@staticn0va/wigolo 0.1.0 → 0.1.2

package/dist/research/branch-exploration.d.ts

@@ -0,0 +1,14 @@
+export interface BranchOptions {
+    maxConcurrent?: number;
+    totalBudgetMs?: number;
+    perQueryBudgetMs?: number;
+}
+export interface BranchResult<T> {
+    query: string;
+    ok: boolean;
+    result?: T;
+    error?: string;
+    timedOut?: boolean;
+}
+export declare function exploreInParallel<T>(queries: string[], executor: (q: string, signal: AbortSignal) => Promise<T>, options?: BranchOptions): Promise<BranchResult<T>[]>;
+//# sourceMappingURL=branch-exploration.d.ts.map

package/dist/research/branch-exploration.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"branch-exploration.d.ts","sourceRoot":"","sources":["../../src/research/branch-exploration.ts"],"names":[],"mappings":"AAIA,MAAM,WAAW,aAAa;IAC5B,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B;AAED,MAAM,WAAW,YAAY,CAAC,CAAC;IAC7B,KAAK,EAAE,MAAM,CAAC;IACd,EAAE,EAAE,OAAO,CAAC;IACZ,MAAM,CAAC,EAAE,CAAC,CAAC;IACX,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,OAAO,CAAC;CACpB;AAMD,wBAAsB,iBAAiB,CAAC,CAAC,EACvC,OAAO,EAAE,MAAM,EAAE,EACjB,QAAQ,EAAE,CAAC,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,WAAW,KAAK,OAAO,CAAC,CAAC,CAAC,EACxD,OAAO,GAAE,aAAkB,GAC1B,OAAO,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CA2D5B"}

package/dist/research/branch-exploration.js

@@ -0,0 +1,100 @@
+import { createLogger } from "../logger.js";
+const log = createLogger("research");
+const DEFAULT_MAX_CONCURRENT = 3;
+const DEFAULT_TOTAL_BUDGET_MS = 3e4;
+const DEFAULT_PER_QUERY_BUDGET_MS = 1e4;
+async function exploreInParallel(queries, executor, options = {}) {
+  if (queries.length === 0) return [];
+  const maxConcurrent = options.maxConcurrent ?? DEFAULT_MAX_CONCURRENT;
+  const totalBudgetMs = options.totalBudgetMs ?? DEFAULT_TOTAL_BUDGET_MS;
+  const perQueryBudgetMs = options.perQueryBudgetMs ?? DEFAULT_PER_QUERY_BUDGET_MS;
+  const results = new Array(queries.length);
+  let totalExceeded = false;
+  const inFlightControllers = /* @__PURE__ */ new Set();
+  const totalTimer = setTimeout(() => {
+    totalExceeded = true;
+    for (const c of inFlightControllers) c.abort();
+  }, totalBudgetMs);
+  try {
+    for (let chunkStart = 0; chunkStart < queries.length; chunkStart += maxConcurrent) {
+      const chunk = [];
+      for (let j = chunkStart; j < Math.min(chunkStart + maxConcurrent, queries.length); j++) {
+        chunk.push(j);
+      }
+      if (totalExceeded) {
+        for (const idx of chunk) {
+          results[idx] = {
+            query: queries[idx],
+            ok: false,
+            error: "total budget exceeded",
+            timedOut: true
+          };
+        }
+        continue;
+      }
+      const settled = await Promise.allSettled(
+        chunk.map(
+          (idx) => runOne(queries[idx], executor, perQueryBudgetMs, inFlightControllers, () => totalExceeded)
+        )
+      );
+      for (let k = 0; k < chunk.length; k++) {
+        const idx = chunk[k];
+        const s = settled[k];
+        if (s.status === "fulfilled") {
+          results[idx] = s.value;
+        } else {
+          results[idx] = {
+            query: queries[idx],
+            ok: false,
+            error: errorMessage(s.reason)
+          };
+        }
+      }
+    }
+  } finally {
+    clearTimeout(totalTimer);
+  }
+  return results;
+}
+async function runOne(query, executor, perQueryBudgetMs, inFlight, isTotalExceeded) {
+  const controller = new AbortController();
+  inFlight.add(controller);
+  let timedOut = false;
+  const timer = setTimeout(() => {
+    timedOut = true;
+    controller.abort();
+  }, perQueryBudgetMs);
+  try {
+    const result = await executor(query, controller.signal);
+    return { query, ok: true, result };
+  } catch (err) {
+    if (isTotalExceeded()) {
+      return {
+        query,
+        ok: false,
+        error: "total budget exceeded",
+        timedOut: true
+      };
+    }
+    if (timedOut) {
+      return {
+        query,
+        ok: false,
+        error: `per-query budget ${perQueryBudgetMs}ms exceeded`,
+        timedOut: true
+      };
+    }
+    log.debug("branch executor failed", { query, error: errorMessage(err) });
+    return { query, ok: false, error: errorMessage(err) };
+  } finally {
+    clearTimeout(timer);
+    inFlight.delete(controller);
+  }
+}
+function errorMessage(err) {
+  return err instanceof Error ? err.message : String(err);
+}
+export {
+  exploreInParallel
+};
+//# sourceMappingURL=branch-exploration.js.map

package/dist/research/branch-exploration.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/research/branch-exploration.ts"],"sourcesContent":["import { createLogger } from '../logger.js';\n\nconst log = createLogger('research');\n\nexport interface BranchOptions {\n  maxConcurrent?: number;\n  totalBudgetMs?: number;\n  perQueryBudgetMs?: number;\n}\n\nexport interface BranchResult<T> {\n  query: string;\n  ok: boolean;\n  result?: T;\n  error?: string;\n  timedOut?: boolean;\n}\n\nconst DEFAULT_MAX_CONCURRENT = 3;\nconst DEFAULT_TOTAL_BUDGET_MS = 30_000;\nconst DEFAULT_PER_QUERY_BUDGET_MS = 10_000;\n\nexport async function exploreInParallel<T>(\n  queries: string[],\n  executor: (q: string, signal: AbortSignal) => Promise<T>,\n  options: BranchOptions = {},\n): Promise<BranchResult<T>[]> {\n  if (queries.length === 0) return [];\n\n  const maxConcurrent = options.maxConcurrent ?? DEFAULT_MAX_CONCURRENT;\n  const totalBudgetMs = options.totalBudgetMs ?? DEFAULT_TOTAL_BUDGET_MS;\n  const perQueryBudgetMs = options.perQueryBudgetMs ?? DEFAULT_PER_QUERY_BUDGET_MS;\n\n  const results: BranchResult<T>[] = new Array(queries.length);\n  let totalExceeded = false;\n  const inFlightControllers = new Set<AbortController>();\n\n  const totalTimer = setTimeout(() => {\n    totalExceeded = true;\n    for (const c of inFlightControllers) c.abort();\n  }, totalBudgetMs);\n\n  try {\n    for (let chunkStart = 0; chunkStart < queries.length; chunkStart += maxConcurrent) {\n      const chunk: number[] = [];\n      for (let j = chunkStart; j < Math.min(chunkStart + maxConcurrent, queries.length); j++) {\n        chunk.push(j);\n      }\n\n      if (totalExceeded) {\n        for (const idx of chunk) {\n          results[idx] = {\n            query: queries[idx],\n            ok: false,\n            error: 'total budget exceeded',\n            timedOut: true,\n          };\n        }\n        continue;\n      }\n\n      const settled = await Promise.allSettled(\n        chunk.map((idx) =>\n          runOne(queries[idx], executor, perQueryBudgetMs, inFlightControllers, () => totalExceeded),\n        ),\n      );\n      for (let k = 0; k < chunk.length; k++) {\n        const idx = chunk[k];\n        const s = settled[k];\n        if (s.status === 'fulfilled') {\n          results[idx] = s.value;\n        } else {\n          results[idx] = {\n            query: queries[idx],\n            ok: false,\n            error: errorMessage(s.reason),\n          };\n        }\n      }\n    }\n  } finally {\n    clearTimeout(totalTimer);\n  }\n\n  return results;\n}\n\nasync function runOne<T>(\n  query: string,\n  executor: (q: string, signal: AbortSignal) => Promise<T>,\n  perQueryBudgetMs: number,\n  inFlight: Set<AbortController>,\n  isTotalExceeded: () => boolean,\n): Promise<BranchResult<T>> {\n  const controller = new AbortController();\n  inFlight.add(controller);\n  let timedOut = false;\n  const timer = setTimeout(() => {\n    timedOut = true;\n    controller.abort();\n  }, perQueryBudgetMs);\n\n  try {\n    const result = await executor(query, controller.signal);\n    return { query, ok: true, result };\n  } catch (err) {\n    if (isTotalExceeded()) {\n      return {\n        query,\n        ok: false,\n        error: 'total budget exceeded',\n        timedOut: true,\n      };\n    }\n    if (timedOut) {\n      return {\n        query,\n        ok: false,\n        error: `per-query budget ${perQueryBudgetMs}ms exceeded`,\n        timedOut: true,\n      };\n    }\n    log.debug('branch executor failed', { query, error: errorMessage(err) });\n    return { query, ok: false, error: errorMessage(err) };\n  } finally {\n    clearTimeout(timer);\n    inFlight.delete(controller);\n  }\n}\n\nfunction errorMessage(err: unknown): string {\n  return err instanceof Error ? err.message : String(err);\n}\n"],"mappings":"AAAA,SAAS,oBAAoB;AAE7B,MAAM,MAAM,aAAa,UAAU;AAgBnC,MAAM,yBAAyB;AAC/B,MAAM,0BAA0B;AAChC,MAAM,8BAA8B;AAEpC,eAAsB,kBACpB,SACA,UACA,UAAyB,CAAC,GACE;AAC5B,MAAI,QAAQ,WAAW,EAAG,QAAO,CAAC;AAElC,QAAM,gBAAgB,QAAQ,iBAAiB;AAC/C,QAAM,gBAAgB,QAAQ,iBAAiB;AAC/C,QAAM,mBAAmB,QAAQ,oBAAoB;AAErD,QAAM,UAA6B,IAAI,MAAM,QAAQ,MAAM;AAC3D,MAAI,gBAAgB;AACpB,QAAM,sBAAsB,oBAAI,IAAqB;AAErD,QAAM,aAAa,WAAW,MAAM;AAClC,oBAAgB;AAChB,eAAW,KAAK,oBAAqB,GAAE,MAAM;AAAA,EAC/C,GAAG,aAAa;AAEhB,MAAI;AACF,aAAS,aAAa,GAAG,aAAa,QAAQ,QAAQ,cAAc,eAAe;AACjF,YAAM,QAAkB,CAAC;AACzB,eAAS,IAAI,YAAY,IAAI,KAAK,IAAI,aAAa,eAAe,QAAQ,MAAM,GAAG,KAAK;AACtF,cAAM,KAAK,CAAC;AAAA,MACd;AAEA,UAAI,eAAe;AACjB,mBAAW,OAAO,OAAO;AACvB,kBAAQ,GAAG,IAAI;AAAA,YACb,OAAO,QAAQ,GAAG;AAAA,YAClB,IAAI;AAAA,YACJ,OAAO;AAAA,YACP,UAAU;AAAA,UACZ;AAAA,QACF;AACA;AAAA,MACF;AAEA,YAAM,UAAU,MAAM,QAAQ;AAAA,QAC5B,MAAM;AAAA,UAAI,CAAC,QACT,OAAO,QAAQ,GAAG,GAAG,UAAU,kBAAkB,qBAAqB,MAAM,aAAa;AAAA,QAC3F;AAAA,MACF;AACA,eAAS,IAAI,GAAG,IAAI,MAAM,QAAQ,KAAK;AACrC,cAAM,MAAM,MAAM,CAAC;AACnB,cAAM,IAAI,QAAQ,CAAC;AACnB,YAAI,EAAE,WAAW,aAAa;AAC5B,kBAAQ,GAAG,IAAI,EAAE;AAAA,QACnB,OAAO;AACL,kBAAQ,GAAG,IAAI;AAAA,YACb,OAAO,QAAQ,GAAG;AAAA,YAClB,IAAI;AAAA,YACJ,OAAO,aAAa,EAAE,MAAM;AAAA,UAC9B;AAAA,QACF;AAAA,MACF;AAAA,IACF;AAAA,EACF,UAAE;AACA,iBAAa,UAAU;AAAA,EACzB;AAEA,SAAO;AACT;AAEA,eAAe,OACb,OACA,UACA,kBACA,UACA,iBAC0B;AAC1B,QAAM,aAAa,IAAI,gBAAgB;AACvC,WAAS,IAAI,UAAU;AACvB,MAAI,WAAW;AACf,QAAM,QAAQ,WAAW,MAAM;AAC7B,eAAW;AACX,eAAW,MAAM;AAAA,EACnB,GAAG,gBAAgB;AAEnB,MAAI;AACF,UAAM,SAAS,MAAM,SAAS,OAAO,WAAW,MAAM;AACtD,WAAO,EAAE,OAAO,IAAI,MAAM,OAAO;AAAA,EACnC,SAAS,KAAK;AACZ,QAAI,gBAAgB,GAAG;AACrB,aAAO;AAAA,QACL;AAAA,QACA,IAAI;AAAA,QACJ,OAAO;AAAA,QACP,UAAU;AAAA,MACZ;AAAA,IACF;AACA,QAAI,UAAU;AACZ,aAAO;AAAA,QACL;AAAA,QACA,IAAI;AAAA,QACJ,OAAO,oBAAoB,gBAAgB;AAAA,QAC3C,UAAU;AAAA,MACZ;AAAA,IACF;AACA,QAAI,MAAM,0BAA0B,EAAE,OAAO,OAAO,aAAa,GAAG,EAAE,CAAC;AACvE,WAAO,EAAE,OAAO,IAAI,OAAO,OAAO,aAAa,GAAG,EAAE;AAAA,EACtD,UAAE;AACA,iBAAa,KAAK;AAClB,aAAS,OAAO,UAAU;AAAA,EAC5B;AACF;AAEA,SAAS,aAAa,KAAsB;AAC1C,SAAO,eAAe,QAAQ,IAAI,UAAU,OAAO,GAAG;AACxD;","names":[]}

package/dist/research/brief.d.ts

@@ -0,0 +1,6 @@
+import type { ResearchBrief, ResearchSource, CrossReference } from '../types.js';
+import type { QueryType } from './decompose.js';
+export declare function buildResearchBrief(question: string, sources: ResearchSource[], subQueries: string[], perSourceCharCap: number, totalSourcesCharCap: number, queryType?: QueryType, comparisonEntities?: string[], synthesisText?: string): Promise<ResearchBrief>;
+export declare function detectCrossReferences(sources: ResearchSource[]): CrossReference[];
+export declare function stripMarkdownLinks(text: string): string;
+//# sourceMappingURL=brief.d.ts.map

package/dist/research/brief.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"brief.d.ts","sourceRoot":"","sources":["../../src/research/brief.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,cAAc,EAAoB,cAAc,EAAE,MAAM,aAAa,CAAC;AACnG,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,gBAAgB,CAAC;AAahD,wBAAsB,kBAAkB,CACtC,QAAQ,EAAE,MAAM,EAChB,OAAO,EAAE,cAAc,EAAE,EACzB,UAAU,EAAE,MAAM,EAAE,EACpB,gBAAgB,EAAE,MAAM,EACxB,mBAAmB,EAAE,MAAM,EAC3B,SAAS,GAAE,SAAqB,EAChC,kBAAkB,GAAE,MAAM,EAAO,EACjC,aAAa,CAAC,EAAE,MAAM,GACrB,OAAO,CAAC,aAAa,CAAC,CAgDxB;AA6BD,wBAAgB,qBAAqB,CAAC,OAAO,EAAE,cAAc,EAAE,GAAG,cAAc,EAAE,CAuCjF;AA6FD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAOvD"}

package/dist/research/brief.js

@@ -0,0 +1,246 @@
+import { extractHighlights } from "../search/highlights.js";
+import { buildCitationGraph } from "./citation-graph.js";
+const MAX_HIGHLIGHTS = 12;
+const MAX_KEY_FINDING_LEN = 280;
+const MAX_TOPICS = 8;
+const MAX_CROSS_REFS = 10;
+const MIN_PHRASE_LEN = 4;
+async function buildResearchBrief(question, sources, subQueries, perSourceCharCap, totalSourcesCharCap, queryType = "general", comparisonEntities = [], synthesisText) {
+  const fetched = sources.filter((s) => s.fetched && s.markdown_content.length > 0);
+  const searchItems = fetched.map((s) => ({
+    title: s.title,
+    url: s.url,
+    snippet: s.markdown_content.slice(0, 200),
+    markdown_content: s.markdown_content,
+    relevance_score: s.relevance_score
+  }));
+  const { highlights } = await extractHighlights(question, searchItems, MAX_HIGHLIGHTS);
+  const topics = buildTopics(subQueries, fetched);
+  const keyFindings = buildKeyFindings(fetched);
+  const crossReferences = detectCrossReferences(fetched);
+  const gaps = detectGaps(subQueries, fetched);
+  const comparison = queryType === "comparison" && comparisonEntities.length >= 2 ? buildComparisonSection(comparisonEntities, fetched) : void 0;
+  const citationGraph = synthesisText && synthesisText.trim().length > 0 && fetched.length > 0 ? buildCitationGraph(
+    synthesisText,
+    fetched.map((s) => ({ url: s.url, title: s.title, markdown: s.markdown_content }))
+  ) : void 0;
+  return {
+    topics,
+    highlights,
+    key_findings: keyFindings,
+    per_source_char_cap: perSourceCharCap,
+    total_sources_char_cap: totalSourcesCharCap,
+    sections: {
+      overview: {
+        key_findings: keyFindings.slice(0, 5),
+        cross_references: crossReferences
+      },
+      ...comparison ? { comparison } : {},
+      gaps
+    },
+    query_type: queryType,
+    ...citationGraph && citationGraph.length > 0 ? { citation_graph: citationGraph } : {}
+  };
+}
+function buildTopics(subQueries, sources) {
+  if (subQueries.length > 0) {
+    return dedupe(subQueries).slice(0, MAX_TOPICS);
+  }
+  const labels = sources.map((s) => s.title.split(/[–|:·-]/)[0].trim()).filter((t) => t.length >= 5 && t.length <= 100);
+  return dedupe(labels).slice(0, MAX_TOPICS);
+}
+function buildKeyFindings(sources) {
+  const out = [];
+  for (const s of [...sources].sort((a, b) => b.relevance_score - a.relevance_score)) {
+    const first = firstSubstantiveParagraph(s.markdown_content);
+    if (!first) continue;
+    const trimmed = first.length > MAX_KEY_FINDING_LEN ? first.slice(0, MAX_KEY_FINDING_LEN - 1).trimEnd() + "\u2026" : first;
+    out.push(trimmed);
+  }
+  return dedupe(out);
+}
+function detectCrossReferences(sources) {
+  if (sources.length < 2) return [];
+  const phraseMap = /* @__PURE__ */ new Map();
+  for (let idx = 0; idx < sources.length; idx++) {
+    const content = sources[idx].markdown_content.toLowerCase();
+    const words = content.replace(/[^a-z0-9\s]/g, " ").split(/\s+/).filter((w) => w.length >= MIN_PHRASE_LEN && !STOP_WORDS.has(w));
+    const seenForSource = /* @__PURE__ */ new Set();
+    for (let i = 0; i < words.length - 2; i++) {
+      const phrase = words.slice(i, i + 3).join(" ");
+      if (seenForSource.has(phrase)) continue;
+      seenForSource.add(phrase);
+      if (!phraseMap.has(phrase)) phraseMap.set(phrase, /* @__PURE__ */ new Set());
+      phraseMap.get(phrase).add(idx);
+    }
+  }
+  const candidates = [];
+  for (const [phrase, sourceIndices] of phraseMap) {
+    if (sourceIndices.size >= 2) {
+      candidates.push({
+        finding: phrase,
+        source_indices: [...sourceIndices].sort(),
+        confidence: sourceIndices.size >= 3 ? "high" : "medium"
+      });
+    }
+  }
+  candidates.sort((a, b) => b.source_indices.length - a.source_indices.length);
+  return deduplicateOverlapping(candidates).slice(0, MAX_CROSS_REFS);
+}
+function deduplicateOverlapping(refs) {
+  const kept = [];
+  const usedWords = /* @__PURE__ */ new Set();
+  for (const ref of refs) {
+    const words = ref.finding.split(" ");
+    const overlapCount = words.filter((w) => usedWords.has(w)).length;
+    if (overlapCount >= words.length - 1 && kept.length > 0) continue;
+    kept.push(ref);
+    for (const w of words) usedWords.add(w);
+  }
+  return kept;
+}
+function detectGaps(subQueries, sources) {
+  if (subQueries.length === 0) return [];
+  const gaps = [];
+  const contentLower = sources.map((s) => s.markdown_content.toLowerCase()).join(" ");
+  for (const query of subQueries) {
+    const words = query.toLowerCase().replace(/[^a-z0-9\s]/g, " ").split(/\s+/).filter((w) => w.length >= MIN_PHRASE_LEN && !STOP_WORDS.has(w));
+    if (words.length === 0) continue;
+    const found = words.filter((w) => contentLower.includes(w)).length;
+    const coverage = found / words.length;
+    if (coverage < 0.5) {
+      gaps.push(`Limited coverage for: "${query}"`);
+    }
+  }
+  return gaps;
+}
+function buildComparisonSection(entities, sources) {
+  const comparisonPoints = [];
+  const contentLower = sources.map((s) => s.markdown_content.toLowerCase()).join("\n");
+  const comparisonTerms = [
+    "faster",
+    "slower",
+    "better",
+    "worse",
+    "more",
+    "less",
+    "easier",
+    "harder",
+    "simpler",
+    "complex",
+    "lightweight",
+    "heavy",
+    "performance",
+    "scalability",
+    "ecosystem",
+    "community",
+    "support"
+  ];
+  for (const term of comparisonTerms) {
+    if (!contentLower.includes(term)) continue;
+    const nearEntity = entities.some((e) => {
+      const entityLower = e.toLowerCase();
+      const idx = contentLower.indexOf(entityLower);
+      if (idx === -1) return false;
+      const neighborhood = contentLower.slice(Math.max(0, idx - 200), idx + e.length + 200);
+      return neighborhood.includes(term);
+    });
+    if (nearEntity) {
+      comparisonPoints.push(term);
+    }
+  }
+  return { entities, comparison_points: [...new Set(comparisonPoints)] };
+}
+function firstSubstantiveParagraph(markdown) {
+  const paragraphs = markdown.split(/\n\n+/).map((p) => p.trim());
+  for (const p of paragraphs) {
+    if (p.length < 80) continue;
+    if (p.startsWith("#") || p.startsWith("|") || p.startsWith("```")) continue;
+    const cleaned = stripMarkdownLinks(p);
+    if (cleaned.length < 80) continue;
+    return cleaned.replace(/\s+/g, " ");
+  }
+  return null;
+}
+function stripMarkdownLinks(text) {
+  return text.replace(/!\[[^\]]*\]\([^)]*\)/g, "").replace(/\[!\[[^\]]*\]\([^)]*\)\]\([^)]*\)/g, "").replace(/\[([^\]]+)\]\([^)]*\)/g, "$1").replace(/<https?:\/\/[^>]+>/g, "").trim();
+}
+function dedupe(list) {
+  const seen = /* @__PURE__ */ new Set();
+  const out = [];
+  for (const item of list) {
+    const key = item.toLowerCase();
+    if (seen.has(key)) continue;
+    seen.add(key);
+    out.push(item);
+  }
+  return out;
+}
+const STOP_WORDS = /* @__PURE__ */ new Set([
+  "about",
+  "after",
+  "also",
+  "been",
+  "before",
+  "being",
+  "between",
+  "both",
+  "could",
+  "does",
+  "doing",
+  "done",
+  "each",
+  "even",
+  "every",
+  "from",
+  "have",
+  "here",
+  "into",
+  "just",
+  "like",
+  "made",
+  "make",
+  "many",
+  "more",
+  "most",
+  "much",
+  "must",
+  "need",
+  "only",
+  "other",
+  "over",
+  "same",
+  "should",
+  "some",
+  "such",
+  "than",
+  "that",
+  "their",
+  "them",
+  "then",
+  "there",
+  "these",
+  "they",
+  "this",
+  "those",
+  "through",
+  "very",
+  "want",
+  "well",
+  "were",
+  "what",
+  "when",
+  "where",
+  "which",
+  "while",
+  "will",
+  "with",
+  "would",
+  "your"
+]);
+export {
+  buildResearchBrief,
+  detectCrossReferences,
+  stripMarkdownLinks
+};
+//# sourceMappingURL=brief.js.map

package/dist/research/brief.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/research/brief.ts"],"sourcesContent":["import type { ResearchBrief, ResearchSource, SearchResultItem, CrossReference } from '../types.js';\nimport type { QueryType } from './decompose.js';\nimport { extractHighlights } from '../search/highlights.js';\nimport { buildCitationGraph } from './citation-graph.js';\n\nconst MAX_HIGHLIGHTS = 12;\nconst MAX_KEY_FINDING_LEN = 280;\nconst MAX_TOPICS = 8;\nconst MAX_CROSS_REFS = 10;\nconst MIN_PHRASE_LEN = 4;\n\n// Build a host-LLM-friendly structured brief when internal sampling is\n// unavailable. The host model (Claude Code / Cursor / etc.) consumes this\n// shape to produce the final report without needing to re-read raw sources.\nexport async function buildResearchBrief(\n  question: string,\n  sources: ResearchSource[],\n  subQueries: string[],\n  perSourceCharCap: number,\n  totalSourcesCharCap: number,\n  queryType: QueryType = 'general',\n  comparisonEntities: string[] = [],\n  synthesisText?: string,\n): Promise<ResearchBrief> {\n  const fetched = sources.filter((s) => s.fetched && s.markdown_content.length > 0);\n\n  // Highlights reuse the ONNX-reranker-or-paragraph scorer so briefs align with\n  // whatever format='highlights' produces for single-query searches.\n  const searchItems: SearchResultItem[] = fetched.map((s) => ({\n    title: s.title,\n    url: s.url,\n    snippet: s.markdown_content.slice(0, 200),\n    markdown_content: s.markdown_content,\n    relevance_score: s.relevance_score,\n  }));\n\n  const { highlights } = await extractHighlights(question, searchItems, MAX_HIGHLIGHTS);\n\n  const topics = buildTopics(subQueries, fetched);\n  const keyFindings = buildKeyFindings(fetched);\n  const crossReferences = detectCrossReferences(fetched);\n  const gaps = detectGaps(subQueries, fetched);\n\n  const comparison = queryType === 'comparison' && comparisonEntities.length >= 2\n    ? buildComparisonSection(comparisonEntities, fetched)\n    : undefined;\n\n  const citationGraph = synthesisText && synthesisText.trim().length > 0 && fetched.length > 0\n    ? buildCitationGraph(\n        synthesisText,\n        fetched.map((s) => ({ url: s.url, title: s.title, markdown: s.markdown_content })),\n      )\n    : undefined;\n\n  return {\n    topics,\n    highlights,\n    key_findings: keyFindings,\n    per_source_char_cap: perSourceCharCap,\n    total_sources_char_cap: totalSourcesCharCap,\n    sections: {\n      overview: {\n        key_findings: keyFindings.slice(0, 5),\n        cross_references: crossReferences,\n      },\n      ...(comparison ? { comparison } : {}),\n      gaps,\n    },\n    query_type: queryType,\n    ...(citationGraph && citationGraph.length > 0 ? { citation_graph: citationGraph } : {}),\n  };\n}\n\n// Prefer sub-queries (planner's view of the topic space) when available;\n// otherwise derive compact topic labels from source titles.\nfunction buildTopics(subQueries: string[], sources: ResearchSource[]): string[] {\n  if (subQueries.length > 0) {\n    return dedupe(subQueries).slice(0, MAX_TOPICS);\n  }\n  const labels = sources\n    .map((s) => s.title.split(/[–|:·-]/)[0].trim())\n    .filter((t) => t.length >= 5 && t.length <= 100);\n  return dedupe(labels).slice(0, MAX_TOPICS);\n}\n\n// First substantive paragraph per source, trimmed to a finding-sized blurb.\n// Ordered by source relevance so the most-weighted finding is first.\nfunction buildKeyFindings(sources: ResearchSource[]): string[] {\n  const out: string[] = [];\n  for (const s of [...sources].sort((a, b) => b.relevance_score - a.relevance_score)) {\n    const first = firstSubstantiveParagraph(s.markdown_content);\n    if (!first) continue;\n    const trimmed = first.length > MAX_KEY_FINDING_LEN\n      ? first.slice(0, MAX_KEY_FINDING_LEN - 1).trimEnd() + '…'\n      : first;\n    out.push(trimmed);\n  }\n  return dedupe(out);\n}\n\nexport function detectCrossReferences(sources: ResearchSource[]): CrossReference[] {\n  if (sources.length < 2) return [];\n\n  // Extract significant phrases from each source's content\n  const phraseMap = new Map<string, Set<number>>();\n\n  for (let idx = 0; idx < sources.length; idx++) {\n    const content = sources[idx].markdown_content.toLowerCase();\n    const words = content\n      .replace(/[^a-z0-9\\s]/g, ' ')\n      .split(/\\s+/)\n      .filter((w) => w.length >= MIN_PHRASE_LEN && !STOP_WORDS.has(w));\n\n    const seenForSource = new Set<string>();\n    for (let i = 0; i < words.length - 2; i++) {\n      const phrase = words.slice(i, i + 3).join(' ');\n      if (seenForSource.has(phrase)) continue;\n      seenForSource.add(phrase);\n\n      if (!phraseMap.has(phrase)) phraseMap.set(phrase, new Set());\n      phraseMap.get(phrase)!.add(idx);\n    }\n  }\n\n  // Phrases found in 2+ sources are cross-references\n  const candidates: CrossReference[] = [];\n  for (const [phrase, sourceIndices] of phraseMap) {\n    if (sourceIndices.size >= 2) {\n      candidates.push({\n        finding: phrase,\n        source_indices: [...sourceIndices].sort(),\n        confidence: sourceIndices.size >= 3 ? 'high' : 'medium',\n      });\n    }\n  }\n\n  // Sort by number of sources (desc), then deduplicate overlapping phrases\n  candidates.sort((a, b) => b.source_indices.length - a.source_indices.length);\n  return deduplicateOverlapping(candidates).slice(0, MAX_CROSS_REFS);\n}\n\nfunction deduplicateOverlapping(refs: CrossReference[]): CrossReference[] {\n  const kept: CrossReference[] = [];\n  const usedWords = new Set<string>();\n\n  for (const ref of refs) {\n    const words = ref.finding.split(' ');\n    // Skip if most words already covered by a higher-ranked cross-reference\n    const overlapCount = words.filter((w) => usedWords.has(w)).length;\n    if (overlapCount >= words.length - 1 && kept.length > 0) continue;\n\n    kept.push(ref);\n    for (const w of words) usedWords.add(w);\n  }\n\n  return kept;\n}\n\nfunction detectGaps(subQueries: string[], sources: ResearchSource[]): string[] {\n  if (subQueries.length === 0) return [];\n\n  const gaps: string[] = [];\n  const contentLower = sources.map((s) => s.markdown_content.toLowerCase()).join(' ');\n\n  for (const query of subQueries) {\n    // Extract significant words from sub-query\n    const words = query.toLowerCase()\n      .replace(/[^a-z0-9\\s]/g, ' ')\n      .split(/\\s+/)\n      .filter((w) => w.length >= MIN_PHRASE_LEN && !STOP_WORDS.has(w));\n\n    if (words.length === 0) continue;\n\n    // Count how many significant words appear in any source\n    const found = words.filter((w) => contentLower.includes(w)).length;\n    const coverage = found / words.length;\n\n    if (coverage < 0.5) {\n      gaps.push(`Limited coverage for: \"${query}\"`);\n    }\n  }\n\n  return gaps;\n}\n\nfunction buildComparisonSection(\n  entities: string[],\n  sources: ResearchSource[],\n): { entities: string[]; comparison_points: string[] } {\n  const comparisonPoints: string[] = [];\n  const contentLower = sources.map((s) => s.markdown_content.toLowerCase()).join('\\n');\n\n  // Look for comparison keywords near entity mentions\n  const comparisonTerms = ['faster', 'slower', 'better', 'worse', 'more', 'less',\n    'easier', 'harder', 'simpler', 'complex', 'lightweight', 'heavy',\n    'performance', 'scalability', 'ecosystem', 'community', 'support'];\n\n  for (const term of comparisonTerms) {\n    if (!contentLower.includes(term)) continue;\n\n    // Check if term appears near any entity\n    const nearEntity = entities.some((e) => {\n      const entityLower = e.toLowerCase();\n      const idx = contentLower.indexOf(entityLower);\n      if (idx === -1) return false;\n      // Check within 200 chars of entity mention\n      const neighborhood = contentLower.slice(Math.max(0, idx - 200), idx + e.length + 200);\n      return neighborhood.includes(term);\n    });\n\n    if (nearEntity) {\n      comparisonPoints.push(term);\n    }\n  }\n\n  return { entities, comparison_points: [...new Set(comparisonPoints)] };\n}\n\nfunction firstSubstantiveParagraph(markdown: string): string | null {\n  const paragraphs = markdown.split(/\\n\\n+/).map((p) => p.trim());\n  for (const p of paragraphs) {\n    if (p.length < 80) continue;\n    if (p.startsWith('#') || p.startsWith('|') || p.startsWith('```')) continue;\n    const cleaned = stripMarkdownLinks(p);\n    if (cleaned.length < 80) continue;\n    return cleaned.replace(/\\s+/g, ' ');\n  }\n  return null;\n}\n\n// Flatten markdown link/image syntax to plain text so a downstream char-slice\n// can't chop mid-link and leak `](/?source=post_page...` into key_findings.\nexport function stripMarkdownLinks(text: string): string {\n  return text\n    .replace(/!\\[[^\\]]*\\]\\([^)]*\\)/g, '')\n    .replace(/\\[!\\[[^\\]]*\\]\\([^)]*\\)\\]\\([^)]*\\)/g, '')\n    .replace(/\\[([^\\]]+)\\]\\([^)]*\\)/g, '$1')\n    .replace(/<https?:\\/\\/[^>]+>/g, '')\n    .trim();\n}\n\nfunction dedupe(list: string[]): string[] {\n  const seen = new Set<string>();\n  const out: string[] = [];\n  for (const item of list) {\n    const key = item.toLowerCase();\n    if (seen.has(key)) continue;\n    seen.add(key);\n    out.push(item);\n  }\n  return out;\n}\n\nconst STOP_WORDS = new Set([\n  'about', 'after', 'also', 'been', 'before', 'being', 'between',\n  'both', 'could', 'does', 'doing', 'done', 'each', 'even', 'every',\n  'from', 'have', 'here', 'into', 'just', 'like', 'made', 'make',\n  'many', 'more', 'most', 'much', 'must', 'need', 'only', 'other',\n  'over', 'same', 'should', 'some', 'such', 'than', 'that', 'their',\n  'them', 'then', 'there', 'these', 'they', 'this', 'those', 'through',\n  'very', 'want', 'well', 'were', 'what', 'when', 'where', 'which',\n  'while', 'will', 'with', 'would', 'your',\n]);\n"],"mappings":"AAEA,SAAS,yBAAyB;AAClC,SAAS,0BAA0B;AAEnC,MAAM,iBAAiB;AACvB,MAAM,sBAAsB;AAC5B,MAAM,aAAa;AACnB,MAAM,iBAAiB;AACvB,MAAM,iBAAiB;AAKvB,eAAsB,mBACpB,UACA,SACA,YACA,kBACA,qBACA,YAAuB,WACvB,qBAA+B,CAAC,GAChC,eACwB;AACxB,QAAM,UAAU,QAAQ,OAAO,CAAC,MAAM,EAAE,WAAW,EAAE,iBAAiB,SAAS,CAAC;AAIhF,QAAM,cAAkC,QAAQ,IAAI,CAAC,OAAO;AAAA,IAC1D,OAAO,EAAE;AAAA,IACT,KAAK,EAAE;AAAA,IACP,SAAS,EAAE,iBAAiB,MAAM,GAAG,GAAG;AAAA,IACxC,kBAAkB,EAAE;AAAA,IACpB,iBAAiB,EAAE;AAAA,EACrB,EAAE;AAEF,QAAM,EAAE,WAAW,IAAI,MAAM,kBAAkB,UAAU,aAAa,cAAc;AAEpF,QAAM,SAAS,YAAY,YAAY,OAAO;AAC9C,QAAM,cAAc,iBAAiB,OAAO;AAC5C,QAAM,kBAAkB,sBAAsB,OAAO;AACrD,QAAM,OAAO,WAAW,YAAY,OAAO;AAE3C,QAAM,aAAa,cAAc,gBAAgB,mBAAmB,UAAU,IAC1E,uBAAuB,oBAAoB,OAAO,IAClD;AAEJ,QAAM,gBAAgB,iBAAiB,cAAc,KAAK,EAAE,SAAS,KAAK,QAAQ,SAAS,IACvF;AAAA,IACE;AAAA,IACA,QAAQ,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,KAAK,OAAO,EAAE,OAAO,UAAU,EAAE,iBAAiB,EAAE;AAAA,EACnF,IACA;AAEJ,SAAO;AAAA,IACL;AAAA,IACA;AAAA,IACA,cAAc;AAAA,IACd,qBAAqB;AAAA,IACrB,wBAAwB;AAAA,IACxB,UAAU;AAAA,MACR,UAAU;AAAA,QACR,cAAc,YAAY,MAAM,GAAG,CAAC;AAAA,QACpC,kBAAkB;AAAA,MACpB;AAAA,MACA,GAAI,aAAa,EAAE,WAAW,IAAI,CAAC;AAAA,MACnC;AAAA,IACF;AAAA,IACA,YAAY;AAAA,IACZ,GAAI,iBAAiB,cAAc,SAAS,IAAI,EAAE,gBAAgB,cAAc,IAAI,CAAC;AAAA,EACvF;AACF;AAIA,SAAS,YAAY,YAAsB,SAAqC;AAC9E,MAAI,WAAW,SAAS,GAAG;AACzB,WAAO,OAAO,UAAU,EAAE,MAAM,GAAG,UAAU;AAAA,EAC/C;AACA,QAAM,SAAS,QACZ,IAAI,CAAC,MAAM,EAAE,MAAM,MAAM,SAAS,EAAE,CAAC,EAAE,KAAK,CAAC,EAC7C,OAAO,CAAC,MAAM,EAAE,UAAU,KAAK,EAAE,UAAU,GAAG;AACjD,SAAO,OAAO,MAAM,EAAE,MAAM,GAAG,UAAU;AAC3C;AAIA,SAAS,iBAAiB,SAAqC;AAC7D,QAAM,MAAgB,CAAC;AACvB,aAAW,KAAK,CAAC,GAAG,OAAO,EAAE,KAAK,CAAC,GAAG,MAAM,EAAE,kBAAkB,EAAE,eAAe,GAAG;AAClF,UAAM,QAAQ,0BAA0B,EAAE,gBAAgB;AAC1D,QAAI,CAAC,MAAO;AACZ,UAAM,UAAU,MAAM,SAAS,sBAC3B,MAAM,MAAM,GAAG,sBAAsB,CAAC,EAAE,QAAQ,IAAI,WACpD;AACJ,QAAI,KAAK,OAAO;AAAA,EAClB;AACA,SAAO,OAAO,GAAG;AACnB;AAEO,SAAS,sBAAsB,SAA6C;AACjF,MAAI,QAAQ,SAAS,EAAG,QAAO,CAAC;AAGhC,QAAM,YAAY,oBAAI,IAAyB;AAE/C,WAAS,MAAM,GAAG,MAAM,QAAQ,QAAQ,OAAO;AAC7C,UAAM,UAAU,QAAQ,GAAG,EAAE,iBAAiB,YAAY;AAC1D,UAAM,QAAQ,QACX,QAAQ,gBAAgB,GAAG,EAC3B,MAAM,KAAK,EACX,OAAO,CAAC,MAAM,EAAE,UAAU,kBAAkB,CAAC,WAAW,IAAI,CAAC,CAAC;AAEjE,UAAM,gBAAgB,oBAAI,IAAY;AACtC,aAAS,IAAI,GAAG,IAAI,MAAM,SAAS,GAAG,KAAK;AACzC,YAAM,SAAS,MAAM,MAAM,GAAG,IAAI,CAAC,EAAE,KAAK,GAAG;AAC7C,UAAI,cAAc,IAAI,MAAM,EAAG;AAC/B,oBAAc,IAAI,MAAM;AAExB,UAAI,CAAC,UAAU,IAAI,MAAM,EAAG,WAAU,IAAI,QAAQ,oBAAI,IAAI,CAAC;AAC3D,gBAAU,IAAI,MAAM,EAAG,IAAI,GAAG;AAAA,IAChC;AAAA,EACF;AAGA,QAAM,aAA+B,CAAC;AACtC,aAAW,CAAC,QAAQ,aAAa,KAAK,WAAW;AAC/C,QAAI,cAAc,QAAQ,GAAG;AAC3B,iBAAW,KAAK;AAAA,QACd,SAAS;AAAA,QACT,gBAAgB,CAAC,GAAG,aAAa,EAAE,KAAK;AAAA,QACxC,YAAY,cAAc,QAAQ,IAAI,SAAS;AAAA,MACjD,CAAC;AAAA,IACH;AAAA,EACF;AAGA,aAAW,KAAK,CAAC,GAAG,MAAM,EAAE,eAAe,SAAS,EAAE,eAAe,MAAM;AAC3E,SAAO,uBAAuB,UAAU,EAAE,MAAM,GAAG,cAAc;AACnE;AAEA,SAAS,uBAAuB,MAA0C;AACxE,QAAM,OAAyB,CAAC;AAChC,QAAM,YAAY,oBAAI,IAAY;AAElC,aAAW,OAAO,MAAM;AACtB,UAAM,QAAQ,IAAI,QAAQ,MAAM,GAAG;AAEnC,UAAM,eAAe,MAAM,OAAO,CAAC,MAAM,UAAU,IAAI,CAAC,CAAC,EAAE;AAC3D,QAAI,gBAAgB,MAAM,SAAS,KAAK,KAAK,SAAS,EAAG;AAEzD,SAAK,KAAK,GAAG;AACb,eAAW,KAAK,MAAO,WAAU,IAAI,CAAC;AAAA,EACxC;AAEA,SAAO;AACT;AAEA,SAAS,WAAW,YAAsB,SAAqC;AAC7E,MAAI,WAAW,WAAW,EAAG,QAAO,CAAC;AAErC,QAAM,OAAiB,CAAC;AACxB,QAAM,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,iBAAiB,YAAY,CAAC,EAAE,KAAK,GAAG;AAElF,aAAW,SAAS,YAAY;AAE9B,UAAM,QAAQ,MAAM,YAAY,EAC7B,QAAQ,gBAAgB,GAAG,EAC3B,MAAM,KAAK,EACX,OAAO,CAAC,MAAM,EAAE,UAAU,kBAAkB,CAAC,WAAW,IAAI,CAAC,CAAC;AAEjE,QAAI,MAAM,WAAW,EAAG;AAGxB,UAAM,QAAQ,MAAM,OAAO,CAAC,MAAM,aAAa,SAAS,CAAC,CAAC,EAAE;AAC5D,UAAM,WAAW,QAAQ,MAAM;AAE/B,QAAI,WAAW,KAAK;AAClB,WAAK,KAAK,0BAA0B,KAAK,GAAG;AAAA,IAC9C;AAAA,EACF;AAEA,SAAO;AACT;AAEA,SAAS,uBACP,UACA,SACqD;AACrD,QAAM,mBAA6B,CAAC;AACpC,QAAM,eAAe,QAAQ,IAAI,CAAC,MAAM,EAAE,iBAAiB,YAAY,CAAC,EAAE,KAAK,IAAI;AAGnF,QAAM,kBAAkB;AAAA,IAAC;AAAA,IAAU;AAAA,IAAU;AAAA,IAAU;AAAA,IAAS;AAAA,IAAQ;AAAA,IACtE;AAAA,IAAU;AAAA,IAAU;AAAA,IAAW;AAAA,IAAW;AAAA,IAAe;AAAA,IACzD;AAAA,IAAe;AAAA,IAAe;AAAA,IAAa;AAAA,IAAa;AAAA,EAAS;AAEnE,aAAW,QAAQ,iBAAiB;AAClC,QAAI,CAAC,aAAa,SAAS,IAAI,EAAG;AAGlC,UAAM,aAAa,SAAS,KAAK,CAAC,MAAM;AACtC,YAAM,cAAc,EAAE,YAAY;AAClC,YAAM,MAAM,aAAa,QAAQ,WAAW;AAC5C,UAAI,QAAQ,GAAI,QAAO;AAEvB,YAAM,eAAe,aAAa,MAAM,KAAK,IAAI,GAAG,MAAM,GAAG,GAAG,MAAM,EAAE,SAAS,GAAG;AACpF,aAAO,aAAa,SAAS,IAAI;AAAA,IACnC,CAAC;AAED,QAAI,YAAY;AACd,uBAAiB,KAAK,IAAI;AAAA,IAC5B;AAAA,EACF;AAEA,SAAO,EAAE,UAAU,mBAAmB,CAAC,GAAG,IAAI,IAAI,gBAAgB,CAAC,EAAE;AACvE;AAEA,SAAS,0BAA0B,UAAiC;AAClE,QAAM,aAAa,SAAS,MAAM,OAAO,EAAE,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC;AAC9D,aAAW,KAAK,YAAY;AAC1B,QAAI,EAAE,SAAS,GAAI;AACnB,QAAI,EAAE,WAAW,GAAG,KAAK,EAAE,WAAW,GAAG,KAAK,EAAE,WAAW,KAAK,EAAG;AACnE,UAAM,UAAU,mBAAmB,CAAC;AACpC,QAAI,QAAQ,SAAS,GAAI;AACzB,WAAO,QAAQ,QAAQ,QAAQ,GAAG;AAAA,EACpC;AACA,SAAO;AACT;AAIO,SAAS,mBAAmB,MAAsB;AACvD,SAAO,KACJ,QAAQ,yBAAyB,EAAE,EACnC,QAAQ,sCAAsC,EAAE,EAChD,QAAQ,0BAA0B,IAAI,EACtC,QAAQ,uBAAuB,EAAE,EACjC,KAAK;AACV;AAEA,SAAS,OAAO,MAA0B;AACxC,QAAM,OAAO,oBAAI,IAAY;AAC7B,QAAM,MAAgB,CAAC;AACvB,aAAW,QAAQ,MAAM;AACvB,UAAM,MAAM,KAAK,YAAY;AAC7B,QAAI,KAAK,IAAI,GAAG,EAAG;AACnB,SAAK,IAAI,GAAG;AACZ,QAAI,KAAK,IAAI;AAAA,EACf;AACA,SAAO;AACT;AAEA,MAAM,aAAa,oBAAI,IAAI;AAAA,EACzB;AAAA,EAAS;AAAA,EAAS;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAU;AAAA,EAAS;AAAA,EACrD;AAAA,EAAQ;AAAA,EAAS;AAAA,EAAQ;AAAA,EAAS;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAC1D;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EACxD;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EACxD;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAU;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAC1D;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAS;AAAA,EAAS;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAS;AAAA,EAC3D;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAS;AAAA,EACzD;AAAA,EAAS;AAAA,EAAQ;AAAA,EAAQ;AAAA,EAAS;AACpC,CAAC;","names":[]}

package/dist/research/citation-graph.d.ts

@@ -0,0 +1,9 @@
+import type { CitationGraphEntry } from '../types.js';
+export type { CitationGraphEntry };
+export interface CitationGraphSource {
+    url: string;
+    title: string;
+    markdown: string;
+}
+export declare function buildCitationGraph(synthesisText: string, sources: CitationGraphSource[]): CitationGraphEntry[];
+//# sourceMappingURL=citation-graph.d.ts.map

package/dist/research/citation-graph.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"citation-graph.d.ts","sourceRoot":"","sources":["../../src/research/citation-graph.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,aAAa,CAAC;AAEtD,YAAY,EAAE,kBAAkB,EAAE,CAAC;AAEnC,MAAM,WAAW,mBAAmB;IAClC,GAAG,EAAE,MAAM,CAAC;IACZ,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;CAClB;AAcD,wBAAgB,kBAAkB,CAChC,aAAa,EAAE,MAAM,EACrB,OAAO,EAAE,mBAAmB,EAAE,GAC7B,kBAAkB,EAAE,CAmDtB"}

package/dist/research/citation-graph.js

@@ -0,0 +1,114 @@
+const MAX_ENTRIES = 50;
+const MAX_SOURCES_PER_CLAIM = 3;
+const JACCARD_THRESHOLD = 0.2;
+const MIN_SENTENCE_LEN = 10;
+const MIN_TOKEN_LEN = 4;
+const STOPWORDS = /* @__PURE__ */ new Set([
+  "a",
+  "an",
+  "the",
+  "is",
+  "are",
+  "was",
+  "were",
+  "and",
+  "or",
+  "but",
+  "in",
+  "on",
+  "at",
+  "to",
+  "for",
+  "of",
+  "with",
+  "by",
+  "from",
+  "as",
+  "that",
+  "this",
+  "it",
+  "its",
+  "be",
+  "been",
+  "has",
+  "have",
+  "had"
+]);
+function buildCitationGraph(synthesisText, sources) {
+  if (!synthesisText || synthesisText.trim().length === 0) return [];
+  const sentenceMatches = synthesisText.match(/[^.!?]+[.!?]+(?:\s|$)/g);
+  const sentences = sentenceMatches && sentenceMatches.length > 0 ? sentenceMatches.map((s) => s.trim()) : [synthesisText.trim()];
+  const sourceTokens = sources.map((s) => tokenize(s.markdown));
+  const entries = [];
+  for (const sentence of sentences) {
+    if (entries.length >= MAX_ENTRIES) break;
+    if (sentence.length < MIN_SENTENCE_LEN) continue;
+    const markerIndices = extractMarkers(sentence, sources.length);
+    if (markerIndices.length > 0) {
+      entries.push({
+        claim: sentence,
+        source_indices: markerIndices,
+        confidence: "high"
+      });
+      continue;
+    }
+    const sentenceTokens = tokenize(sentence);
+    if (sentenceTokens.size === 0 || sourceTokens.length === 0) {
+      entries.push({ claim: sentence, source_indices: [], confidence: "low" });
+      continue;
+    }
+    const scored = [];
+    for (let i = 0; i < sourceTokens.length; i++) {
+      const score = jaccard(sentenceTokens, sourceTokens[i]);
+      if (score >= JACCARD_THRESHOLD) scored.push({ idx: i, score });
+    }
+    scored.sort((a, b) => b.score - a.score);
+    if (scored.length === 0) {
+      entries.push({ claim: sentence, source_indices: [], confidence: "low" });
+    } else {
+      entries.push({
+        claim: sentence,
+        source_indices: scored.slice(0, MAX_SOURCES_PER_CLAIM).map((s) => s.idx),
+        confidence: "medium"
+      });
+    }
+  }
+  return entries;
+}
+function extractMarkers(sentence, sourceCount) {
+  const matches = sentence.match(/\[(\d+)\]/g);
+  if (!matches) return [];
+  const indices = [];
+  const seen = /* @__PURE__ */ new Set();
+  for (const m of matches) {
+    const n = Number(m.slice(1, -1));
+    if (!Number.isFinite(n) || n < 1) continue;
+    const idx = n - 1;
+    if (idx >= sourceCount) continue;
+    if (seen.has(idx)) continue;
+    seen.add(idx);
+    indices.push(idx);
+  }
+  return indices;
+}
+function tokenize(text) {
+  const tokens = /* @__PURE__ */ new Set();
+  const words = text.toLowerCase().replace(/[^a-z0-9\s]/g, " ").split(/\s+/);
+  for (const w of words) {
+    if (w.length < MIN_TOKEN_LEN) continue;
+    if (STOPWORDS.has(w)) continue;
+    tokens.add(w);
+  }
+  return tokens;
+}
+function jaccard(a, b) {
+  if (a.size === 0 || b.size === 0) return 0;
+  let intersection = 0;
+  for (const t of a) if (b.has(t)) intersection++;
+  const union = a.size + b.size - intersection;
+  return union === 0 ? 0 : intersection / union;
+}
+export {
+  buildCitationGraph
+};
+//# sourceMappingURL=citation-graph.js.map

package/dist/research/citation-graph.js.map

@@ -0,0 +1 @@
+{"version":3,"sources":["../../src/research/citation-graph.ts"],"sourcesContent":["import type { CitationGraphEntry } from '../types.js';\n\nexport type { CitationGraphEntry };\n\nexport interface CitationGraphSource {\n  url: string;\n  title: string;\n  markdown: string;\n}\n\nconst MAX_ENTRIES = 50;\nconst MAX_SOURCES_PER_CLAIM = 3;\nconst JACCARD_THRESHOLD = 0.2;\nconst MIN_SENTENCE_LEN = 10;\nconst MIN_TOKEN_LEN = 4;\n\nconst STOPWORDS = new Set([\n  'a', 'an', 'the', 'is', 'are', 'was', 'were', 'and', 'or', 'but', 'in', 'on',\n  'at', 'to', 'for', 'of', 'with', 'by', 'from', 'as', 'that', 'this', 'it',\n  'its', 'be', 'been', 'has', 'have', 'had',\n]);\n\nexport function buildCitationGraph(\n  synthesisText: string,\n  sources: CitationGraphSource[],\n): CitationGraphEntry[] {\n  if (!synthesisText || synthesisText.trim().length === 0) return [];\n\n  const sentenceMatches = synthesisText.match(/[^.!?]+[.!?]+(?:\\s|$)/g);\n  const sentences = sentenceMatches && sentenceMatches.length > 0\n    ? sentenceMatches.map((s) => s.trim())\n    : [synthesisText.trim()];\n\n  const sourceTokens = sources.map((s) => tokenize(s.markdown));\n  const entries: CitationGraphEntry[] = [];\n\n  for (const sentence of sentences) {\n    if (entries.length >= MAX_ENTRIES) break;\n    if (sentence.length < MIN_SENTENCE_LEN) continue;\n\n    const markerIndices = extractMarkers(sentence, sources.length);\n    if (markerIndices.length > 0) {\n      entries.push({\n        claim: sentence,\n        source_indices: markerIndices,\n        confidence: 'high',\n      });\n      continue;\n    }\n\n    // No (valid) markers -> Jaccard overlap\n    const sentenceTokens = tokenize(sentence);\n    if (sentenceTokens.size === 0 || sourceTokens.length === 0) {\n      entries.push({ claim: sentence, source_indices: [], confidence: 'low' });\n      continue;\n    }\n\n    const scored: { idx: number; score: number }[] = [];\n    for (let i = 0; i < sourceTokens.length; i++) {\n      const score = jaccard(sentenceTokens, sourceTokens[i]);\n      if (score >= JACCARD_THRESHOLD) scored.push({ idx: i, score });\n    }\n    scored.sort((a, b) => b.score - a.score);\n\n    if (scored.length === 0) {\n      entries.push({ claim: sentence, source_indices: [], confidence: 'low' });\n    } else {\n      entries.push({\n        claim: sentence,\n        source_indices: scored.slice(0, MAX_SOURCES_PER_CLAIM).map((s) => s.idx),\n        confidence: 'medium',\n      });\n    }\n  }\n\n  return entries;\n}\n\nfunction extractMarkers(sentence: string, sourceCount: number): number[] {\n  const matches = sentence.match(/\\[(\\d+)\\]/g);\n  if (!matches) return [];\n  const indices: number[] = [];\n  const seen = new Set<number>();\n  for (const m of matches) {\n    const n = Number(m.slice(1, -1));\n    if (!Number.isFinite(n) || n < 1) continue;\n    const idx = n - 1;\n    if (idx >= sourceCount) continue;\n    if (seen.has(idx)) continue;\n    seen.add(idx);\n    indices.push(idx);\n  }\n  return indices;\n}\n\nfunction tokenize(text: string): Set<string> {\n  const tokens = new Set<string>();\n  const words = text.toLowerCase().replace(/[^a-z0-9\\s]/g, ' ').split(/\\s+/);\n  for (const w of words) {\n    if (w.length < MIN_TOKEN_LEN) continue;\n    if (STOPWORDS.has(w)) continue;\n    tokens.add(w);\n  }\n  return tokens;\n}\n\nfunction jaccard(a: Set<string>, b: Set<string>): number {\n  if (a.size === 0 || b.size === 0) return 0;\n  let intersection = 0;\n  for (const t of a) if (b.has(t)) intersection++;\n  const union = a.size + b.size - intersection;\n  return union === 0 ? 0 : intersection / union;\n}\n"],"mappings":"AAUA,MAAM,cAAc;AACpB,MAAM,wBAAwB;AAC9B,MAAM,oBAAoB;AAC1B,MAAM,mBAAmB;AACzB,MAAM,gBAAgB;AAEtB,MAAM,YAAY,oBAAI,IAAI;AAAA,EACxB;AAAA,EAAK;AAAA,EAAM;AAAA,EAAO;AAAA,EAAM;AAAA,EAAO;AAAA,EAAO;AAAA,EAAQ;AAAA,EAAO;AAAA,EAAM;AAAA,EAAO;AAAA,EAAM;AAAA,EACxE;AAAA,EAAM;AAAA,EAAM;AAAA,EAAO;AAAA,EAAM;AAAA,EAAQ;AAAA,EAAM;AAAA,EAAQ;AAAA,EAAM;AAAA,EAAQ;AAAA,EAAQ;AAAA,EACrE;AAAA,EAAO;AAAA,EAAM;AAAA,EAAQ;AAAA,EAAO;AAAA,EAAQ;AACtC,CAAC;AAEM,SAAS,mBACd,eACA,SACsB;AACtB,MAAI,CAAC,iBAAiB,cAAc,KAAK,EAAE,WAAW,EAAG,QAAO,CAAC;AAEjE,QAAM,kBAAkB,cAAc,MAAM,wBAAwB;AACpE,QAAM,YAAY,mBAAmB,gBAAgB,SAAS,IAC1D,gBAAgB,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,IACnC,CAAC,cAAc,KAAK,CAAC;AAEzB,QAAM,eAAe,QAAQ,IAAI,CAAC,MAAM,SAAS,EAAE,QAAQ,CAAC;AAC5D,QAAM,UAAgC,CAAC;AAEvC,aAAW,YAAY,WAAW;AAChC,QAAI,QAAQ,UAAU,YAAa;AACnC,QAAI,SAAS,SAAS,iBAAkB;AAExC,UAAM,gBAAgB,eAAe,UAAU,QAAQ,MAAM;AAC7D,QAAI,cAAc,SAAS,GAAG;AAC5B,cAAQ,KAAK;AAAA,QACX,OAAO;AAAA,QACP,gBAAgB;AAAA,QAChB,YAAY;AAAA,MACd,CAAC;AACD;AAAA,IACF;AAGA,UAAM,iBAAiB,SAAS,QAAQ;AACxC,QAAI,eAAe,SAAS,KAAK,aAAa,WAAW,GAAG;AAC1D,cAAQ,KAAK,EAAE,OAAO,UAAU,gBAAgB,CAAC,GAAG,YAAY,MAAM,CAAC;AACvE;AAAA,IACF;AAEA,UAAM,SAA2C,CAAC;AAClD,aAAS,IAAI,GAAG,IAAI,aAAa,QAAQ,KAAK;AAC5C,YAAM,QAAQ,QAAQ,gBAAgB,aAAa,CAAC,CAAC;AACrD,UAAI,SAAS,kBAAmB,QAAO,KAAK,EAAE,KAAK,GAAG,MAAM,CAAC;AAAA,IAC/D;AACA,WAAO,KAAK,CAAC,GAAG,MAAM,EAAE,QAAQ,EAAE,KAAK;AAEvC,QAAI,OAAO,WAAW,GAAG;AACvB,cAAQ,KAAK,EAAE,OAAO,UAAU,gBAAgB,CAAC,GAAG,YAAY,MAAM,CAAC;AAAA,IACzE,OAAO;AACL,cAAQ,KAAK;AAAA,QACX,OAAO;AAAA,QACP,gBAAgB,OAAO,MAAM,GAAG,qBAAqB,EAAE,IAAI,CAAC,MAAM,EAAE,GAAG;AAAA,QACvE,YAAY;AAAA,MACd,CAAC;AAAA,IACH;AAAA,EACF;AAEA,SAAO;AACT;AAEA,SAAS,eAAe,UAAkB,aAA+B;AACvE,QAAM,UAAU,SAAS,MAAM,YAAY;AAC3C,MAAI,CAAC,QAAS,QAAO,CAAC;AACtB,QAAM,UAAoB,CAAC;AAC3B,QAAM,OAAO,oBAAI,IAAY;AAC7B,aAAW,KAAK,SAAS;AACvB,UAAM,IAAI,OAAO,EAAE,MAAM,GAAG,EAAE,CAAC;AAC/B,QAAI,CAAC,OAAO,SAAS,CAAC,KAAK,IAAI,EAAG;AAClC,UAAM,MAAM,IAAI;AAChB,QAAI,OAAO,YAAa;AACxB,QAAI,KAAK,IAAI,GAAG,EAAG;AACnB,SAAK,IAAI,GAAG;AACZ,YAAQ,KAAK,GAAG;AAAA,EAClB;AACA,SAAO;AACT;AAEA,SAAS,SAAS,MAA2B;AAC3C,QAAM,SAAS,oBAAI,IAAY;AAC/B,QAAM,QAAQ,KAAK,YAAY,EAAE,QAAQ,gBAAgB,GAAG,EAAE,MAAM,KAAK;AACzE,aAAW,KAAK,OAAO;AACrB,QAAI,EAAE,SAAS,cAAe;AAC9B,QAAI,UAAU,IAAI,CAAC,EAAG;AACtB,WAAO,IAAI,CAAC;AAAA,EACd;AACA,SAAO;AACT;AAEA,SAAS,QAAQ,GAAgB,GAAwB;AACvD,MAAI,EAAE,SAAS,KAAK,EAAE,SAAS,EAAG,QAAO;AACzC,MAAI,eAAe;AACnB,aAAW,KAAK,EAAG,KAAI,EAAE,IAAI,CAAC,EAAG;AACjC,QAAM,QAAQ,EAAE,OAAO,EAAE,OAAO;AAChC,SAAO,UAAU,IAAI,IAAI,eAAe;AAC1C;","names":[]}

package/dist/research/decompose.d.ts

@@ -0,0 +1,14 @@
+import { type SamplingCapableServer } from '../search/sampling.js';
+export type QueryType = 'comparison' | 'how-to' | 'concept' | 'general';
+export interface DecomposeResult {
+    subQueries: string[];
+    samplingUsed: boolean;
+    queryType: QueryType;
+}
+export declare function decomposeQuestion(question: string, depth: 'quick' | 'standard' | 'comprehensive', server?: SamplingCapableServer): Promise<DecomposeResult>;
+export declare function detectQueryType(question: string): QueryType;
+export declare function extractComparisonEntities(question: string): {
+    entities: string[];
+    context: string;
+};
+//# sourceMappingURL=decompose.d.ts.map

package/dist/research/decompose.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"decompose.d.ts","sourceRoot":"","sources":["../../src/research/decompose.ts"],"names":[],"mappings":"AACA,OAAO,EACL,KAAK,qBAAqB,EAG3B,MAAM,uBAAuB,CAAC;AAU/B,MAAM,MAAM,SAAS,GAAG,YAAY,GAAG,QAAQ,GAAG,SAAS,GAAG,SAAS,CAAC;AAExE,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,MAAM,EAAE,CAAC;IACrB,YAAY,EAAE,OAAO,CAAC;IACtB,SAAS,EAAE,SAAS,CAAC;CACtB;AAED,wBAAsB,iBAAiB,CACrC,QAAQ,EAAE,MAAM,EAChB,KAAK,EAAE,OAAO,GAAG,UAAU,GAAG,eAAe,EAC7C,MAAM,CAAC,EAAE,qBAAqB,GAC7B,OAAO,CAAC,eAAe,CAAC,CAuC1B;AA8ED,wBAAgB,eAAe,CAAC,QAAQ,EAAE,MAAM,GAAG,SAAS,CA0B3D;AA8BD,wBAAgB,yBAAyB,CAAC,QAAQ,EAAE,MAAM,GAAG;IAAE,QAAQ,EAAE,MAAM,EAAE,CAAC;IAAC,OAAO,EAAE,MAAM,CAAA;CAAE,CAqCnG"}