spectra-cli 4.1.1 → 4.3.0

package/dist/scaffold-kb/recall-eval.d.ts

@@ -0,0 +1,48 @@
+/**
+ * F190 FR-015 — recall@k 机械判定（基于冻结 manifest + searchKbCore）
+ *
+ * 反过拟合：判定只看 query → top-k docId 是否命中 expected_doc_ids，
+ * 无任何按 query 文本的特例分支。systematic zero-recall（命中 0 且目标存在）= BLOCKER。
+ */
+import type { SqliteDb } from './sqlite-engine.js';
+export type RecallCategory = 'chinese_word' | 'mixed' | 'api_symbol' | 'error_code' | 'synonym';
+export interface RecallEntry {
+    id: string;
+    query: string;
+    fixture: 'zh' | 'en';
+    category: RecallCategory;
+    /** 标准答案文档 id 集；null = 占位（未冻结），跳过判定 */
+    expected_doc_ids: string[] | null;
+    expected_chunk_ids?: string[] | null;
+}
+export interface RecallManifest {
+    manifest_version: string;
+    created?: string;
+    description?: string;
+    entries: RecallEntry[];
+}
+export interface EntryOutcome {
+    id: string;
+    category: RecallCategory;
+    hit: boolean;
+    /** systematic zero-recall：返回 0 结果但 expected 文档确实在库中 → BLOCKER */
+    blocker: boolean;
+    skipped: boolean;
+}
+export interface CategoryRecall {
+    category: RecallCategory;
+    total: number;
+    hits: number;
+    recall: number;
+}
+export interface RecallReport {
+    byCategory: CategoryRecall[];
+    outcomes: EntryOutcome[];
+    blockers: string[];
+}
+/**
+ * 对单库（fixture 对应的 vendor DB）跑整份 manifest，按 category 计 recall@k。
+ * @param dbFor 按 fixture 返回对应已加载 DB（zh/en 各一）
+ */
+export declare function computeRecall(dbFor: (fixture: 'zh' | 'en') => SqliteDb, manifest: RecallManifest, k: number): RecallReport;
+//# sourceMappingURL=recall-eval.d.ts.map

package/dist/scaffold-kb/recall-eval.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"recall-eval.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/recall-eval.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAInD,MAAM,MAAM,cAAc,GACtB,cAAc,GACd,OAAO,GACP,YAAY,GACZ,YAAY,GACZ,SAAS,CAAC;AAEd,MAAM,WAAW,WAAW;IAC1B,EAAE,EAAE,MAAM,CAAC;IACX,KAAK,EAAE,MAAM,CAAC;IACd,OAAO,EAAE,IAAI,GAAG,IAAI,CAAC;IACrB,QAAQ,EAAE,cAAc,CAAC;IACzB,sCAAsC;IACtC,gBAAgB,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAClC,kBAAkB,CAAC,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;CACtC;AAED,MAAM,WAAW,cAAc;IAC7B,gBAAgB,EAAE,MAAM,CAAC;IACzB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,WAAW,EAAE,CAAC;CACxB;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,MAAM,CAAC;IACX,QAAQ,EAAE,cAAc,CAAC;IACzB,GAAG,EAAE,OAAO,CAAC;IACb,iEAAiE;IACjE,OAAO,EAAE,OAAO,CAAC;IACjB,OAAO,EAAE,OAAO,CAAC;CAClB;AAED,MAAM,WAAW,cAAc;IAC7B,QAAQ,EAAE,cAAc,CAAC;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,YAAY;IAC3B,UAAU,EAAE,cAAc,EAAE,CAAC;IAC7B,QAAQ,EAAE,YAAY,EAAE,CAAC;IACzB,QAAQ,EAAE,MAAM,EAAE,CAAC;CACpB;AAQD;;;GAGG;AACH,wBAAgB,aAAa,CAC3B,KAAK,EAAE,CAAC,OAAO,EAAE,IAAI,GAAG,IAAI,KAAK,QAAQ,EACzC,QAAQ,EAAE,cAAc,EACxB,CAAC,EAAE,MAAM,GACR,YAAY,CAsCd"}

package/dist/scaffold-kb/recall-eval.js

@@ -0,0 +1,55 @@
+/**
+ * F190 FR-015 — recall@k 机械判定（基于冻结 manifest + searchKbCore）
+ *
+ * 反过拟合：判定只看 query → top-k docId 是否命中 expected_doc_ids，
+ * 无任何按 query 文本的特例分支。systematic zero-recall（命中 0 且目标存在）= BLOCKER。
+ */
+import { queryRows } from './sqlite-engine.js';
+import { searchKbCore } from './search-core.js';
+/** 某 docId 是否存在于库（用于 zero-recall BLOCKER 判定） */
+function docExists(db, docId) {
+    const rows = queryRows(db, 'SELECT 1 FROM chunk_meta WHERE doc_id = ? LIMIT 1', [docId]);
+    return rows.length > 0;
+}
+/**
+ * 对单库（fixture 对应的 vendor DB）跑整份 manifest，按 category 计 recall@k。
+ * @param dbFor 按 fixture 返回对应已加载 DB（zh/en 各一）
+ */
+export function computeRecall(dbFor, manifest, k) {
+    const outcomes = [];
+    const blockers = [];
+    for (const entry of manifest.entries) {
+        if (entry.expected_doc_ids === null || entry.expected_doc_ids.length === 0) {
+            outcomes.push({ id: entry.id, category: entry.category, hit: false, blocker: false, skipped: true });
+            continue;
+        }
+        const db = dbFor(entry.fixture);
+        const res = searchKbCore(db, entry.query, k);
+        const topDocIds = res.ok ? res.results.slice(0, k).map((r) => r.docId) : [];
+        const expected = new Set(entry.expected_doc_ids);
+        const hit = topDocIds.some((d) => expected.has(d));
+        // systematic zero-recall：0 命中 + 目标确实在库 → BLOCKER
+        const blocker = topDocIds.length === 0 && entry.expected_doc_ids.some((d) => docExists(db, d));
+        if (blocker)
+            blockers.push(entry.id);
+        outcomes.push({ id: entry.id, category: entry.category, hit, blocker, skipped: false });
+    }
+    const cats = new Map();
+    for (const o of outcomes) {
+        if (o.skipped)
+            continue;
+        const c = cats.get(o.category) ?? { total: 0, hits: 0 };
+        c.total += 1;
+        if (o.hit)
+            c.hits += 1;
+        cats.set(o.category, c);
+    }
+    const byCategory = [...cats.entries()].map(([category, { total, hits }]) => ({
+        category,
+        total,
+        hits,
+        recall: total === 0 ? 0 : hits / total,
+    }));
+    return { byCategory, outcomes, blockers };
+}
+//# sourceMappingURL=recall-eval.js.map

package/dist/scaffold-kb/recall-eval.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"recall-eval.js","sourceRoot":"","sources":["../../src/scaffold-kb/recall-eval.ts"],"names":[],"mappings":"AAAA;;;;;GAKG;AAGH,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAC/C,OAAO,EAAE,YAAY,EAAE,MAAM,kBAAkB,CAAC;AAgDhD,gDAAgD;AAChD,SAAS,SAAS,CAAC,EAAY,EAAE,KAAa;IAC5C,MAAM,IAAI,GAAG,SAAS,CAAC,EAAE,EAAE,mDAAmD,EAAE,CAAC,KAAK,CAAC,CAAC,CAAC;IACzF,OAAO,IAAI,CAAC,MAAM,GAAG,CAAC,CAAC;AACzB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,aAAa,CAC3B,KAAyC,EACzC,QAAwB,EACxB,CAAS;IAET,MAAM,QAAQ,GAAmB,EAAE,CAAC;IACpC,MAAM,QAAQ,GAAa,EAAE,CAAC;IAE9B,KAAK,MAAM,KAAK,IAAI,QAAQ,CAAC,OAAO,EAAE,CAAC;QACrC,IAAI,KAAK,CAAC,gBAAgB,KAAK,IAAI,IAAI,KAAK,CAAC,gBAAgB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3E,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;YACrG,SAAS;QACX,CAAC;QACD,MAAM,EAAE,GAAG,KAAK,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC;QAChC,MAAM,GAAG,GAAG,YAAY,CAAC,EAAE,EAAE,KAAK,CAAC,KAAK,EAAE,CAAC,CAAC,CAAC;QAC7C,MAAM,SAAS,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QAC5E,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC;QACjD,MAAM,GAAG,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QAEnD,iDAAiD;QACjD,MAAM,OAAO,GAAG,SAAS,CAAC,MAAM,KAAK,CAAC,IAAI,KAAK,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC;QAC/F,IAAI,OAAO;YAAE,QAAQ,CAAC,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;QAErC,QAAQ,CAAC,IAAI,CAAC,EAAE,EAAE,EAAE,KAAK,CAAC,EAAE,EAAE,QAAQ,EAAE,KAAK,CAAC,QAAQ,EAAE,GAAG,EAAE,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,CAAC,CAAC;IAC1F,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,GAAG,EAAmD,CAAC;IACxE,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACzB,IAAI,CAAC,CAAC,OAAO;YAAE,SAAS;QACxB,MAAM,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,IAAI,EAAE,KAAK,EAAE,CAAC,EAAE,IAAI,EAAE,CAAC,EAAE,CAAC;QACxD,CAAC,CAAC,KAAK,IAAI,CAAC,CAAC;QACb,IAAI,CAAC,CAAC,GAAG;YAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;QACvB,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC;IAC1B,CAAC;IACD,MAAM,UAAU,GAAqB,CAAC,GAAG,IAAI,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;QAC7F,QAAQ;QACR,KAAK;QACL,IAAI;QACJ,MAAM,EAAE,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,GAAG,KAAK;KACvC,CAAC,CAAC,CAAC;IAEJ,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAC;AAC5C,CAAC"}

package/dist/scaffold-kb/schema-compat.d.ts

@@ -0,0 +1,22 @@
+/**
+ * F192 — chunk_meta schema 向后兼容（C-1 / R-COMPAT-1）
+ *
+ * F190 已分发的 sqlite（厂商/demo 旧库）无 provenance 三列（ingest_source_type/
+ * ingest_origin/ingested_at）。对其显式 SELECT 这些列会 `no such column` 抛错。
+ *
+ * 本模块用 `PRAGMA table_info(chunk_meta)` 探测列是否存在，生成"始终产出 3 列"的
+ * SELECT 片段（旧库用 `NULL AS`）—— 使行索引布局对新旧库稳定，所有读取路径
+ * （search-core / kb_search / kb_doc_lookup / kb_api_lookup 的 provenance 投影）
+ * 一律走本片段，禁止任何处手写新列名（plan §6）。
+ */
+import type { SqliteDb } from './sqlite-engine.js';
+/** provenance 列名（F192 §3.3）；与 sqlite-writer CREATE_META 保持一致 */
+export declare const PROVENANCE_COLUMNS: readonly ["ingest_source_type", "ingest_origin", "ingested_at"];
+/** 探测 chunk_meta 是否含全部 provenance 列（F190 旧库返回 false） */
+export declare function hasProvenanceColumns(db: SqliteDb): boolean;
+/**
+ * 生成 provenance 的 SELECT 片段（含前导逗号，拼接在已有列之后）。
+ * 始终产出 3 列：新库选真实列，旧库用 `NULL AS`，保证调用方行索引布局不变。
+ */
+export declare function provenanceSelectFragment(db: SqliteDb): string;
+//# sourceMappingURL=schema-compat.d.ts.map

package/dist/scaffold-kb/schema-compat.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"schema-compat.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/schema-compat.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAGnD,gEAAgE;AAChE,eAAO,MAAM,kBAAkB,iEAAkE,CAAC;AAElG,wDAAwD;AACxD,wBAAgB,oBAAoB,CAAC,EAAE,EAAE,QAAQ,GAAG,OAAO,CAK1D;AAED;;;GAGG;AACH,wBAAgB,wBAAwB,CAAC,EAAE,EAAE,QAAQ,GAAG,MAAM,CAK7D"}

package/dist/scaffold-kb/schema-compat.js

@@ -0,0 +1,32 @@
+/**
+ * F192 — chunk_meta schema 向后兼容（C-1 / R-COMPAT-1）
+ *
+ * F190 已分发的 sqlite（厂商/demo 旧库）无 provenance 三列（ingest_source_type/
+ * ingest_origin/ingested_at）。对其显式 SELECT 这些列会 `no such column` 抛错。
+ *
+ * 本模块用 `PRAGMA table_info(chunk_meta)` 探测列是否存在，生成"始终产出 3 列"的
+ * SELECT 片段（旧库用 `NULL AS`）—— 使行索引布局对新旧库稳定，所有读取路径
+ * （search-core / kb_search / kb_doc_lookup / kb_api_lookup 的 provenance 投影）
+ * 一律走本片段，禁止任何处手写新列名（plan §6）。
+ */
+import { queryRows } from './sqlite-engine.js';
+/** provenance 列名（F192 §3.3）；与 sqlite-writer CREATE_META 保持一致 */
+export const PROVENANCE_COLUMNS = ['ingest_source_type', 'ingest_origin', 'ingested_at'];
+/** 探测 chunk_meta 是否含全部 provenance 列（F190 旧库返回 false） */
+export function hasProvenanceColumns(db) {
+    // PRAGMA table_info 返回行：cid, name, type, notnull, dflt_value, pk
+    const rows = queryRows(db, 'PRAGMA table_info(chunk_meta)', []);
+    const cols = new Set(rows.map((r) => String(r[1] ?? '')));
+    return PROVENANCE_COLUMNS.every((c) => cols.has(c));
+}
+/**
+ * 生成 provenance 的 SELECT 片段（含前导逗号，拼接在已有列之后）。
+ * 始终产出 3 列：新库选真实列，旧库用 `NULL AS`，保证调用方行索引布局不变。
+ */
+export function provenanceSelectFragment(db) {
+    if (hasProvenanceColumns(db)) {
+        return ', chunk_meta.ingest_source_type, chunk_meta.ingest_origin, chunk_meta.ingested_at';
+    }
+    return ', NULL AS ingest_source_type, NULL AS ingest_origin, NULL AS ingested_at';
+}
+//# sourceMappingURL=schema-compat.js.map

package/dist/scaffold-kb/schema-compat.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"schema-compat.js","sourceRoot":"","sources":["../../src/scaffold-kb/schema-compat.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAGH,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAE/C,gEAAgE;AAChE,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,oBAAoB,EAAE,eAAe,EAAE,aAAa,CAAU,CAAC;AAElG,wDAAwD;AACxD,MAAM,UAAU,oBAAoB,CAAC,EAAY;IAC/C,iEAAiE;IACjE,MAAM,IAAI,GAAG,SAAS,CAAC,EAAE,EAAE,+BAA+B,EAAE,EAAE,CAAC,CAAC;IAChE,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC;IAC1D,OAAO,kBAAkB,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;AACtD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,wBAAwB,CAAC,EAAY;IACnD,IAAI,oBAAoB,CAAC,EAAE,CAAC,EAAE,CAAC;QAC7B,OAAO,mFAAmF,CAAC;IAC7F,CAAC;IACD,OAAO,0EAA0E,CAAC;AACpF,CAAC"}

package/dist/scaffold-kb/search-core.d.ts

@@ -0,0 +1,42 @@
+/**
+ * F190 scaffold-kb — 单库查询核 searchKbCore（Phase A，recall 验收依赖此，非 kb_search）
+ *
+ * 修 Codex C-2（Phase A/B 边界）：recall 门禁在 Phase A，故查询核也在 Phase A。
+ * kb_search（Phase B）= 包装本核 + 双库 result-merger + envelope + telemetry + MCP 注册。
+ *
+ * 流程：sanitizeQuery → FTS5 MATCH（bm25 排序，越负越相关）→ chunk_meta 联表
+ *      → 短 CJK 查询 LIKE 兜底（EC-001）。返回带原始 bm25 score 的结果（供上层归一）。
+ */
+import type { SqliteDb } from './sqlite-engine.js';
+import type { IngestSourceType } from './types.js';
+export interface CoreResult {
+    chunkId: string;
+    docId: string;
+    contentRaw: string;
+    docTitle: string;
+    anchor: string | null;
+    sourceUrl: string | null;
+    sdkVersion: string | null;
+    builtAt: string;
+    /** F192 provenance（旧库 / 厂商构建为 null，经 schema-compat 投影，C-1） */
+    ingestSourceType: IngestSourceType | null;
+    ingestOrigin: string | null;
+    ingestedAt: string | null;
+    /** 原始 bm25 分（越负越相关）；LIKE 兜底命中记为 1.0（最不相关） */
+    score: number;
+    via: 'fts' | 'like';
+}
+export interface SearchCoreError {
+    ok: false;
+    code: 'INVALID_QUERY';
+}
+export interface SearchCoreOk {
+    ok: true;
+    results: CoreResult[];
+}
+/**
+ * 单库检索核。
+ * @returns ok+results（按相关度排序，FTS 优先、LIKE 兜底在后）或 INVALID_QUERY 错误
+ */
+export declare function searchKbCore(db: SqliteDb, query: string, topK: number, sdkVersion?: string, preTokenized?: boolean): SearchCoreOk | SearchCoreError;
+//# sourceMappingURL=search-core.d.ts.map

package/dist/scaffold-kb/search-core.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"search-core.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/search-core.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AACnD,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAKnD,MAAM,WAAW,UAAU;IACzB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,MAAM,CAAC;IACnB,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,OAAO,EAAE,MAAM,CAAC;IAChB,8DAA8D;IAC9D,gBAAgB,EAAE,gBAAgB,GAAG,IAAI,CAAC;IAC1C,YAAY,EAAE,MAAM,GAAG,IAAI,CAAC;IAC5B,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,6CAA6C;IAC7C,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,EAAE,KAAK,GAAG,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,KAAK,CAAC;IACV,IAAI,EAAE,eAAe,CAAC;CACvB;AAED,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,IAAI,CAAC;IACT,OAAO,EAAE,UAAU,EAAE,CAAC;CACvB;AAwCD;;;GAGG;AACH,wBAAgB,YAAY,CAC1B,EAAE,EAAE,QAAQ,EACZ,KAAK,EAAE,MAAM,EACb,IAAI,EAAE,MAAM,EACZ,UAAU,CAAC,EAAE,MAAM,EACnB,YAAY,UAAQ,GACnB,YAAY,GAAG,eAAe,CA4ChC"}

package/dist/scaffold-kb/search-core.js

@@ -0,0 +1,89 @@
+/**
+ * F190 scaffold-kb — 单库查询核 searchKbCore（Phase A，recall 验收依赖此，非 kb_search）
+ *
+ * 修 Codex C-2（Phase A/B 边界）：recall 门禁在 Phase A，故查询核也在 Phase A。
+ * kb_search（Phase B）= 包装本核 + 双库 result-merger + envelope + telemetry + MCP 注册。
+ *
+ * 流程：sanitizeQuery → FTS5 MATCH（bm25 排序，越负越相关）→ chunk_meta 联表
+ *      → 短 CJK 查询 LIKE 兜底（EC-001）。返回带原始 bm25 score 的结果（供上层归一）。
+ */
+import { queryRows } from './sqlite-engine.js';
+import { sanitizeQuery } from './query-sanitizer.js';
+import { provenanceSelectFragment } from './schema-compat.js';
+const HAN = /\p{Script=Han}/gu;
+const LIKE_FALLBACK_FLOOR = 3;
+const SELECT_COLS = `chunks.chunk_id, chunks.doc_id, chunks.content_raw,
+  chunk_meta.doc_title, chunk_meta.anchor, chunk_meta.source_url,
+  chunk_meta.sdk_version, chunk_meta.built_at`;
+/** 转义 LIKE 通配符 `%` `_` `\`，配合 `ESCAPE '\'`（修 Codex WARNING：召回污染） */
+function escapeLike(s) {
+    return s.replace(/[\\%_]/g, '\\$&');
+}
+function rowToResult(row, score, via) {
+    const str = (v) => (v == null ? '' : String(v));
+    const strOrNull = (v) => (v == null ? null : String(v));
+    return {
+        chunkId: str(row[0]),
+        docId: str(row[1]),
+        contentRaw: str(row[2]),
+        docTitle: str(row[3]),
+        anchor: strOrNull(row[4]),
+        sourceUrl: strOrNull(row[5]),
+        sdkVersion: strOrNull(row[6]),
+        builtAt: str(row[7]),
+        // provenance（schema-compat 片段固定产出 3 列：索引 8/9/10）
+        ingestSourceType: strOrNull(row[8]),
+        ingestOrigin: strOrNull(row[9]),
+        ingestedAt: strOrNull(row[10]),
+        score,
+        via,
+    };
+}
+/** 统计查询中的 CJK 字符数（决定是否触发短词 LIKE 兜底） */
+function countHan(query) {
+    const m = query.match(HAN);
+    return m ? m.length : 0;
+}
+/**
+ * 单库检索核。
+ * @returns ok+results（按相关度排序，FTS 优先、LIKE 兜底在后）或 INVALID_QUERY 错误
+ */
+export function searchKbCore(db, query, topK, sdkVersion, preTokenized = false) {
+    const sanitized = sanitizeQuery(query, 'OR', preTokenized);
+    if (!sanitized.ok) {
+        return { ok: false, code: 'INVALID_QUERY' };
+    }
+    // provenance 投影片段（始终 3 列，旧库 NULL AS；C-1 兼容）
+    const prov = provenanceSelectFragment(db);
+    // 可选 sdk_version 过滤（FR-007，修 Codex WARNING：原 sdk_version 暴露但未用）
+    const versionClause = sdkVersion !== undefined ? ' AND chunk_meta.sdk_version = ?' : '';
+    // 列顺序：SELECT_COLS(0-7) + prov(8-10) + bm25 score(11)
+    const ftsSql = `SELECT ${SELECT_COLS}${prov}, bm25(chunks) AS score
+FROM chunks JOIN chunk_meta ON chunk_meta.chunk_id = chunks.chunk_id
+WHERE chunks MATCH ?${versionClause}
+ORDER BY bm25(chunks)
+LIMIT ?`;
+    const ftsBind = sdkVersion !== undefined ? [sanitized.match, sdkVersion, topK] : [sanitized.match, topK];
+    const ftsRows = queryRows(db, ftsSql, ftsBind);
+    const results = ftsRows.map((row) => rowToResult(row, Number(row[11] ?? 0), 'fts'));
+    // EC-001：短 CJK 查询（≤2 字）且 FTS 命中不足 → content_raw LIKE 兜底（转义通配符），合并去重
+    const hanCount = countHan(query);
+    if (hanCount > 0 && hanCount <= 2 && results.length < LIKE_FALLBACK_FLOOR) {
+        const likeSql = `SELECT ${SELECT_COLS}${prov}
+FROM chunks JOIN chunk_meta ON chunk_meta.chunk_id = chunks.chunk_id
+WHERE chunks.content_raw LIKE ? ESCAPE '\\'${versionClause}
+LIMIT ?`;
+        const likePattern = `%${escapeLike(query.trim())}%`;
+        const likeBind = sdkVersion !== undefined ? [likePattern, sdkVersion, topK] : [likePattern, topK];
+        const seen = new Set(results.map((r) => r.chunkId));
+        for (const row of queryRows(db, likeSql, likeBind)) {
+            const chunkId = String(row[0] ?? '');
+            if (!seen.has(chunkId)) {
+                seen.add(chunkId);
+                results.push(rowToResult(row, 1.0, 'like'));
+            }
+        }
+    }
+    return { ok: true, results: results.slice(0, topK) };
+}
+//# sourceMappingURL=search-core.js.map

package/dist/scaffold-kb/search-core.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"search-core.js","sourceRoot":"","sources":["../../src/scaffold-kb/search-core.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAIH,OAAO,EAAE,SAAS,EAAE,MAAM,oBAAoB,CAAC;AAC/C,OAAO,EAAE,aAAa,EAAE,MAAM,sBAAsB,CAAC;AACrD,OAAO,EAAE,wBAAwB,EAAE,MAAM,oBAAoB,CAAC;AA8B9D,MAAM,GAAG,GAAG,kBAAkB,CAAC;AAC/B,MAAM,mBAAmB,GAAG,CAAC,CAAC;AAC9B,MAAM,WAAW,GAAG;;8CAE0B,CAAC;AAE/C,oEAAoE;AACpE,SAAS,UAAU,CAAC,CAAS;IAC3B,OAAO,CAAC,CAAC,OAAO,CAAC,SAAS,EAAE,MAAM,CAAC,CAAC;AACtC,CAAC;AAED,SAAS,WAAW,CAAC,GAAc,EAAE,KAAa,EAAE,GAAmB;IACrE,MAAM,GAAG,GAAG,CAAC,CAAU,EAAU,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IACjE,MAAM,SAAS,GAAG,CAAC,CAAU,EAAiB,EAAE,CAAC,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IAChF,OAAO;QACL,OAAO,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACpB,KAAK,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAClB,UAAU,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACvB,QAAQ,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACrB,MAAM,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACzB,SAAS,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC5B,UAAU,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC7B,OAAO,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACpB,iDAAiD;QACjD,gBAAgB,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAA4B;QAC9D,YAAY,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC/B,UAAU,EAAE,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QAC9B,KAAK;QACL,GAAG;KACJ,CAAC;AACJ,CAAC;AAED,uCAAuC;AACvC,SAAS,QAAQ,CAAC,KAAa;IAC7B,MAAM,CAAC,GAAG,KAAK,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC;IAC3B,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;AAC1B,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,YAAY,CAC1B,EAAY,EACZ,KAAa,EACb,IAAY,EACZ,UAAmB,EACnB,YAAY,GAAG,KAAK;IAEpB,MAAM,SAAS,GAAG,aAAa,CAAC,KAAK,EAAE,IAAI,EAAE,YAAY,CAAC,CAAC;IAC3D,IAAI,CAAC,SAAS,CAAC,EAAE,EAAE,CAAC;QAClB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,eAAe,EAAE,CAAC;IAC9C,CAAC;IAED,4CAA4C;IAC5C,MAAM,IAAI,GAAG,wBAAwB,CAAC,EAAE,CAAC,CAAC;IAE1C,gEAAgE;IAChE,MAAM,aAAa,GAAG,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,iCAAiC,CAAC,CAAC,CAAC,EAAE,CAAC;IAExF,qDAAqD;IACrD,MAAM,MAAM,GAAG,UAAU,WAAW,GAAG,IAAI;;sBAEvB,aAAa;;QAE3B,CAAC;IACP,MAAM,OAAO,GACX,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;IAC3F,MAAM,OAAO,GAAG,SAAS,CAAC,EAAE,EAAE,MAAM,EAAE,OAAO,CAAC,CAAC;IAC/C,MAAM,OAAO,GAAiB,OAAO,CAAC,GAAG,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,WAAW,CAAC,GAAG,EAAE,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,CAAC,EAAE,KAAK,CAAC,CAAC,CAAC;IAElG,oEAAoE;IACpE,MAAM,QAAQ,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC;IACjC,IAAI,QAAQ,GAAG,CAAC,IAAI,QAAQ,IAAI,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,mBAAmB,EAAE,CAAC;QAC1E,MAAM,OAAO,GAAG,UAAU,WAAW,GAAG,IAAI;;6CAEH,aAAa;QAClD,CAAC;QACL,MAAM,WAAW,GAAG,IAAI,UAAU,CAAC,KAAK,CAAC,IAAI,EAAE,CAAC,GAAG,CAAC;QACpD,MAAM,QAAQ,GACZ,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,UAAU,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,WAAW,EAAE,IAAI,CAAC,CAAC;QACnF,MAAM,IAAI,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC;QACpD,KAAK,MAAM,GAAG,IAAI,SAAS,CAAC,EAAE,EAAE,OAAO,EAAE,QAAQ,CAAC,EAAE,CAAC;YACnD,MAAM,OAAO,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;YACrC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAC;gBACvB,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;gBAClB,OAAO,CAAC,IAAI,CAAC,WAAW,CAAC,GAAG,EAAE,GAAG,EAAE,MAAM,CAAC,CAAC,CAAC;YAC9C,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CAAC;AACvD,CAAC"}

package/dist/scaffold-kb/sqlite-engine.d.ts

@@ -0,0 +1,58 @@
+/**
+ * F190 scaffold-kb — @sqlite.org/sqlite-wasm 封装（WASM 单例 + 内存 DB 落盘/加载）
+ *
+ * 实证（codebase-grounding §6）：官方 sqlite-wasm 含 FTS5；纯内存 DB 经
+ * `sqlite3_js_db_export` 导出字节落盘，经 `sqlite3_deserialize` 从字节重建。
+ * WASM module 初始化较重，进程内缓存为单例（cold/warm 性能前提，plan §4.9）。
+ *
+ * sqlite-wasm 的上层 oo1 API 有类型声明，但低层 capi/wasm 句柄类型不全，
+ * 故在本模块内用最小本地接口收口这些调用，避免污染上层类型。
+ */
+/** 一行查询结果（列值数组） */
+export type Row = unknown[];
+/** 对 sqlite-wasm oo1.DB 的最小类型视图（只声明本模块用到的成员） */
+export interface SqliteDb {
+    exec(opts: {
+        sql: string;
+        bind?: unknown[];
+        rowMode?: 'array' | 'object';
+        resultRows?: Row[];
+    }): void;
+    exec(sql: string): void;
+    close(): void;
+    readonly pointer: number;
+    checkRc(rc: number): void;
+}
+/** 低层 capi/wasm 句柄（sqlite-wasm 未完整声明的部分，本地最小收口） */
+interface Sqlite3Api {
+    oo1: {
+        DB: new (filename?: string) => SqliteDb;
+    };
+    capi: {
+        sqlite3_js_db_export(db: SqliteDb): Uint8Array;
+        sqlite3_deserialize(dbPtr: number, schema: string, data: number, szDb: number, szBuf: number, flags: number): number;
+        SQLITE_DESERIALIZE_FREEONCLOSE: number;
+        SQLITE_DESERIALIZE_RESIZEABLE: number;
+    };
+    wasm: {
+        allocFromTypedArray(bytes: Uint8Array): number;
+    };
+}
+/** 初始化（或复用）WASM module 单例 */
+export declare function initSqlite(): Promise<Sqlite3Api>;
+/** 打开一个空的内存 DB */
+export declare function openMemoryDb(): Promise<{
+    sqlite3: Sqlite3Api;
+    db: SqliteDb;
+}>;
+/** 把内存 DB 导出为字节（供落盘 chunks.sqlite） */
+export declare function exportDb(sqlite3: Sqlite3Api, db: SqliteDb): Uint8Array;
+/** 从字节重建内存 DB（供 serve 加载已构建的 chunks.sqlite） */
+export declare function loadDbFromBytes(bytes: Uint8Array): Promise<{
+    sqlite3: Sqlite3Api;
+    db: SqliteDb;
+}>;
+/** 便捷查询：返回行数组 */
+export declare function queryRows(db: SqliteDb, sql: string, bind?: unknown[]): Row[];
+export {};
+//# sourceMappingURL=sqlite-engine.d.ts.map

package/dist/scaffold-kb/sqlite-engine.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sqlite-engine.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/sqlite-engine.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAIH,mBAAmB;AACnB,MAAM,MAAM,GAAG,GAAG,OAAO,EAAE,CAAC;AAE5B,gDAAgD;AAChD,MAAM,WAAW,QAAQ;IACvB,IAAI,CAAC,IAAI,EAAE;QACT,GAAG,EAAE,MAAM,CAAC;QACZ,IAAI,CAAC,EAAE,OAAO,EAAE,CAAC;QACjB,OAAO,CAAC,EAAE,OAAO,GAAG,QAAQ,CAAC;QAC7B,UAAU,CAAC,EAAE,GAAG,EAAE,CAAC;KACpB,GAAG,IAAI,CAAC;IACT,IAAI,CAAC,GAAG,EAAE,MAAM,GAAG,IAAI,CAAC;IACxB,KAAK,IAAI,IAAI,CAAC;IACd,QAAQ,CAAC,OAAO,EAAE,MAAM,CAAC;IACzB,OAAO,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAAC;CAC3B;AAED,mDAAmD;AACnD,UAAU,UAAU;IAClB,GAAG,EAAE;QAAE,EAAE,EAAE,KAAK,QAAQ,CAAC,EAAE,MAAM,KAAK,QAAQ,CAAA;KAAE,CAAC;IACjD,IAAI,EAAE;QACJ,oBAAoB,CAAC,EAAE,EAAE,QAAQ,GAAG,UAAU,CAAC;QAC/C,mBAAmB,CACjB,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,EACd,IAAI,EAAE,MAAM,EACZ,IAAI,EAAE,MAAM,EACZ,KAAK,EAAE,MAAM,EACb,KAAK,EAAE,MAAM,GACZ,MAAM,CAAC;QACV,8BAA8B,EAAE,MAAM,CAAC;QACvC,6BAA6B,EAAE,MAAM,CAAC;KACvC,CAAC;IACF,IAAI,EAAE;QAAE,mBAAmB,CAAC,KAAK,EAAE,UAAU,GAAG,MAAM,CAAA;KAAE,CAAC;CAC1D;AAID,6BAA6B;AAC7B,wBAAsB,UAAU,IAAI,OAAO,CAAC,UAAU,CAAC,CAOtD;AAED,kBAAkB;AAClB,wBAAsB,YAAY,IAAI,OAAO,CAAC;IAAE,OAAO,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,QAAQ,CAAA;CAAE,CAAC,CAInF;AAED,sCAAsC;AACtC,wBAAgB,QAAQ,CAAC,OAAO,EAAE,UAAU,EAAE,EAAE,EAAE,QAAQ,GAAG,UAAU,CAEtE;AAED,+CAA+C;AAC/C,wBAAsB,eAAe,CAAC,KAAK,EAAE,UAAU,GAAG,OAAO,CAAC;IAAE,OAAO,EAAE,UAAU,CAAC;IAAC,EAAE,EAAE,QAAQ,CAAA;CAAE,CAAC,CA4BvG;AAED,iBAAiB;AACjB,wBAAgB,SAAS,CAAC,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE,OAAO,EAAE,GAAG,GAAG,EAAE,CAI5E"}

package/dist/scaffold-kb/sqlite-engine.js

@@ -0,0 +1,61 @@
+/**
+ * F190 scaffold-kb — @sqlite.org/sqlite-wasm 封装（WASM 单例 + 内存 DB 落盘/加载）
+ *
+ * 实证（codebase-grounding §6）：官方 sqlite-wasm 含 FTS5；纯内存 DB 经
+ * `sqlite3_js_db_export` 导出字节落盘，经 `sqlite3_deserialize` 从字节重建。
+ * WASM module 初始化较重，进程内缓存为单例（cold/warm 性能前提，plan §4.9）。
+ *
+ * sqlite-wasm 的上层 oo1 API 有类型声明，但低层 capi/wasm 句柄类型不全，
+ * 故在本模块内用最小本地接口收口这些调用，避免污染上层类型。
+ */
+import sqlite3InitModule from '@sqlite.org/sqlite-wasm';
+let modulePromise = null;
+/** 初始化（或复用）WASM module 单例 */
+export async function initSqlite() {
+    if (modulePromise === null) {
+        modulePromise = sqlite3InitModule().then((m) => m);
+    }
+    return modulePromise;
+}
+/** 打开一个空的内存 DB */
+export async function openMemoryDb() {
+    const sqlite3 = await initSqlite();
+    const db = new sqlite3.oo1.DB(':memory:');
+    return { sqlite3, db };
+}
+/** 把内存 DB 导出为字节（供落盘 chunks.sqlite） */
+export function exportDb(sqlite3, db) {
+    return sqlite3.capi.sqlite3_js_db_export(db);
+}
+/** 从字节重建内存 DB（供 serve 加载已构建的 chunks.sqlite） */
+export async function loadDbFromBytes(bytes) {
+    const sqlite3 = await initSqlite();
+    const db = new sqlite3.oo1.DB();
+    // Node 的 readFileSync 返回 Buffer，其底层 ArrayBuffer 是池化的（byteOffset≠0 / 超长），
+    // 直接喂给 allocFromTypedArray 会触发 heapForSize 错误。复制为紧凑 Uint8Array 规避。
+    const tight = new Uint8Array(bytes.byteLength);
+    tight.set(bytes);
+    try {
+        const ptr = sqlite3.wasm.allocFromTypedArray(tight);
+        const rc = sqlite3.capi.sqlite3_deserialize(db.pointer, 'main', ptr, tight.byteLength, tight.byteLength, sqlite3.capi.SQLITE_DESERIALIZE_FREEONCLOSE | sqlite3.capi.SQLITE_DESERIALIZE_RESIZEABLE);
+        db.checkRc(rc);
+    }
+    catch (err) {
+        // 失败时关闭已创建的 DB 句柄，避免 WASM 内存泄漏（修 Codex WARNING）
+        try {
+            db.close();
+        }
+        catch {
+            // close 本身失败忽略，优先抛原始错误
+        }
+        throw err;
+    }
+    return { sqlite3, db };
+}
+/** 便捷查询：返回行数组 */
+export function queryRows(db, sql, bind) {
+    const resultRows = [];
+    db.exec({ sql, bind: bind ?? [], rowMode: 'array', resultRows });
+    return resultRows;
+}
+//# sourceMappingURL=sqlite-engine.js.map

package/dist/scaffold-kb/sqlite-engine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sqlite-engine.js","sourceRoot":"","sources":["../../src/scaffold-kb/sqlite-engine.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,iBAAiB,MAAM,yBAAyB,CAAC;AAsCxD,IAAI,aAAa,GAA+B,IAAI,CAAC;AAErD,6BAA6B;AAC7B,MAAM,CAAC,KAAK,UAAU,UAAU;IAC9B,IAAI,aAAa,KAAK,IAAI,EAAE,CAAC;QAC3B,aAAa,GAAI,iBAAiB,EAAuB,CAAC,IAAI,CAC5D,CAAC,CAAC,EAAE,EAAE,CAAC,CAAe,CACvB,CAAC;IACJ,CAAC;IACD,OAAO,aAAa,CAAC;AACvB,CAAC;AAED,kBAAkB;AAClB,MAAM,CAAC,KAAK,UAAU,YAAY;IAChC,MAAM,OAAO,GAAG,MAAM,UAAU,EAAE,CAAC;IACnC,MAAM,EAAE,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC,EAAE,CAAC,UAAU,CAAC,CAAC;IAC1C,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;AACzB,CAAC;AAED,sCAAsC;AACtC,MAAM,UAAU,QAAQ,CAAC,OAAmB,EAAE,EAAY;IACxD,OAAO,OAAO,CAAC,IAAI,CAAC,oBAAoB,CAAC,EAAE,CAAC,CAAC;AAC/C,CAAC;AAED,+CAA+C;AAC/C,MAAM,CAAC,KAAK,UAAU,eAAe,CAAC,KAAiB;IACrD,MAAM,OAAO,GAAG,MAAM,UAAU,EAAE,CAAC;IACnC,MAAM,EAAE,GAAG,IAAI,OAAO,CAAC,GAAG,CAAC,EAAE,EAAE,CAAC;IAChC,yEAAyE;IACzE,mEAAmE;IACnE,MAAM,KAAK,GAAG,IAAI,UAAU,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC;IAC/C,KAAK,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC;IACjB,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,OAAO,CAAC,IAAI,CAAC,mBAAmB,CAAC,KAAK,CAAC,CAAC;QACpD,MAAM,EAAE,GAAG,OAAO,CAAC,IAAI,CAAC,mBAAmB,CACzC,EAAE,CAAC,OAAO,EACV,MAAM,EACN,GAAG,EACH,KAAK,CAAC,UAAU,EAChB,KAAK,CAAC,UAAU,EAChB,OAAO,CAAC,IAAI,CAAC,8BAA8B,GAAG,OAAO,CAAC,IAAI,CAAC,6BAA6B,CACzF,CAAC;QACF,EAAE,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;IACjB,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,gDAAgD;QAChD,IAAI,CAAC;YACH,EAAE,CAAC,KAAK,EAAE,CAAC;QACb,CAAC;QAAC,MAAM,CAAC;YACP,uBAAuB;QACzB,CAAC;QACD,MAAM,GAAG,CAAC;IACZ,CAAC;IACD,OAAO,EAAE,OAAO,EAAE,EAAE,EAAE,CAAC;AACzB,CAAC;AAED,iBAAiB;AACjB,MAAM,UAAU,SAAS,CAAC,EAAY,EAAE,GAAW,EAAE,IAAgB;IACnE,MAAM,UAAU,GAAU,EAAE,CAAC;IAC7B,EAAE,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,IAAI,EAAE,EAAE,OAAO,EAAE,OAAO,EAAE,UAAU,EAAE,CAAC,CAAC;IACjE,OAAO,UAAU,CAAC;AACpB,CAAC"}

package/dist/scaffold-kb/sqlite-writer.d.ts

@@ -0,0 +1,18 @@
+/**
+ * F190 scaffold-kb — chunks.sqlite 写入（FTS5 建表 + 写入 + 落盘）
+ *
+ * 表结构（plan §4.1.2 / spec §3.3）：
+ * - `chunks` FTS5 虚拟表：chunk_id/doc_id/content_raw 为 UNINDEXED（不参与匹配，
+ *   content_raw 供 envelope 原样返回，R-003/Codex C4）；content_tokenized 是唯一检索列
+ *   （写入前过 normalizeForIndex，与查询侧同构）
+ * - `chunk_meta` 普通表：冗余 doc_title/source_url，使 kb_search 不依赖 doc-graph.json
+ */
+import type { Chunk, ChunkMeta } from './types.js';
+/**
+ * 在内存 DB 构建 chunks + chunk_meta，导出为字节。
+ * 纯函数（不落盘），便于测试。
+ */
+export declare function buildChunksDbBytes(chunks: Chunk[], meta: ChunkMeta[]): Promise<Uint8Array>;
+/** 构建并落盘 chunks.sqlite 到 outputPath */
+export declare function writeChunksToSqlite(chunks: Chunk[], meta: ChunkMeta[], outputPath: string): Promise<void>;
+//# sourceMappingURL=sqlite-writer.d.ts.map

package/dist/scaffold-kb/sqlite-writer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"sqlite-writer.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/sqlite-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAGH,OAAO,KAAK,EAAE,KAAK,EAAE,SAAS,EAAE,MAAM,YAAY,CAAC;AA0BnD;;;GAGG;AACH,wBAAsB,kBAAkB,CACtC,MAAM,EAAE,KAAK,EAAE,EACf,IAAI,EAAE,SAAS,EAAE,GAChB,OAAO,CAAC,UAAU,CAAC,CAiCrB;AAED,uCAAuC;AACvC,wBAAsB,mBAAmB,CACvC,MAAM,EAAE,KAAK,EAAE,EACf,IAAI,EAAE,SAAS,EAAE,EACjB,UAAU,EAAE,MAAM,GACjB,OAAO,CAAC,IAAI,CAAC,CAGf"}

package/dist/scaffold-kb/sqlite-writer.js

@@ -0,0 +1,76 @@
+/**
+ * F190 scaffold-kb — chunks.sqlite 写入（FTS5 建表 + 写入 + 落盘）
+ *
+ * 表结构（plan §4.1.2 / spec §3.3）：
+ * - `chunks` FTS5 虚拟表：chunk_id/doc_id/content_raw 为 UNINDEXED（不参与匹配，
+ *   content_raw 供 envelope 原样返回，R-003/Codex C4）；content_tokenized 是唯一检索列
+ *   （写入前过 normalizeForIndex，与查询侧同构）
+ * - `chunk_meta` 普通表：冗余 doc_title/source_url，使 kb_search 不依赖 doc-graph.json
+ */
+import { writeFileSync } from 'node:fs';
+import { normalizeForIndex } from './tokenizer.js';
+import { openMemoryDb, exportDb } from './sqlite-engine.js';
+const CREATE_CHUNKS = `CREATE VIRTUAL TABLE chunks USING fts5(
+  chunk_id UNINDEXED,
+  doc_id UNINDEXED,
+  content_raw UNINDEXED,
+  content_tokenized,
+  tokenize = 'unicode61 remove_diacritics 1'
+)`;
+// F192：新建/项目库的 chunk_meta 含 provenance 三列（F190 旧库无，读取走 schema-compat）
+const CREATE_META = `CREATE TABLE chunk_meta (
+  chunk_id TEXT PRIMARY KEY,
+  doc_id TEXT NOT NULL,
+  doc_title TEXT NOT NULL,
+  source_url TEXT,
+  anchor TEXT,
+  sdk_version TEXT,
+  built_at TEXT NOT NULL,
+  ingest_source_type TEXT,
+  ingest_origin TEXT,
+  ingested_at TEXT
+)`;
+/**
+ * 在内存 DB 构建 chunks + chunk_meta，导出为字节。
+ * 纯函数（不落盘），便于测试。
+ */
+export async function buildChunksDbBytes(chunks, meta) {
+    const { sqlite3, db } = await openMemoryDb();
+    try {
+        db.exec(CREATE_CHUNKS);
+        db.exec(CREATE_META);
+        for (const c of chunks) {
+            db.exec({
+                sql: 'INSERT INTO chunks(chunk_id, doc_id, content_raw, content_tokenized) VALUES(?,?,?,?)',
+                bind: [c.chunkId, c.docId, c.contentRaw, normalizeForIndex(c.contentRaw)],
+            });
+        }
+        for (const m of meta) {
+            db.exec({
+                sql: 'INSERT INTO chunk_meta(chunk_id, doc_id, doc_title, source_url, anchor, sdk_version, built_at, ingest_source_type, ingest_origin, ingested_at) VALUES(?,?,?,?,?,?,?,?,?,?)',
+                bind: [
+                    m.chunkId,
+                    m.docId,
+                    m.docTitle,
+                    m.sourceUrl,
+                    m.anchor,
+                    m.sdkVersion,
+                    m.builtAt,
+                    m.ingestSourceType ?? null,
+                    m.ingestOrigin ?? null,
+                    m.ingestedAt ?? null,
+                ],
+            });
+        }
+        return exportDb(sqlite3, db);
+    }
+    finally {
+        db.close();
+    }
+}
+/** 构建并落盘 chunks.sqlite 到 outputPath */
+export async function writeChunksToSqlite(chunks, meta, outputPath) {
+    const bytes = await buildChunksDbBytes(chunks, meta);
+    writeFileSync(outputPath, bytes);
+}
+//# sourceMappingURL=sqlite-writer.js.map

package/dist/scaffold-kb/sqlite-writer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"sqlite-writer.js","sourceRoot":"","sources":["../../src/scaffold-kb/sqlite-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAExC,OAAO,EAAE,iBAAiB,EAAE,MAAM,gBAAgB,CAAC;AACnD,OAAO,EAAE,YAAY,EAAE,QAAQ,EAAE,MAAM,oBAAoB,CAAC;AAE5D,MAAM,aAAa,GAAG;;;;;;EAMpB,CAAC;AAEH,sEAAsE;AACtE,MAAM,WAAW,GAAG;;;;;;;;;;;EAWlB,CAAC;AAEH;;;GAGG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CACtC,MAAe,EACf,IAAiB;IAEjB,MAAM,EAAE,OAAO,EAAE,EAAE,EAAE,GAAG,MAAM,YAAY,EAAE,CAAC;IAC7C,IAAI,CAAC;QACH,EAAE,CAAC,IAAI,CAAC,aAAa,CAAC,CAAC;QACvB,EAAE,CAAC,IAAI,CAAC,WAAW,CAAC,CAAC;QAErB,KAAK,MAAM,CAAC,IAAI,MAAM,EAAE,CAAC;YACvB,EAAE,CAAC,IAAI,CAAC;gBACN,GAAG,EAAE,sFAAsF;gBAC3F,IAAI,EAAE,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,KAAK,EAAE,CAAC,CAAC,UAAU,EAAE,iBAAiB,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;aAC1E,CAAC,CAAC;QACL,CAAC;QACD,KAAK,MAAM,CAAC,IAAI,IAAI,EAAE,CAAC;YACrB,EAAE,CAAC,IAAI,CAAC;gBACN,GAAG,EAAE,4KAA4K;gBACjL,IAAI,EAAE;oBACJ,CAAC,CAAC,OAAO;oBACT,CAAC,CAAC,KAAK;oBACP,CAAC,CAAC,QAAQ;oBACV,CAAC,CAAC,SAAS;oBACX,CAAC,CAAC,MAAM;oBACR,CAAC,CAAC,UAAU;oBACZ,CAAC,CAAC,OAAO;oBACT,CAAC,CAAC,gBAAgB,IAAI,IAAI;oBAC1B,CAAC,CAAC,YAAY,IAAI,IAAI;oBACtB,CAAC,CAAC,UAAU,IAAI,IAAI;iBACrB;aACF,CAAC,CAAC;QACL,CAAC;QACD,OAAO,QAAQ,CAAC,OAAO,EAAE,EAAE,CAAC,CAAC;IAC/B,CAAC;YAAS,CAAC;QACT,EAAE,CAAC,KAAK,EAAE,CAAC;IACb,CAAC;AACH,CAAC;AAED,uCAAuC;AACvC,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,MAAe,EACf,IAAiB,EACjB,UAAkB;IAElB,MAAM,KAAK,GAAG,MAAM,kBAAkB,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;IACrD,aAAa,CAAC,UAAU,EAAE,KAAK,CAAC,CAAC;AACnC,CAAC"}

package/dist/scaffold-kb/tokenizer.d.ts

@@ -0,0 +1,26 @@
+/**
+ * F190 scaffold-kb — CJK/符号感知的索引规范化（唯一 canonical 形式）
+ *
+ * 写入侧与查询侧共用同一个 `tokenize` / `normalizeForIndex`，保证同构（plan §4.1.2）。
+ * 设计要点：
+ * - SQLite FTS5 默认 unicode61 不切中文、对 `.`/`-`/`_` 等符号按分隔处理；
+ *   故不依赖 tokenizer 的 CJK 感知，而是在写入/查询前把文本规范化成"空格分隔、
+ *   unicode61 能逐 token 索引"的形式。
+ * - CJK 连续段 → unigram + bigram（单字保证短查询不零召回，bigram 提精度）。
+ * - ASCII 符号段（含 `._-<>@()/`）→ 各组件 + 拼接形（既能按组件查，也能按整符号查）；
+ *   不保留原始 `sdk.Init()` 字面（unicode61 下必被切），不使用任何特殊编码。
+ */
+/**
+ * 把文本切成有序 token 数组（写入侧与查询侧共用）。
+ *
+ * - 按字符类别（han / ascii / sep）切分为 run；sep run 丢弃
+ * - han run → unigram + bigram
+ * - ascii run → 组件 + 拼接形
+ */
+export declare function tokenize(rawText: string): string[];
+/**
+ * 规范化为 FTS5 索引/查询用的空格分隔字符串。
+ * `normalizeForIndex(t) === tokenize(t).join(' ')`，写入与查询同构的单一事实源。
+ */
+export declare function normalizeForIndex(text: string): string;
+//# sourceMappingURL=tokenizer.d.ts.map

package/dist/scaffold-kb/tokenizer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"tokenizer.d.ts","sourceRoot":"","sources":["../../src/scaffold-kb/tokenizer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AA2CH;;;;;;GAMG;AACH,wBAAgB,QAAQ,CAAC,OAAO,EAAE,MAAM,GAAG,MAAM,EAAE,CA4BlD;AAED;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,CAEtD"}