNormalized: false,
+    exactRaw: false,
+    exactSubstring: false,
+    matchEnd: -1,
+    matchStart: -1,
+    score: 0
+  };
+}
+function charNgramScore(left, right) {
+  const leftBigrams = charNgrams(left, 2);
+  const rightBigrams = charNgrams(right, 2);
+  const leftTrigrams = charNgrams(left, 3);
+  const rightTrigrams = charNgrams(right, 3);
+  return (diceCoefficient(leftBigrams, rightBigrams) + diceCoefficient(leftTrigrams, rightTrigrams)) / 2;
+}
+function charNgrams(text, size) {
+  if (text.length <= size) {
+    return text === "" ? /* @__PURE__ */ new Set() : /* @__PURE__ */ new Set([text]);
+  }
+  const result = /* @__PURE__ */ new Set();
+  for (let index = 0; index <= text.length - size; index += 1) {
+    result.add(text.slice(index, index + size));
+  }
+  return result;
+}
+function diceCoefficient(left, right) {
+  if (left.size === 0 || right.size === 0) {
+    return 0;
+  }
+  let overlap = 0;
+  for (const value of left) {
+    if (right.has(value)) {
+      overlap += 1;
+    }
+  }
+  return 2 * overlap / (left.size + right.size);
+}
+function levenshteinSimilarity(left, right) {
+  if (left === right) {
+    return 1;
+  }
+  if (left === "" || right === "") {
+    return 0;
+  }
+  let previous = Array.from({ length: right.length + 1 }, (_, index) => index);
+  let current = new Array(right.length + 1).fill(0);
+  for (const [leftIndex, leftChar] of [...left].entries()) {
+    current[0] = leftIndex + 1;
+    for (const [rightIndex, rightChar] of [...right].entries()) {
+      const insertCost = (current[rightIndex] ?? 0) + 1;
+      const deleteCost = (previous[rightIndex + 1] ?? 0) + 1;
+      const replaceCost = (previous[rightIndex] ?? 0) + (leftChar === rightChar ? 0 : 1);
+      current[rightIndex + 1] = Math.min(insertCost, deleteCost, replaceCost);
+    }
+    [previous, current] = [current, previous];
+  }
+  const distance = previous[right.length] ?? Math.max(left.length, right.length);
+  return 1 - distance / Math.max(left.length, right.length);
+}
+function lengthPenalty(left, right) {
+  return 1 - Math.abs(left.length - right.length) / Math.max(left.length, right.length);
+}
+function sequenceSimilarity(left, right) {
+  if (left === right) {
+    return 1;
+  }
+  const leftLength = left.length;
+  const rightLength = right.length;
+  if (leftLength === 0 || rightLength === 0) {
+    return 0;
+  }
+  const matrix = Array.from(
+    { length: leftLength + 1 },
+    () => new Array(rightLength + 1).fill(0)
+  );
+  let longest = 0;
+  for (let leftIndex = 1; leftIndex <= leftLength; leftIndex += 1) {
+    for (let rightIndex = 1; rightIndex <= rightLength; rightIndex += 1) {
+      if (left[leftIndex - 1] !== right[rightIndex - 1]) {
+        continue;
+      }
+      const currentRow = matrix[leftIndex];
+      if (currentRow === void 0) {
+        continue;
+      }
+      currentRow[rightIndex] = (matrix[leftIndex - 1]?.[rightIndex - 1] ?? 0) + 1;
+      longest = Math.max(longest, currentRow[rightIndex] ?? 0);
+    }
+  }
+  return 2 * longest / (leftLength + rightLength);
+}
+function formatCandidate(candidate) {
+  return `  - ${candidate.occurrenceId} score=${candidate.score.toFixed(3)}
+    prev: ${truncate(candidate.prevText)}
+    text: ${truncate(candidate.text, 120)}
+    next: ${truncate(candidate.nextText)}`;
+}
+function truncate(text, limit = 80) {
+  const stripped = text.replace(/\s+/gu, " ").trim();
+  if (stripped === "") {
+    return "(none)";
+  }
+  if (stripped.length <= limit) {
+    return stripped;
+  }
+  return `${stripped.slice(0, limit - 3)}...`;
+}
+export {
+  charNgramScore,
+  createEmptyTextMatchScore,
+  formatCandidate,
+  lengthPenalty,
+  levenshteinSimilarity,
+  sequenceSimilarity,
+  truncate
+};
+//# sourceMappingURL=scoring.js.map

package/dist/graph/evidence-selection/anchor/resolver/scoring.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../../../src/graph/evidence-selection/anchor/resolver/scoring.ts"],"sourcesContent":["import type { RankedSentenceCandidate } from \"../types.js\";\n\nexport interface TextMatchScore {\n  readonly score: number;\n  readonly exactRaw: boolean;\n  readonly exactNormalized: boolean;\n  readonly exactSubstring: boolean;\n  readonly matchStart: number;\n  readonly matchEnd: number;\n}\n\nexport function createEmptyTextMatchScore(): TextMatchScore {\n  return {\n    exactNormalized: false,\n    exactRaw: false,\n    exactSubstring: false,\n    matchEnd: -1,\n    matchStart: -1,\n    score: 0,\n  };\n}\n\nexport function charNgramScore(left: string, right: string): number {\n  const leftBigrams = charNgrams(left, 2);\n  const rightBigrams = charNgrams(right, 2);\n  const leftTrigrams = charNgrams(left, 3);\n  const rightTrigrams = charNgrams(right, 3);\n\n  return (\n    (diceCoefficient(leftBigrams, rightBigrams) +\n      diceCoefficient(leftTrigrams, rightTrigrams)) /\n    2\n  );\n}\n\nfunction charNgrams(text: string, size: number): Set<string> {\n  if (text.length <= size) {\n    return text === \"\" ? new Set() : new Set([text]);\n  }\n\n  const result = new Set<string>();\n\n  for (let index = 0; index <= text.length - size; index += 1) {\n    result.add(text.slice(index, index + size));\n  }\n\n  return result;\n}\n\nfunction diceCoefficient(\n  left: ReadonlySet<string>,\n  right: ReadonlySet<string>,\n): number {\n  if (left.size === 0 || right.size === 0) {\n    return 0;\n  }\n\n  let overlap = 0;\n\n  for (const value of left) {\n    if (right.has(value)) {\n      overlap += 1;\n    }\n  }\n\n  return (2 * overlap) / (left.size + right.size);\n}\n\nexport function levenshteinSimilarity(left: string, right: string): number {\n  if (left === right) {\n    return 1;\n  }\n\n  if (left === \"\" || right === \"\") {\n    return 0;\n  }\n\n  let previous = Array.from({ length: right.length + 1 }, (_, index) => index);\n  let current = new Array<number>(right.length + 1).fill(0);\n\n  for (const [leftIndex, leftChar] of [...left].entries()) {\n    current[0] = leftIndex + 1;\n\n    for (const [rightIndex, rightChar] of [...right].entries()) {\n      const insertCost = (current[rightIndex] ?? 0) + 1;\n      const deleteCost = (previous[rightIndex + 1] ?? 0) + 1;\n      const replaceCost =\n        (previous[rightIndex] ?? 0) + (leftChar === rightChar ? 0 : 1);\n\n      current[rightIndex + 1] = Math.min(insertCost, deleteCost, replaceCost);\n    }\n\n    [previous, current] = [current, previous];\n  }\n\n  const distance =\n    previous[right.length] ?? Math.max(left.length, right.length);\n\n  return 1 - distance / Math.max(left.length, right.length);\n}\n\nexport function lengthPenalty(left: string, right: string): number {\n  return (\n    1 -\n    Math.abs(left.length - right.length) / Math.max(left.length, right.length)\n  );\n}\n\nexport function sequenceSimilarity(left: string, right: string): number {\n  if (left === right) {\n    return 1;\n  }\n\n  const leftLength = left.length;\n  const rightLength = right.length;\n\n  if (leftLength === 0 || rightLength === 0) {\n    return 0;\n  }\n\n  const matrix = Array.from({ length: leftLength + 1 }, () =>\n    new Array<number>(rightLength + 1).fill(0),\n  );\n  let longest = 0;\n\n  for (let leftIndex = 1; leftIndex <= leftLength; leftIndex += 1) {\n    for (let rightIndex = 1; rightIndex <= rightLength; rightIndex += 1) {\n      if (left[leftIndex - 1] !== right[rightIndex - 1]) {\n        continue;\n      }\n\n      const currentRow = matrix[leftIndex];\n\n      if (currentRow === undefined) {\n        continue;\n      }\n\n      currentRow[rightIndex] =\n        (matrix[leftIndex - 1]?.[rightIndex - 1] ?? 0) + 1;\n      longest = Math.max(longest, currentRow[rightIndex] ?? 0);\n    }\n  }\n\n  return (2 * longest) / (leftLength + rightLength);\n}\n\nexport function formatCandidate(candidate: RankedSentenceCandidate): string {\n  return (\n    `  - ${candidate.occurrenceId} score=${candidate.score.toFixed(3)}\\n` +\n    `    prev: ${truncate(candidate.prevText)}\\n` +\n    `    text: ${truncate(candidate.text, 120)}\\n` +\n    `    next: ${truncate(candidate.nextText)}`\n  );\n}\n\nexport function truncate(text: string, limit = 80): string {\n  const stripped = text.replace(/\\s+/gu, \" \").trim();\n\n  if (stripped === \"\") {\n    return \"(none)\";\n  }\n\n  if (stripped.length <= limit) {\n    return stripped;\n  }\n\n  return `${stripped.slice(0, limit - 3)}...`;\n}\n"],"mappings":"AAWO,SAAS,4BAA4C;AAC1D,SAAO;AAAA,IACL,iBAAiB;AAAA,IACjB,UAAU;AAAA,IACV,gBAAgB;AAAA,IAChB,UAAU;AAAA,IACV,YAAY;AAAA,IACZ,OAAO;AAAA,EACT;AACF;AAEO,SAAS,eAAe,MAAc,OAAuB;AAClE,QAAM,cAAc,WAAW,MAAM,CAAC;AACtC,QAAM,eAAe,WAAW,OAAO,CAAC;AACxC,QAAM,eAAe,WAAW,MAAM,CAAC;AACvC,QAAM,gBAAgB,WAAW,OAAO,CAAC;AAEzC,UACG,gBAAgB,aAAa,YAAY,IACxC,gBAAgB,cAAc,aAAa,KAC7C;AAEJ;AAEA,SAAS,WAAW,MAAc,MAA2B;AAC3D,MAAI,KAAK,UAAU,MAAM;AACvB,WAAO,SAAS,KAAK,oBAAI,IAAI,IAAI,oBAAI,IAAI,CAAC,IAAI,CAAC;AAAA,EACjD;AAEA,QAAM,SAAS,oBAAI,IAAY;AAE/B,WAAS,QAAQ,GAAG,SAAS,KAAK,SAAS,MAAM,SAAS,GAAG;AAC3D,WAAO,IAAI,KAAK,MAAM,OAAO,QAAQ,IAAI,CAAC;AAAA,EAC5C;AAEA,SAAO;AACT;AAEA,SAAS,gBACP,MACA,OACQ;AACR,MAAI,KAAK,SAAS,KAAK,MAAM,SAAS,GAAG;AACvC,WAAO;AAAA,EACT;AAEA,MAAI,UAAU;AAEd,aAAW,SAAS,MAAM;AACxB,QAAI,MAAM,IAAI,KAAK,GAAG;AACpB,iBAAW;AAAA,IACb;AAAA,EACF;AAEA,SAAQ,IAAI,WAAY,KAAK,OAAO,MAAM;AAC5C;AAEO,SAAS,sBAAsB,MAAc,OAAuB;AACzE,MAAI,SAAS,OAAO;AAClB,WAAO;AAAA,EACT;AAEA,MAAI,SAAS,MAAM,UAAU,IAAI;AAC/B,WAAO;AAAA,EACT;AAEA,MAAI,WAAW,MAAM,KAAK,EAAE,QAAQ,MAAM,SAAS,EAAE,GAAG,CAAC,GAAG,UAAU,KAAK;AAC3E,MAAI,UAAU,IAAI,MAAc,MAAM,SAAS,CAAC,EAAE,KAAK,CAAC;AAExD,aAAW,CAAC,WAAW,QAAQ,KAAK,CAAC,GAAG,IAAI,EAAE,QAAQ,GAAG;AACvD,YAAQ,CAAC,IAAI,YAAY;AAEzB,eAAW,CAAC,YAAY,SAAS,KAAK,CAAC,GAAG,KAAK,EAAE,QAAQ,GAAG;AAC1D,YAAM,cAAc,QAAQ,UAAU,KAAK,KAAK;AAChD,YAAM,cAAc,SAAS,aAAa,CAAC,KAAK,KAAK;AACrD,YAAM,eACH,SAAS,UAAU,KAAK,MAAM,aAAa,YAAY,IAAI;AAE9D,cAAQ,aAAa,CAAC,IAAI,KAAK,IAAI,YAAY,YAAY,WAAW;AAAA,IACxE;AAEA,KAAC,UAAU,OAAO,IAAI,CAAC,SAAS,QAAQ;AAAA,EAC1C;AAEA,QAAM,WACJ,SAAS,MAAM,MAAM,KAAK,KAAK,IAAI,KAAK,QAAQ,MAAM,MAAM;AAE9D,SAAO,IAAI,WAAW,KAAK,IAAI,KAAK,QAAQ,MAAM,MAAM;AAC1D;AAEO,SAAS,cAAc,MAAc,OAAuB;AACjE,SACE,IACA,KAAK,IAAI,KAAK,SAAS,MAAM,MAAM,IAAI,KAAK,IAAI,KAAK,QAAQ,MAAM,MAAM;AAE7E;AAEO,SAAS,mBAAmB,MAAc,OAAuB;AACtE,MAAI,SAAS,OAAO;AAClB,WAAO;AAAA,EACT;AAEA,QAAM,aAAa,KAAK;AACxB,QAAM,cAAc,MAAM;AAE1B,MAAI,eAAe,KAAK,gBAAgB,GAAG;AACzC,WAAO;AAAA,EACT;AAEA,QAAM,SAAS,MAAM;AAAA,IAAK,EAAE,QAAQ,aAAa,EAAE;AAAA,IAAG,MACpD,IAAI,MAAc,cAAc,CAAC,EAAE,KAAK,CAAC;AAAA,EAC3C;AACA,MAAI,UAAU;AAEd,WAAS,YAAY,GAAG,aAAa,YAAY,aAAa,GAAG;AAC/D,aAAS,aAAa,GAAG,cAAc,aAAa,cAAc,GAAG;AACnE,UAAI,KAAK,YAAY,CAAC,MAAM,MAAM,aAAa,CAAC,GAAG;AACjD;AAAA,MACF;AAEA,YAAM,aAAa,OAAO,SAAS;AAEnC,UAAI,eAAe,QAAW;AAC5B;AAAA,MACF;AAEA,iBAAW,UAAU,KAClB,OAAO,YAAY,CAAC,IAAI,aAAa,CAAC,KAAK,KAAK;AACnD,gBAAU,KAAK,IAAI,SAAS,WAAW,UAAU,KAAK,CAAC;AAAA,IACzD;AAAA,EACF;AAEA,SAAQ,IAAI,WAAY,aAAa;AACvC;AAEO,SAAS,gBAAgB,WAA4C;AAC1E,SACE,OAAO,UAAU,YAAY,UAAU,UAAU,MAAM,QAAQ,CAAC,CAAC;AAAA,YACpD,SAAS,UAAU,QAAQ,CAAC;AAAA,YAC5B,SAAS,UAAU,MAAM,GAAG,CAAC;AAAA,YAC7B,SAAS,UAAU,QAAQ,CAAC;AAE7C;AAEO,SAAS,SAAS,MAAc,QAAQ,IAAY;AACzD,QAAM,WAAW,KAAK,QAAQ,SAAS,GAAG,EAAE,KAAK;AAEjD,MAAI,aAAa,IAAI;AACnB,WAAO;AAAA,EACT;AAEA,MAAI,SAAS,UAAU,OAAO;AAC5B,WAAO;AAAA,EACT;AAEA,SAAO,GAAG,SAAS,MAAM,GAAG,QAAQ,CAAC,CAAC;AACxC;","names":[]}

package/dist/graph/evidence-selection/anchor/resolver/types.d.ts

@@ -0,0 +1,15 @@
+declare const MIN_AUTO_RESOLVE_GAP = 0.07;
+declare const MIN_AUTO_RESOLVE_SCORE = 0.9;
+declare const MIN_BOUNDARY_BONUS_LENGTH = 12;
+declare const MIN_CANDIDATE_SCORE = 0.55;
+declare const VERY_HIGH_CONFIDENCE_SCORE = 0.97;
+declare const MAX_BOUNDARY_BONUS = 0.08;
+declare const MAX_CANDIDATE_DISPLAY = 3;
+interface AnchorSpec {
+    readonly mode: "full" | "head_tail";
+    readonly text?: string;
+    readonly head?: string;
+    readonly tail?: string;
+}
+
+export { type AnchorSpec, MAX_BOUNDARY_BONUS, MAX_CANDIDATE_DISPLAY, MIN_AUTO_RESOLVE_GAP, MIN_AUTO_RESOLVE_SCORE, MIN_BOUNDARY_BONUS_LENGTH, MIN_CANDIDATE_SCORE, VERY_HIGH_CONFIDENCE_SCORE };

package/dist/graph/evidence-selection/anchor/resolver/types.js

@@ -0,0 +1,17 @@
+const MIN_AUTO_RESOLVE_GAP = 0.07;
+const MIN_AUTO_RESOLVE_SCORE = 0.9;
+const MIN_BOUNDARY_BONUS_LENGTH = 12;
+const MIN_CANDIDATE_SCORE = 0.55;
+const VERY_HIGH_CONFIDENCE_SCORE = 0.97;
+const MAX_BOUNDARY_BONUS = 0.08;
+const MAX_CANDIDATE_DISPLAY = 3;
+export {
+  MAX_BOUNDARY_BONUS,
+  MAX_CANDIDATE_DISPLAY,
+  MIN_AUTO_RESOLVE_GAP,
+  MIN_AUTO_RESOLVE_SCORE,
+  MIN_BOUNDARY_BONUS_LENGTH,
+  MIN_CANDIDATE_SCORE,
+  VERY_HIGH_CONFIDENCE_SCORE
+};
+//# sourceMappingURL=types.js.map

package/dist/graph/evidence-selection/anchor/resolver/types.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../../../src/graph/evidence-selection/anchor/resolver/types.ts"],"sourcesContent":["export const MIN_AUTO_RESOLVE_GAP = 0.07;\nexport const MIN_AUTO_RESOLVE_SCORE = 0.9;\nexport const MIN_BOUNDARY_BONUS_LENGTH = 12;\nexport const MIN_CANDIDATE_SCORE = 0.55;\nexport const VERY_HIGH_CONFIDENCE_SCORE = 0.97;\nexport const MAX_BOUNDARY_BONUS = 0.08;\nexport const MAX_CANDIDATE_DISPLAY = 3;\n\nexport interface AnchorSpec {\n  readonly mode: \"full\" | \"head_tail\";\n  readonly text?: string;\n  readonly head?: string;\n  readonly tail?: string;\n}\n"],"mappings":"AAAO,MAAM,uBAAuB;AAC7B,MAAM,yBAAyB;AAC/B,MAAM,4BAA4B;AAClC,MAAM,sBAAsB;AAC5B,MAAM,6BAA6B;AACnC,MAAM,qBAAqB;AAC3B,MAAM,wBAAwB;","names":[]}

package/dist/graph/evidence-selection/anchor/types.d.ts

@@ -0,0 +1,27 @@
+import { SentenceId } from '../../../document/types.js';
+
+interface RankedSentenceCandidate {
+    readonly occurrenceId: string;
+    readonly sentenceId: SentenceId;
+    readonly index: number;
+    readonly text: string;
+    readonly prevText: string;
+    readonly nextText: string;
+    readonly score: number;
+    readonly exactRaw: boolean;
+    readonly exactNormalized: boolean;
+    readonly exactSubstring: boolean;
+}
+interface EvidenceResolutionResult {
+    readonly sentenceIds: SentenceId[];
+    readonly strategy: string;
+    readonly confidence: number;
+}
+interface EvidenceResolutionFailure {
+    readonly fieldName: string;
+    readonly code: string;
+    readonly message: string;
+    readonly candidates: readonly RankedSentenceCandidate[];
+}
+
+export type { EvidenceResolutionFailure, EvidenceResolutionResult, RankedSentenceCandidate };

package/dist/graph/evidence-selection/index.d.ts

@@ -0,0 +1,9 @@
+export { EvidenceResolver } from './anchor/resolver/index.js';
+export { EvidenceResolutionFailure, EvidenceResolutionResult, RankedSentenceCandidate } from './anchor/types.js';
+export { EVIDENCE_SELECTION_JSON_SHAPE, EVIDENCE_SELECTION_PROMPT_FRAGMENT, formatEvidenceSelectionChoicePrompt } from './prompt.js';
+export { EvidenceQuoteMatchStrategy, EvidenceQuoteScore, normalizeEvidenceDisplayText, normalizeEvidenceText, scoreEvidenceQuote } from './quote-score.js';
+export { rankEvidenceQuote, resolveEvidenceSelection, resolveEvidenceSelectionList } from './resolver.js';
+export { normalizeText, splitTextIntoSentences } from './text.js';
+export { EvidenceSelection, EvidenceSelectionCandidate, EvidenceSelectionFailure, EvidenceSelectionList, EvidenceSelectionResolution, EvidenceSelectionSentence, EvidenceSentenceId } from './types.js';
+import '../../document/types.js';
+import './anchor/resolver/types.js';

package/dist/graph/evidence-selection/index.js

@@ -0,0 +1,32 @@
+import { EvidenceResolver } from "./anchor/index.js";
+import {
+  EVIDENCE_SELECTION_JSON_SHAPE,
+  EVIDENCE_SELECTION_PROMPT_FRAGMENT,
+  formatEvidenceSelectionChoicePrompt
+} from "./prompt.js";
+import {
+  normalizeEvidenceDisplayText,
+  normalizeEvidenceText,
+  scoreEvidenceQuote
+} from "./quote-score.js";
+import {
+  rankEvidenceQuote,
+  resolveEvidenceSelection,
+  resolveEvidenceSelectionList
+} from "./resolver.js";
+import { normalizeText, splitTextIntoSentences } from "./text.js";
+export {
+  EVIDENCE_SELECTION_JSON_SHAPE,
+  EVIDENCE_SELECTION_PROMPT_FRAGMENT,
+  EvidenceResolver,
+  formatEvidenceSelectionChoicePrompt,
+  normalizeEvidenceDisplayText,
+  normalizeEvidenceText,
+  normalizeText,
+  rankEvidenceQuote,
+  resolveEvidenceSelection,
+  resolveEvidenceSelectionList,
+  scoreEvidenceQuote,
+  splitTextIntoSentences
+};
+//# sourceMappingURL=index.js.map

package/dist/graph/evidence-selection/index.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../src/graph/evidence-selection/index.ts"],"sourcesContent":["export { EvidenceResolver } from \"./anchor/index.js\";\nexport type {\n  EvidenceResolutionFailure,\n  EvidenceResolutionResult,\n  RankedSentenceCandidate,\n} from \"./anchor/types.js\";\nexport {\n  EVIDENCE_SELECTION_JSON_SHAPE,\n  EVIDENCE_SELECTION_PROMPT_FRAGMENT,\n  formatEvidenceSelectionChoicePrompt,\n} from \"./prompt.js\";\nexport {\n  normalizeEvidenceDisplayText,\n  normalizeEvidenceText,\n  scoreEvidenceQuote,\n  type EvidenceQuoteMatchStrategy,\n  type EvidenceQuoteScore,\n} from \"./quote-score.js\";\nexport {\n  rankEvidenceQuote,\n  resolveEvidenceSelection,\n  resolveEvidenceSelectionList,\n} from \"./resolver.js\";\nexport { normalizeText, splitTextIntoSentences } from \"./text.js\";\nexport type {\n  EvidenceSelection,\n  EvidenceSelectionCandidate,\n  EvidenceSelectionFailure,\n  EvidenceSelectionList,\n  EvidenceSelectionResolution,\n  EvidenceSelectionSentence,\n  EvidenceSentenceId,\n} from \"./types.js\";\n"],"mappings":"AAAA,SAAS,wBAAwB;AAMjC;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OAGK;AACP;AAAA,EACE;AAAA,EACA;AAAA,EACA;AAAA,OACK;AACP,SAAS,eAAe,8BAA8B;","names":[]}

package/dist/graph/evidence-selection/prompt.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../../src/graph/evidence-selection/prompt.ts"],"sourcesContent":["export const EVIDENCE_SELECTION_PROMPT_FRAGMENT = [\n  \"Evidence selection:\",\n  '- Return evidence as [{\"sentence_id\":\"S1\",\"quote\":\"exact short source quote\"}].',\n  \"- evidence.quote is a locator for the source sentence, not a proof excerpt.\",\n  \"- Choose the shortest exact quote that still uniquely identifies the intended sentence.\",\n  \"- If the quote is too short, it may match multiple sentences; if it is too long, it may include unnecessary text or cross sentence boundaries.\",\n  \"- Do not use a whole sentence by default.\",\n  \"- evidence is an array. Use multiple items when the claim needs multiple source sentences.\",\n  \"- A source sentence means exactly one displayed source line labeled S1, S2, S3, etc.\",\n  \"- Each evidence item must point to exactly one source sentence.\",\n  \"- Each evidence item must quote one continuous span from exactly one labeled source line, for example only from S2.\",\n  \"- Never merge text from two labeled source lines into one evidence quote.\",\n  \"- sentence_id must be one of the sentence IDs shown in the source context.\",\n  \"- quote must be copied from the untagged original source sentence.\",\n  \"- If the source context contains XML-like tags, do not copy the tags into quote; copy only the visible text.\",\n].join(\"\\n\");\n\nexport const EVIDENCE_SELECTION_JSON_SHAPE = [\n  {\n    quote: \"exact short source quote copied from the selected sentence\",\n    sentence_id: \"sentence id from the source context, such as S1\",\n  },\n] as const;\n\nexport function formatEvidenceSelectionChoicePrompt(input: {\n  readonly evidenceLabel: string;\n  readonly generatedEvidence: unknown;\n  readonly candidates: readonly {\n    readonly nextText: string;\n    readonly occurrenceId: string;\n    readonly prevText: string;\n    readonly score: number;\n    readonly text: string;\n  }[];\n}): string {\n  return [\n    `Resolve only this evidence selection: ${input.evidenceLabel}.`,\n    \"The generated evidence did not match its sentence_id and quote confidently.\",\n    \"Choose exactly one candidate occurrence ID from the list below.\",\n    'Return JSON only: {\"choice\":\"S1\"}',\n    \"\",\n    \"Generated evidence:\",\n    JSON.stringify(input.generatedEvidence, null, 2),\n    \"\",\n    \"Candidates:\",\n    input.candidates.map(formatChoiceCandidate).join(\"\\n\"),\n  ].join(\"\\n\");\n}\n\nfunction formatChoiceCandidate(input: {\n  readonly nextText: string;\n  readonly occurrenceId: string;\n  readonly prevText: string;\n  readonly score: number;\n  readonly text: string;\n}): string {\n  return [\n    `${input.occurrenceId} score=${input.score.toFixed(3)}`,\n    `prev: ${formatChoiceText(input.prevText)}`,\n    `text: ${formatChoiceText(input.text)}`,\n    `next: ${formatChoiceText(input.nextText)}`,\n  ].join(\"\\n\");\n}\n\nfunction formatChoiceText(text: string): string {\n  const collapsed = text.replace(/\\s+/gu, \" \").trim();\n\n  return collapsed === \"\" ? \"(none)\" : collapsed;\n}\n"],"mappings":"AAAO,MAAM,qCAAqC;AAAA,EAChD;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AAAA,EACA;AACF,EAAE,KAAK,IAAI;AAEJ,MAAM,gCAAgC;AAAA,EAC3C;AAAA,IACE,OAAO;AAAA,IACP,aAAa;AAAA,EACf;AACF;AAEO,SAAS,oCAAoC,OAUzC;AACT,SAAO;AAAA,IACL,yCAAyC,MAAM,aAAa;AAAA,IAC5D;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA;AAAA,IACA,KAAK,UAAU,MAAM,mBAAmB,MAAM,CAAC;AAAA,IAC/C;AAAA,IACA;AAAA,IACA,MAAM,WAAW,IAAI,qBAAqB,EAAE,KAAK,IAAI;AAAA,EACvD,EAAE,KAAK,IAAI;AACb;AAEA,SAAS,sBAAsB,OAMpB;AACT,SAAO;AAAA,IACL,GAAG,MAAM,YAAY,UAAU,MAAM,MAAM,QAAQ,CAAC,CAAC;AAAA,IACrD,SAAS,iBAAiB,MAAM,QAAQ,CAAC;AAAA,IACzC,SAAS,iBAAiB,MAAM,IAAI,CAAC;AAAA,IACrC,SAAS,iBAAiB,MAAM,QAAQ,CAAC;AAAA,EAC3C,EAAE,KAAK,IAAI;AACb;AAEA,SAAS,iBAAiB,MAAsB;AAC9C,QAAM,YAAY,KAAK,QAAQ,SAAS,GAAG,EAAE,KAAK;AAElD,SAAO,cAAc,KAAK,WAAW;AACvC;","names":[]}