@theseus-whale/dsh-continuity 0.0.0-stage → 4.12.2

package/bin/bc-acceptance-full.mjs

@@ -0,0 +1,91 @@
+import fs from 'node:fs';
+import { loadIndexJsonl } from '../src/memory/bc-index.mjs';
+import {
+  createCMap, expandWindow, learnFromExplanation, markAsked, renderNote, retrieve,
+} from '../src/memory/bc-recall.mjs';
+
+const root = 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity';
+const indexPath = `${root}/projects/bc-recall/data/index.jsonl`;
+const sourcePath = `${root}/backups/continuity-tags-20261031/online400/continuity-1500.jsonl`;
+const records = loadIndexJsonl(indexPath);
+const rawRows = fs.readFileSync(sourcePath, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line));
+const rawByTurn = new Map(rawRows.map((row) => [row.turn, row]));
+const relevant = new Set(rawRows.filter((row) => row.facets?.activity === '跑团').map((row) => row.turn));
+const cmap = createCMap();
+
+function inspect(hit) {
+  if (!hit) return null;
+  const row = rawByTurn.get(hit.turn);
+  return {
+    turn: hit.turn,
+    seqStart: hit.seqStart,
+    seqEnd: hit.seqEnd,
+    sourceFacet: row?.facets?.activity ?? null,
+    text: String(row?.text ?? '').replace(/\s+/g, ' ').slice(0, 180),
+  };
+}
+
+function summarize(name, result) {
+  const turns = (result.hits ?? []).map((hit) => hit.turn);
+  const relevantHits = turns.filter((turn) => relevant.has(turn)).length;
+  return {
+    name,
+    kind: result.kind,
+    via: result.via ?? null,
+    hits: turns.length,
+    relevantHits,
+    extraHits: turns.length - relevantHits,
+    recall: relevant.size === 0 ? 0 : Number((relevantHits / relevant.size).toFixed(3)),
+    note: renderNote(result),
+    samples: (result.hits ?? []).slice(0, 3).map(inspect),
+  };
+}
+
+const out = { records: records.length, relevantTurns: relevant.size, cases: [] };
+
+// 1. B direct game
+out.cases.push(summarize('游戏 B direct', retrieve({ records, cmap, query: '游戏', minConceptScore: 0.35 })));
+
+// 2. Cold start + once
+const cold = retrieve({ records, cmap, query: '跑团' });
+out.cases.push(summarize('跑团 cold', cold));
+markAsked(cmap, '跑团');
+out.cases.push(summarize('跑团 after ask', retrieve({ records, cmap, query: '跑团' })));
+
+// 3. Learn C and retrieve
+const learned = learnFromExplanation(cmap, '跑团', '就是游戏、故事、对话', { records });
+const retrieved = retrieve({ records: learned.records, cmap, query: '跑团', minConceptScore: 0.35 });
+const window = expandWindow(learned.records, retrieved.hits?.[0], { maxTurns: 3 });
+out.cases.push({
+  ...summarize('跑团 after C', retrieved),
+  window,
+  windowSeqs: window.turns.map((turn) => rawByTurn.get(turn)?.seqStart ?? null),
+});
+
+// 4. Technical concept
+out.cases.push(summarize('技术 B direct', retrieve({ records, cmap, query: '技术', minConceptScore: 0.35 })));
+
+// 5. Unseen concept + synthetic record
+const synthetic = {
+  turn: 9999, seqStart: 999900, seqEnd: 999900,
+  generic: ['病毒', '公共卫生'], terms: ['COVID-19'],
+  conceptScores: { 病毒: 0.61, 公共卫生: 0.55 }, confidence: 0.61, status: 'weak',
+};
+const withSynthetic = [...learned.records, synthetic];
+const virusCold = retrieve({ records: withSynthetic, cmap, query: 'COVID-19' });
+out.cases.push(summarize('COVID-19 cold', virusCold));
+markAsked(cmap, 'COVID-19');
+learnFromExplanation(cmap, 'COVID-19', '就是病毒、公共卫生', { records: withSynthetic });
+const virusAfter = retrieve({ records: withSynthetic, cmap, query: 'COVID-19', minConceptScore: 0.35 });
+out.cases.push({
+  ...summarize('COVID-19 after C', virusAfter),
+  syntheticText: 'COVID-19 起源报告，实验室泄漏，自然宿主，冷链传播。',
+});
+
+// 6. Truly unknown term
+out.cases.push(summarize('量子计算 cold', retrieve({ records, cmap, query: '量子计算' })));
+markAsked(cmap, '量子计算');
+out.cases.push(summarize('量子计算 after ask', retrieve({ records, cmap, query: '量子计算' })));
+
+fs.writeFileSync(`${root}/projects/bc-recall/acceptance-full.json`, JSON.stringify(out, null, 2));
+console.log(JSON.stringify(out, null, 2));

package/bin/bc-acceptance-tight.mjs

@@ -0,0 +1,66 @@
+import fs from 'node:fs';
+import { loadIndexJsonl } from '../src/memory/bc-index.mjs';
+import {
+  createCMap, learnFromExplanation, markAsked, renderNote, retrieve,
+} from '../src/memory/bc-recall.mjs';
+
+const root = 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity';
+const records = loadIndexJsonl(`${root}/projects/bc-recall/data/index.jsonl`);
+const rows = fs.readFileSync(`${root}/backups/continuity-tags-20261031/online400/continuity-1500.jsonl`, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line));
+const rawByTurn = new Map(rows.map((row) => [row.turn, row]));
+const relevant = new Set(rows.filter((row) => row.facets?.activity === '跑团').map((row) => row.turn));
+const cmap = createCMap();
+
+function sample(turn) {
+  const row = rawByTurn.get(turn);
+  return row === undefined ? null : {
+    turn,
+    seqStart: row.seqStart,
+    sourceFacet: row.facets?.activity ?? null,
+    text: String(row.text).replace(/\s+/g, ' ').slice(0, 120),
+  };
+}
+
+function caseOf(name, result) {
+  const turns = (result.hits ?? []).map((hit) => hit.turn);
+  return {
+    name,
+    kind: result.kind,
+    via: result.via ?? null,
+    hits: turns.length,
+    relevantHits: turns.filter((turn) => relevant.has(turn)).length,
+    extraHits: turns.filter((turn) => !relevant.has(turn)).length,
+    note: renderNote(result),
+    seedSamples: turns.slice(0, 3).map(sample),
+  };
+}
+
+const out = { records: records.length, relevantTurns: relevant.size, cases: [] };
+out.cases.push(caseOf('游戏 B direct', retrieve({ records, cmap, query: '游戏', minConceptScore: 0.35 })));
+out.cases.push(caseOf('跑团 cold', retrieve({ records, cmap, query: '跑团' })));
+markAsked(cmap, '跑团');
+out.cases.push(caseOf('跑团 after ask', retrieve({ records, cmap, query: '跑团' })));
+const learned = learnFromExplanation(cmap, '跑团', '就是游戏、故事、对话', { records });
+const run = retrieve({ records: learned.records, cmap, query: '跑团', minConceptScore: 0.35 });
+out.cases.push(caseOf('跑团 after C', run));
+// Only an explicit LLM read, not an automatic expansion.
+const seedTurn = run.hits?.[0]?.turn;
+if (seedTurn !== undefined) {
+  const turns = [];
+  for (let turn = seedTurn - 3; turn <= seedTurn + 3; turn += 1) {
+    if (rawByTurn.has(turn)) turns.push(turn);
+  }
+  out.llmRead = { seedTurn, chosenTurns: turns, samples: turns.map(sample) };
+}
+out.cases.push(caseOf('技术 B direct', retrieve({ records, cmap, query: '技术', minConceptScore: 0.35 })));
+const synthetic = { turn: 9999, seqStart: 999900, seqEnd: 999900, generic: ['病毒', '公共卫生'], terms: ['COVID-19'], keywords: ['COVID-19', '实验室泄漏', '自然宿主', '冷链传播'], conceptScores: { 病毒: 0.61, 公共卫生: 0.55 }, confidence: 0.61, status: 'weak' };
+const withSynthetic = [...learned.records, synthetic];
+out.cases.push(caseOf('COVID-19 cold', retrieve({ records: withSynthetic, cmap, query: 'COVID-19' })));
+markAsked(cmap, 'COVID-19');
+learnFromExplanation(cmap, 'COVID-19', '就是病毒、公共卫生', { records: withSynthetic });
+out.cases.push(caseOf('COVID-19 after C', retrieve({ records: withSynthetic, cmap, query: 'COVID-19', minConceptScore: 0.35 })));
+out.cases.push(caseOf('量子计算 cold', retrieve({ records, cmap, query: '量子计算' })));
+markAsked(cmap, '量子计算');
+out.cases.push(caseOf('量子计算 after ask', retrieve({ records, cmap, query: '量子计算' })));
+fs.writeFileSync(`${root}/projects/bc-recall/acceptance-tight.json`, JSON.stringify(out, null, 2));
+console.log(JSON.stringify(out, null, 2));

package/bin/bc-acceptance.mjs

@@ -0,0 +1,22 @@
+import fs from 'node:fs';
+import { loadBCStore } from '../src/memory/bc-store.mjs';
+import { learnFromExplanation, retrieve, renderNote } from '../src/memory/bc-recall.mjs';
+
+const file = process.argv[2] ?? 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/projects/bc-recall/data/store.json';
+const store = loadBCStore(file);
+const before = {};
+for (const query of ['游戏', '跑团', '病毒']) {
+  const result = retrieve({ records: store.records, cmap: store.cmap, query, minConceptScore: 0.35 });
+  before[query] = { kind: result.kind, via: result.via ?? null, hits: result.hits?.length ?? 0, note: renderNote(result) };
+}
+const learned = learnFromExplanation(store.cmap, '跑团', '就是游戏、故事、对话', { records: store.records });
+const after = retrieve({ records: learned.records, cmap: store.cmap, query: '跑团', minConceptScore: 0.35 });
+const report = {
+  store: file,
+  records: store.records.length,
+  before,
+  learned: { ok: learned.ok, generic: learned.generic },
+  after: { kind: after.kind, via: after.via, hits: after.hits?.length ?? 0, note: renderNote(after) },
+};
+fs.writeFileSync('C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/projects/bc-recall/acceptance.json', JSON.stringify(report, null, 2));
+console.log(JSON.stringify(report, null, 2));

package/bin/bc-add-keywords.mjs

@@ -0,0 +1,21 @@
+import fs from 'node:fs';
+import { extractKeywords } from '../src/memory/bc-recall.mjs';
+import { loadSpecialKeywords, matchSpecialKeywords } from '../src/memory/special-keywords.mjs';
+import { loadIndexJsonl, saveIndexJsonl } from '../src/memory/bc-index.mjs';
+
+const root = 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity';
+const indexFile = `${root}/projects/bc-recall/data/index.jsonl`;
+const sourceFile = `${root}/backups/continuity-tags-20261031/online400/continuity-1500.jsonl`;
+const records = loadIndexJsonl(indexFile);
+const rows = fs.readFileSync(sourceFile, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line));
+const textByTurn = new Map(rows.map((row) => [row.turn, row.text]));
+const specialKeywords = loadSpecialKeywords();
+for (const record of records) {
+  const text = textByTurn.get(record.turn) ?? '';
+  record.keywords = [...new Set([
+    ...extractKeywords(text, { limit: 12 }),
+    ...matchSpecialKeywords(text, specialKeywords),
+  ])];
+}
+saveIndexJsonl(indexFile, records);
+console.log(JSON.stringify({ records: records.length, sample: records.slice(0, 2).map((record) => ({ turn: record.turn, keywords: record.keywords })) }, null, 2));

package/bin/bc-demo.mjs

@@ -0,0 +1,47 @@
+import fs from 'node:fs';
+import { backfillRecords, createCMap, learnTerm, makeRecord, renderNote, retrieve } from '../src/memory/bc-recall.mjs';
+
+const rows = fs.readFileSync('C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/backups/continuity-tags-20261031/online400/continuity-1500.jsonl', 'utf8')
+  .split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line))
+  .filter((row) => row.facets?.activity === '跑团');
+
+const genericOf = (facets) => {
+  const out = [];
+  if (facets.activity === '跑团') out.push('游戏');
+  if (facets.activity === '对话') out.push('对话');
+  if (['开发', '调试', '测试', '文档'].includes(facets.activity)) out.push('技术');
+  if (facets.topic === '剧情') out.push('故事');
+  if (['技术', '设备', '记忆', '规则'].includes(facets.topic)) out.push(facets.topic);
+  return [...new Set(out)];
+};
+
+let records = rows.map((row) => makeRecord({
+  turn: row.turn,
+  seqStart: row.seqStart,
+  seqEnd: row.seqEnd,
+  text: row.text,
+  generic: genericOf(row.facets),
+  terms: row.facets.activity === '跑团' ? ['跑团'] : [],
+}));
+records = backfillRecords(records, '跑团', ['游戏', '故事']);
+records.push(makeRecord({
+  turn: 900,
+  seqStart: 5000,
+  seqEnd: 5001,
+  text: 'COVID-19 起源报告，实验室泄漏，自然宿主，冷链传播。',
+  generic: ['病毒', '公共卫生'],
+  terms: ['COVID-19', '实验室泄漏'],
+  status: 'weak',
+  confidence: 0.33,
+}));
+
+const cmap = createCMap();
+const out = [];
+for (const query of ['游戏', '跑团', '病毒', 'COVID-19', '疫情']) {
+  const result = retrieve({ records, cmap, query });
+  out.push({ query, kind: result.kind, via: result.via ?? null, note: renderNote(result), hits: result.hits?.length ?? 0 });
+}
+learnTerm(cmap, '跑团', ['游戏', '故事', '对话']);
+const learned = retrieve({ records, cmap, query: '跑团' });
+out.push({ query: '跑团(after C)', kind: learned.kind, via: learned.via, note: renderNote(learned), hits: learned.hits?.length ?? 0 });
+console.log(JSON.stringify({ records: records.length, examples: out }, null, 2));

package/bin/bc-expand-hint.mjs

@@ -0,0 +1,179 @@
+#!/usr/bin/env node
+/**
+ * Offline expand-hint reminder tool.
+ * Reads an index + turn texts, clusters, and prints candidate topic clusters.
+ * Never writes concepts/index. --write only appends to expand-hint.jsonl.
+ */
+import fs from 'node:fs';
+import path from 'node:path';
+import os from 'node:os';
+import { embedTexts } from '../src/facets/embedding.mjs';
+import { extractKeywords } from '../src/memory/bc-recall.mjs';
+import { loadIndexJsonl } from '../src/memory/bc-index.mjs';
+import { loadConcepts } from '../src/memory/concepts-file.mjs';
+
+function arg(name, fallback = null) {
+  const at = process.argv.indexOf(`--${name}`);
+  return at >= 0 ? process.argv[at + 1] ?? fallback : fallback;
+}
+const has = (name) => process.argv.includes(`--${name}`);
+const indexFile = arg('index', path.join(os.homedir(), '.dsh', 'continuity', 'bc-index.jsonl'));
+const conceptsFile = arg('concepts', path.join(os.homedir(), '.dsh', 'continuity', 'concepts.md'));
+const turnsFile = arg('turns');
+const outFile = arg('out', path.join(os.homedir(), '.dsh', 'continuity', 'expand-hint.jsonl'));
+const verdict = arg('record-verdict');
+if (verdict) {
+  if (!['accepted', 'rejected'].includes(verdict)) {
+    console.error('--record-verdict must be accepted or rejected');
+    process.exit(2);
+  }
+  const event = {
+    at: new Date().toISOString(),
+    event: 'verdict',
+    by: 'cli',
+    verdict,
+    auditAt: arg('audit-at', ''),
+    reason: arg('reason', ''),
+  };
+  fs.mkdirSync(path.dirname(outFile), { recursive: true });
+  fs.appendFileSync(outFile, `${JSON.stringify(event)}\n`, 'utf8');
+  console.log(`recorded ${verdict}`);
+  process.exit(0);
+}
+if (!turnsFile) {
+  console.error('missing --turns <turns.jsonl>');
+  process.exit(2);
+}
+const until = Number(arg('until', Infinity));
+const records = loadIndexJsonl(indexFile).filter((record) => !Number.isFinite(until) || record.turn <= until);
+if (records.length === 0) {
+  console.error('index has no records');
+  process.exit(2);
+}
+const rows = loadConcepts(conceptsFile);
+if (rows.length === 0) {
+  console.error('concepts.md has no valid rows');
+  process.exit(2);
+}
+const concepts = rows.map((row) => row.label);
+const textByTurn = new Map(fs.readFileSync(turnsFile, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => {
+  const row = JSON.parse(line);
+  return [row.turn, row.text];
+}));
+const missing = records.filter((record) => !textByTurn.has(record.turn));
+if (missing.length > 0) {
+  console.error(`missing turn text for ${missing.length} records: ${missing.slice(0, 8).map((r) => r.turn).join(', ')}`);
+  process.exit(3);
+}
+
+const N = records.length;
+const k = Number(arg('k', Math.max(20, Math.min(60, Math.round(N / 50))))) || 30;
+const minSize = Number(arg('min-size', 10)) || 10;
+const topN = Number(arg('top', 5)) || 5;
+const maxTopic = Number(arg('max-topic', 0.5));
+
+const norm = (v) => {
+  const n = Math.sqrt(v.reduce((sum, x) => sum + x * x, 0)) || 1;
+  return v.map((x) => x / n);
+};
+const cos = (a, b) => {
+  let dot = 0;
+  for (let i = 0; i < Math.min(a.length, b.length); i += 1) dot += a[i] * b[i];
+  return dot;
+};
+const texts = records.map((record) => String(textByTurn.get(record.turn) ?? ''));
+const vectors = [];
+for (let i = 0; i < texts.length; i += 16) vectors.push(...await embedTexts(texts.slice(i, i + 16)));
+const points = records.map((record, i) => ({ turn: record.turn, v: norm(vectors[i]), record }));
+
+// chatterScore
+const df = new Map();
+const tokensByTurn = new Map();
+for (const [turn, text] of textByTurn) {
+  const tokens = [...String(text).matchAll(/[\u4e00-\u9fff]{2,6}/g)].map((m) => m[0]);
+  tokensByTurn.set(turn, tokens);
+  for (const token of new Set(tokens)) df.set(token, (df.get(token) ?? 0) + 1);
+}
+const functionRe = /[浜嗙殑鐫€鍚楀憿鍚у晩]/g;
+const chatterScore = (turn) => {
+  const text = String(textByTurn.get(turn) ?? '');
+  const tokens = tokensByTurn.get(turn) ?? [];
+  const cjkTotal = tokens.length;
+  const proper = tokens.filter((token) => (df.get(token) ?? 0) <= 3).length;
+  const properDensity = cjkTotal ? proper / cjkTotal : 0;
+  const fnCount = (text.match(functionRe) ?? []).length;
+  const fnDensity = fnCount / Math.max(1, text.length);
+  const keywordSparse = extractKeywords(text, { limit: 10 }).length < 3 ? 1 : 0;
+  const quotes = (text.match(/[「」“”"——]/g) ?? []).length;
+  return 0.30 * (text.length < 60 ? 1 : 0)
+    + 0.25 * Math.min(1, properDensity)
+    + 0.20 * Math.min(1, fnDensity * 20)
+    + 0.15 * keywordSparse
+    + 0.10 * (quotes > 0 ? 1 : 0);
+};
+
+// Deterministic k-means (seeded init, cosine).
+let seed = 42;
+const rand = () => {
+  seed = (seed * 1664525 + 1013904223) >>> 0;
+  return seed / 2 ** 32;
+};
+const centers = [];
+const used = new Set();
+while (centers.length < k) {
+  const index = Math.floor(rand() * points.length);
+  if (used.has(index)) continue;
+  used.add(index);
+  centers.push(points[index].v);
+}
+let groups = [];
+for (let iteration = 0; iteration < 15; iteration += 1) {
+  groups = centers.map((center) => ({ center, members: [] }));
+  for (const point of points) {
+    let best = 0;
+    let bestScore = -Infinity;
+    for (let c = 0; c < centers.length; c += 1) {
+      const score = cos(point.v, centers[c]);
+      if (score > bestScore) { bestScore = score; best = c; }
+    }
+    groups[best].members.push(point);
+  }
+  for (let c = 0; c < centers.length; c += 1) {
+    if (groups[c].members.length === 0) continue;
+    const sum = new Array(points[0].v.length).fill(0);
+    for (const point of groups[c].members) for (let i = 0; i < sum.length; i += 1) sum[i] += point.v[i];
+    centers[c] = norm(sum);
+  }
+}
+const clusters = groups.filter((group) => group.members.length >= minSize).map((group) => {
+  const turns = group.members.map((point) => point.turn);
+  const topicMax = group.members.reduce((sum, point) => {
+    const scores = point.record.conceptScores ?? {};
+    return sum + Math.max(0, ...concepts.map((name) => scores[name] ?? 0));
+  }, 0) / group.members.length;
+  const chatter = group.members.reduce((sum, point) => sum + chatterScore(point.turn), 0) / group.members.length;
+  const kw = [...new Set(extractKeywords(turns.map((turn) => textByTurn.get(turn) ?? '').join('\n'), { limit: 30 })
+    .map((word) => String(word).toLowerCase()))].filter((word) => word.length >= 2).slice(0, 6);
+  return { turns, size: turns.length, start: Math.min(...turns), end: Math.max(...turns), topicMax, chatter, kw };
+});
+const candidates = clusters.filter((cluster) => cluster.topicMax < maxTopic)
+  .sort((a, b) => a.topicMax - b.topicMax || b.size - a.size)
+  .slice(0, topN);
+if (candidates.length === 0) {
+  console.log('no candidates');
+  process.exit(0);
+}
+for (const item of candidates) {
+  console.log(`turn ${item.start}-${item.end} - ${item.size}轮 - topicMax ${item.topicMax.toFixed(2)} - chatter ${item.chatter.toFixed(2)} - kw ${item.kw.join('/')}`);
+  console.log('    → topicMax<0.5 且 kw 是话题词（非人名/道具/英文标识符）才考虑加；加就一步：concepts.md 追加 `名字 | 词1、词2… | 覆盖估算`');
+}
+if (has('write')) {
+  const at = new Date().toISOString();
+  const lines = candidates.map((item) => JSON.stringify({
+    at, event: 'pending', range: [item.start, item.end], size: item.size,
+    topicMax: Number(item.topicMax.toFixed(2)), chatter: Number(item.chatter.toFixed(2)), kw: item.kw,
+  }));
+  fs.mkdirSync(path.dirname(outFile), { recursive: true });
+  fs.appendFileSync(outFile, `${lines.join('\n')}\n`, 'utf8');
+}
+

package/bin/bc-ingest-index.mjs

@@ -0,0 +1,35 @@
+import fs from 'node:fs';
+import { embedTexts } from '../src/facets/embedding.mjs';
+import { createBLabeler } from '../src/memory/bc-label.mjs';
+import { makeRecord } from '../src/memory/bc-recall.mjs';
+import { loadSpecialKeywords, matchSpecialKeywords } from '../src/memory/special-keywords.mjs';
+import { saveIndexJsonl } from '../src/memory/bc-index.mjs';
+
+const source = 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/backups/continuity-tags-20261031/online400/continuity-1500.jsonl';
+const target = 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/projects/bc-recall/data/index.jsonl';
+const rows = fs.readFileSync(source, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line));
+const specialKeywords = loadSpecialKeywords();
+const labeler = createBLabeler({ embed: embedTexts });
+const records = [];
+for (const row of rows) {
+  if (!Number.isInteger(row.turn) || !Number.isInteger(row.seqStart)) continue;
+  const label = await labeler.label(row.text, { topK: 3, minScore: 0.25 });
+  const generic = label.candidates.map((candidate) => candidate.concept);
+  const conceptScores = Object.fromEntries(label.candidates.map((candidate) => [candidate.concept, candidate.score]));
+  const terms = [];
+  if (row.facets?.activity === '跑团') terms.push('跑团');
+  if (/COVID|实验室泄漏|自然宿主|冷链传播/.test(row.text)) terms.push('COVID-19');
+  records.push(makeRecord({
+    turn: row.turn,
+    seqStart: row.seqStart,
+    seqEnd: row.seqEnd,
+    generic,
+    terms,
+    keywords: matchSpecialKeywords(row.text, specialKeywords),
+    conceptScores,
+    confidence: label.top?.score ?? 0,
+    status: 'weak',
+  }));
+}
+saveIndexJsonl(target, records);
+console.log(JSON.stringify({ source, target, rows: rows.length, records: records.length }, null, 2));

package/bin/bc-ingest.mjs

@@ -0,0 +1,31 @@
+import fs from 'node:fs';
+import { embedTexts } from '../src/facets/embedding.mjs';
+import { createBLabeler } from '../src/memory/bc-label.mjs';
+import { makeRecord } from '../src/memory/bc-recall.mjs';
+import { emptyStore, saveBCStore } from '../src/memory/bc-store.mjs';
+
+const source = process.argv[2] ?? 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/backups/continuity-tags-20261031/online400/continuity-1500.jsonl';
+const target = process.argv[3] ?? 'C:/Users/35794/Documents/DeepSeek\'s Sandbox/coopanion-continuity/projects/bc-recall/data/store.json';
+const rows = fs.readFileSync(source, 'utf8').split(/\r?\n/).filter(Boolean).map((line) => JSON.parse(line))
+  .filter((row) => row.facets?.activity === '跑团');
+const labeler = createBLabeler({ embed: embedTexts });
+const store = emptyStore();
+for (const row of rows) {
+  const label = await labeler.label(row.text, { topK: 3, minScore: 0.25 });
+  const generic = label.candidates.map((candidate) => candidate.concept);
+  const conceptScores = Object.fromEntries(label.candidates.map((candidate) => [candidate.concept, candidate.score]));
+  const terms = ['跑团'];
+  store.records.push(makeRecord({
+    turn: row.turn,
+    seqStart: row.seqStart,
+    seqEnd: row.seqEnd,
+    text: row.text,
+    generic,
+    terms,
+    confidence: label.top?.score ?? 0,
+    conceptScores,
+    status: 'weak',
+  }));
+}
+saveBCStore(target, store);
+console.log(JSON.stringify({ source, target, records: store.records.length, sample: store.records.slice(0, 3) }, null, 2));

package/bin/bc-label-demo.mjs

@@ -0,0 +1,18 @@
+import { embedTexts } from '../src/facets/embedding.mjs';
+import { createBLabeler } from '../src/memory/bc-label.mjs';
+
+const labeler = createBLabeler({ embed: embedTexts });
+for (const text of [
+  '我把七把钥匙收进兜里，走到门口，用鞋尖量了量潮线到船位的距离。',
+  '我昨天看了COVID-19的起源报告，病毒可能来自实验室泄漏，也提到自然宿主和冷链传播。',
+  '我们接下来去跑团吧。',
+  '帮我看一下这个函数的 bug。',
+]) {
+  const result = await labeler.label(text);
+  console.log(JSON.stringify({
+    text: text.slice(0, 40),
+    top: result.top,
+    candidates: result.candidates,
+    unknown: result.unknown,
+  }, null, 2));
+}

package/bin/bc-relabel.mjs

@@ -0,0 +1,150 @@
+#!/usr/bin/env node
+/**
+ * One-shot concept migration: re-label every turn in an existing index using
+ * concepts.md, write a NEW file, and validate before replacing anything.
+ *
+ * Usage:
+ *   node bin/bc-relabel.mjs --index <bc-index.jsonl> --turns <turns.jsonl> \
+ *     --concepts <concepts.md> --out <new-index.jsonl> [--write]
+ *
+ * turns.jsonl: one JSON object per line: { "turn": number, "text": string }
+ * The tool never edits the source index. Without --write it only reports.
+ */
+import fs from 'node:fs';
+import path from 'node:path';
+import { createBLabeler } from '../src/memory/bc-label.mjs';
+import { embedTexts } from '../src/facets/embedding.mjs';
+import { loadIndexJsonl, saveIndexJsonl } from '../src/memory/bc-index.mjs';
+import { loadConcepts } from '../src/memory/concepts-file.mjs';
+import { buildDf, chatterScore } from '../src/memory/chatter.mjs';
+
+function arg(name, fallback = null) {
+  const index = process.argv.indexOf(`--${name}`);
+  return index >= 0 ? process.argv[index + 1] ?? fallback : fallback;
+}
+function has(name) {
+  return process.argv.includes(`--${name}`);
+}
+
+const indexFile = arg('index');
+const turnsFile = arg('turns');
+const conceptsFile = arg('concepts');
+const outFile = arg('out');
+if (!indexFile || !turnsFile || !outFile) {
+  console.error('missing --index / --turns / --out');
+  process.exit(2);
+}
+const rows = loadConcepts(conceptsFile ?? undefined);
+console.log(JSON.stringify({ concepts: rows.length, names: rows.map((row) => row.label) }, null, 2));
+
+const labels = Object.fromEntries(rows.map((row) => [row.label, row.description]));
+const turns = new Map();
+for (const line of fs.readFileSync(turnsFile, 'utf8').split(/\r?\n/)) {
+  if (!line.trim()) continue;
+  const item = JSON.parse(line);
+  if (Number.isInteger(item.turn) && typeof item.text === 'string') turns.set(item.turn, item.text);
+}
+const records = loadIndexJsonl(indexFile);
+const missing = records.filter((record) => !turns.has(record.turn)).map((record) => record.turn);
+if (missing.length > 0) {
+  console.error(`missing turn text for ${missing.length} records: ${missing.slice(0, 10).join(', ')}`);
+  process.exit(3);
+}
+if (has('vectors')) {
+  const df = buildDf(records.map((record) => turns.get(record.turn) ?? ''));
+  const next = [];
+  for (let i = 0; i < records.length; i += 16) {
+    const batch = records.slice(i, i + 16);
+    const todo = batch.filter((record) => typeof record.vec !== 'string' || !record.vec);
+    const vectors = todo.length > 0 ? await embedTexts(todo.map((record) => turns.get(record.turn) ?? '')) : [];
+    let cursor = 0;
+    for (const record of batch) {
+      if (typeof record.vec === 'string' && record.vec) {
+        next.push(record);
+      } else {
+        const vector = vectors[cursor++];
+        next.push({
+          ...record,
+          vec: Buffer.from(new Float32Array(vector).buffer).toString('base64'),
+          chatter: Number(chatterScore(turns.get(record.turn) ?? '', df).toFixed(4)),
+        });
+      }
+    }
+  }
+  const bad = next.filter((record) => {
+    const buffer = Buffer.from(record.vec ?? '', 'base64');
+    return !buffer.length || !(record.chatter >= 0 && record.chatter <= 1);
+  });
+  console.log(JSON.stringify({ records: next.length, existing: records.length - next.filter((record) => !record.vec).length, bad: bad.length }, null, 2));
+  if (bad.length > 0) process.exit(4);
+  if (!has('write')) {
+    console.log('dry-run only; add --write to emit the vector index');
+    process.exit(0);
+  }
+  saveIndexJsonl(outFile, next);
+  console.log(`wrote ${outFile}`);
+  process.exit(0);
+}
+if (rows.length === 0) {
+  console.error('concepts.md has no valid rows');
+  process.exit(2);
+}
+const labeler = createBLabeler({ embed: embedTexts, labels, concepts: Object.keys(labels) });
+const checks = [
+  [1323, '跑团叙事'],
+  [6, '插件开发'],
+  [866, '线程标签'],
+  [1584, '手机与流量'],
+  [1598, '权限与通知'],
+  [1612, '权限与通知'],
+  [827, '补丁与验证'],
+  [827, '备份与重启'],
+  [866, '关键词与标签'],
+];
+const next = [];
+for (const record of records) {
+  const label = await labeler.label(turns.get(record.turn), { topK: Object.keys(labels).length, minScore: 0 });
+  const scores = Object.fromEntries(label.candidates.map((candidate) => [candidate.concept, candidate.score]));
+  const generic = label.candidates.filter((candidate) => candidate.score >= 0.25).slice(0, 3).map((candidate) => candidate.concept);
+  next.push({
+    ...record,
+    generic,
+    conceptScores: scores,
+    confidence: label.top?.score ?? 0,
+    status: 'weak',
+  });
+}
+const checkResults = checks.map(([turn, concept]) => {
+  const record = next.find((item) => item.turn === turn);
+  if (!record) return { turn, concept, missing: true };
+  const ranked = Object.entries(record.conceptScores ?? {}).sort((a, b) => b[1] - a[1]);
+  const rank = ranked.findIndex(([name]) => name === concept) + 1;
+  return {
+    turn,
+    concept,
+    textLength: String(turns.get(turn) ?? '').length,
+    score: record.conceptScores?.[concept] ?? null,
+    rank,
+    top3: ranked.slice(0, 3).map(([name, score]) => ({ name, score: Number(score.toFixed(3)) })),
+    vector: turn === 827 ? ranked.map(([name, score]) => ({ name, score: Number(score.toFixed(3)) })) : undefined,
+  };
+});
+const bad = next.filter((record) => (record.generic ?? []).some((name) => !labels[name]));
+if (bad.length > 0) {
+  console.error(`records with unknown concept names: ${bad.length}`);
+  process.exit(4);
+}
+console.log(JSON.stringify({ index: indexFile, records: records.length, out: outFile, missing: 0, bad: 0, checks: checkResults }, null, 2));
+if (!has('write')) {
+  console.log('dry-run only; add --write to emit the new index');
+  process.exit(0);
+}
+const temp = `${outFile}.tmp`;
+saveIndexJsonl(temp, next);
+const parsed = loadIndexJsonl(temp);
+if (parsed.length !== records.length) {
+  console.error(`validation failed: wrote ${parsed.length}, expected ${records.length}`);
+  process.exit(5);
+}
+fs.renameSync(temp, outFile);
+console.log(`wrote ${outFile}`);