pi-mega-compact 0.8.24 → 0.8.26

package/src/vectorStore.ts

@@ -11,431 +11,492 @@
 import { createHash } from "node:crypto";
 import type { Embedder } from "./embedder.js";
 import { cosineSimilarity, defaultEmbedder } from "./embedder.js";
-import { loadDedupConfig, type DedupConfigShape, type DedupTier } from "./config/dedup.js";
+import {
+	loadDedupConfig,
+	type DedupConfigShape,
+	type DedupTier,
+} from "./config/dedup.js";
 import { logDecision } from "./monitoring.js";
 import type { StoredCheckpoint } from "./store.js";
 import { getStateDir, normalizeSessionId, compressSmart } from "./store.js";
 import { computeContentDigest } from "./dedup/digest.js";
-import { minhashSignature, SIGNATURE_VERSION, NUM_HASHES } from "./dedup/l1-minhash.js";
+import type { RaptorTree } from "./dedup/raptor/tree.js";
+import {
+	minhashSignature,
+	SIGNATURE_VERSION,
+	NUM_HASHES,
+} from "./dedup/l1-minhash.js";
 import { lshBands } from "./dedup/l1-lsh.js";
 import { isNearDuplicate } from "./dedup/l1-verify.js";
 import { openBloom, saveBloom } from "./store/bloom.js";
 import {
-  listCheckpoints,
-  nextCheckpointId,
-  upsertCheckpoint,
-  loadSessionState,
-  saveSessionState,
-  upsertMinhashSignature,
-  insertLshBuckets,
-  lshCandidateChunks,
-  addTokensSaved,
-  bumpDedupStats,
+	listCheckpoints,
+	nextCheckpointId,
+	upsertCheckpoint,
+	loadSessionState,
+	saveSessionState,
+	upsertMinhashSignature,
+	insertLshBuckets,
+	lshCandidateChunks,
+	addTokensSaved,
+	bumpDedupStats,
 } from "./store/sqlite.js";
 import { migrateJsonToSqlite } from "./store/migrate.js";
 
 export interface SearchHit {
-  checkpoint: StoredCheckpoint;
-  score: number;
-  /** Source repo id (the foreign repo's stateDir) for cross-repo hits, set by
-   *  `searchAsync` so the recall block can label foreign checkpoints. Undefined
-   *  for same-repo hits (the default path). */
-  repoId?: string;
+	checkpoint: StoredCheckpoint;
+	score: number;
+	/** Source repo id (the foreign repo's stateDir) for cross-repo hits, set by
+	 *  `searchAsync` so the recall block can label foreign checkpoints. Undefined
+	 *  for same-repo hits (the default path). */
+	repoId?: string;
+	/** S42B: when set, this hit is a RAPTOR cluster node (not a stored checkpoint).
+	 *  The recall block uses `raptorSummary` instead of checkpoint.summary. */
+	raptorSummary?: string;
+	/** S42B: the tree level of the cluster node (0 = leaves, 1+ = internal). */
+	raptorLevel?: number;
 }
 
 export interface AddInput {
-  sessionId: string;
-  summary: string;
-  /** Compressed topic summary (extractive). When present, embedded instead of regionText. */
-  topicSummary?: string;
-  keyDecisions?: string[];
-  nextSteps?: string[];
-  filesModified?: string[];
-  tokenEstimate?: number;
-  /** Token count of the ORIGINAL dropped region (before compaction). Drives the
-   *  honest "tokens saved" = originalTokenEstimate − tokenEstimate (stored), or
-   *  the full originalTokenEstimate when the region dedups (nothing new stored).
-   *  Optional for back-compat with direct add() callers; defaults to stored. */
-  originalTokenEstimate?: number;
-  /** Raw text of the compacted region — used to derive the regionHash + vector. */
-  regionText: string;
-  timestamp: number;
-  /** Sync progress callback fired as each dedup tier is evaluated (L0→L1→L2→new).
-   *  Lets the UI render live per-tier progress during compaction. Never awaited;
-   *  must be cheap. Optional for back-compat. */
-  onTier?: (ev: { tier: "L0" | "L1" | "L2" | "new"; status: "scanning" | "deduped" | "passed" | "stored"; detail?: string }) => void;
-  /** Context-window pressure (0–1) — escalates the stored checkpoint's sync
-   *  compression strength (Fix E). Optional; defaults to 0 (brotli-4). */
-  compressionPressure?: number;
+	sessionId: string;
+	summary: string;
+	/** Compressed topic summary (extractive). When present, embedded instead of regionText. */
+	topicSummary?: string;
+	keyDecisions?: string[];
+	nextSteps?: string[];
+	filesModified?: string[];
+	tokenEstimate?: number;
+	/** Token count of the ORIGINAL dropped region (before compaction). Drives the
+	 *  honest "tokens saved" = originalTokenEstimate − tokenEstimate (stored), or
+	 *  the full originalTokenEstimate when the region dedups (nothing new stored).
+	 *  Optional for back-compat with direct add() callers; defaults to stored. */
+	originalTokenEstimate?: number;
+	/** Raw text of the compacted region — used to derive the regionHash + vector. */
+	regionText: string;
+	timestamp: number;
+	/** Sync progress callback fired as each dedup tier is evaluated (L0→L1→L2→new).
+	 *  Lets the UI render live per-tier progress during compaction. Never awaited;
+	 *  must be cheap. Optional for back-compat. */
+	onTier?: (ev: {
+		tier: "L0" | "L1" | "L2" | "new";
+		status: "scanning" | "deduped" | "passed" | "stored";
+		detail?: string;
+	}) => void;
+	/** Context-window pressure (0–1) — escalates the stored checkpoint's sync
+	 *  compression strength (Fix E). Optional; defaults to 0 (brotli-4). */
+	compressionPressure?: number;
 }
 
 /** Default L2 semantic-dedup enable flag (trigram embedder is local, zero-network). */
 export const L2_ENABLED = true;
 
 export interface AddResult {
-  checkpoint: StoredCheckpoint;
-  deduped: boolean; // true when an equivalent region already existed (skipped embed)
-  /** Which dedup tier matched: regionHash | summaryHash | contentSimilarity | undefined (new). */
-  reason?: string;
+	checkpoint: StoredCheckpoint;
+	deduped: boolean; // true when an equivalent region already existed (skipped embed)
+	/** Which dedup tier matched: regionHash | summaryHash | contentSimilarity | undefined (new). */
+	reason?: string;
 }
 
 /** Stable hash of a compacted region, the dedup sentinel key. */
 export function computeRegionHash(regionText: string): string {
-  // Normalize whitespace before hashing so "foo  bar" and "foo bar" dedup.
-  const normalized = regionText.replace(/\s+/g, " ").trim();
-  return createHash("sha256").update(normalized).digest("hex").slice(0, 16);
+	// Normalize whitespace before hashing so "foo  bar" and "foo bar" dedup.
+	const normalized = regionText.replace(/\s+/g, " ").trim();
+	return createHash("sha256").update(normalized).digest("hex").slice(0, 16);
 }
 
 export class VectorStore {
-  // These fields are `readonly` (set once in the constructor) but NOT private:
-  // the read/search/dedup helpers split into vector-read.ts, vector-search.ts,
-  // and vector-dedup.ts access them directly. Marking them private would force
-  // ugly `as unknown as` casts in those modules; keeping them package-public
-  // makes VectorStore a thin barrel whose helpers live in sibling files.
-  readonly embedder: Embedder;
-  readonly stateDir: string;
-  private readonly l2Threshold: number;
-  /** Single source of truth for tier flags + thresholds (Sprint 14). */
-  readonly cfg: DedupConfigShape;
-  /** Optional monitoring target (Sprint 14). Undefined → no monitoring. */
-  private readonly eventsPath?: string;
-  /**
-   * Repo key for the async PGlite vector index (Slice 2). We use the stateDir
-   * itself as the repo id — it is already unique per repo and available here
-   * without crossing into the pi-runtime layer (src/ stays pi-agnostic). The
-   * global index keys on repoId so recall can span repos.
-   */
-  readonly repoId: string;
+	// These fields are `readonly` (set once in the constructor) but NOT private:
+	// the read/search/dedup helpers split into vector-read.ts, vector-search.ts,
+	// and vector-dedup.ts access them directly. Marking them private would force
+	// ugly `as unknown as` casts in those modules; keeping them package-public
+	// makes VectorStore a thin barrel whose helpers live in sibling files.
+	readonly embedder: Embedder;
+	readonly stateDir: string;
+	private readonly l2Threshold: number;
+	/** Single source of truth for tier flags + thresholds (Sprint 14). */
+	readonly cfg: DedupConfigShape;
+	/** Optional monitoring target (Sprint 14). Undefined → no monitoring. */
+	private readonly eventsPath?: string;
+	/**
+	 * Repo key for the async PGlite vector index (Slice 2). We use the stateDir
+	 * itself as the repo id — it is already unique per repo and available here
+	 * without crossing into the pi-runtime layer (src/ stays pi-agnostic). The
+	 * global index keys on repoId so recall can span repos.
+	 */
+	readonly repoId: string;
+	/** S25: per-session cached rehydrated RaptorTree. Keyed by sessionId.
+	 *  Freshness-validated via maxRaptorNodeBuiltAt on each lookup — a cheap
+	 *  indexed MAX query replaces the O(n·leaves) rehydrate on every search. */
+	readonly raptorCache = new Map<
+		string,
+		{ tree: RaptorTree; builtAt: number }
+	>();
 
-  constructor(
-    opts: {
-      embedder?: Embedder;
-      dedupSim?: number;
-      stateDir?: string;
-      l2Enabled?: boolean;
-      l2Threshold?: number;
-      /** Override the dedup config (defaults to env/file snapshot). */
-      config?: DedupConfigShape;
-      /** Optional events.log path for decision monitoring (Sprint 14). */
-      eventsPath?: string;
-      /** Repo id for the async cross-repo vector index. Defaults to stateDir. */
-      repoId?: string;
-    } = {},
-  ) {
-    this.embedder = opts.embedder ?? defaultEmbedder();
-    this.stateDir = opts.stateDir ?? getStateDir();
-    this.repoId = opts.repoId ?? this.stateDir;
-    // Sprint 14: all tier flags/thresholds flow from the single config source
-    // (DedupConfig). The legacy opts.dedupSim / opts.l2Enabled remain accepted
-    // for backward-compat callers but flags are authoritative via `cfg`.
-    void opts.dedupSim;
-    void opts.l2Enabled;
-    // Sprint 12 L2 semantic tier. Threshold 0.85 is the default trigram
-    // embedder's honest firing point; a direct override is allowed for tests.
-    this.cfg = opts.config ?? loadDedupConfig();
-    this.l2Threshold = opts.l2Threshold ?? this.cfg.L2_COSINE;
-    this.eventsPath = opts.eventsPath;
-    // Sprint 8: bring any v0.1.0 JSON checkpoint files into SQLite (idempotent).
-    migrateJsonToSqlite(this.stateDir);
-    // Sprint 10: warm the bloom accelerator (accelerator only — SQLite stays
-    // source of truth; a bloom hit is always confirmed by a query below).
-    openBloom(this.stateDir);
-  }
+	constructor(
+		opts: {
+			embedder?: Embedder;
+			dedupSim?: number;
+			stateDir?: string;
+			l2Enabled?: boolean;
+			l2Threshold?: number;
+			/** Override the dedup config (defaults to env/file snapshot). */
+			config?: DedupConfigShape;
+			/** Optional events.log path for decision monitoring (Sprint 14). */
+			eventsPath?: string;
+			/** Repo id for the async cross-repo vector index. Defaults to stateDir. */
+			repoId?: string;
+		} = {},
+	) {
+		this.embedder = opts.embedder ?? defaultEmbedder();
+		this.stateDir = opts.stateDir ?? getStateDir();
+		this.repoId = opts.repoId ?? this.stateDir;
+		// Sprint 14: all tier flags/thresholds flow from the single config source
+		// (DedupConfig). The legacy opts.dedupSim / opts.l2Enabled remain accepted
+		// for backward-compat callers but flags are authoritative via `cfg`.
+		void opts.dedupSim;
+		void opts.l2Enabled;
+		// Sprint 12 L2 semantic tier. Threshold 0.85 is the default trigram
+		// embedder's honest firing point; a direct override is allowed for tests.
+		this.cfg = opts.config ?? loadDedupConfig();
+		this.l2Threshold = opts.l2Threshold ?? this.cfg.L2_COSINE;
+		this.eventsPath = opts.eventsPath;
+		// Sprint 8: bring any v0.1.0 JSON checkpoint files into SQLite (idempotent).
+		migrateJsonToSqlite(this.stateDir);
+		// Sprint 10: warm the bloom accelerator (accelerator only — SQLite stays
+		// source of truth; a bloom hit is always confirmed by a query below).
+		openBloom(this.stateDir);
+	}
 
-  /** Emit a structured dedup-decision event (best-effort, never throws). */
-  record(tier: DedupTier, result: "deduped" | "new" | "mark_only", reason: string | undefined, latencyMs: number): void {
-    if (!this.eventsPath) return;
-    logDecision(this.eventsPath, {
-      ts: Date.now(),
-      tier,
-      result,
-      reason,
-      latencyMs: Math.round(latencyMs * 100) / 100,
-    });
-  }
+	/** Emit a structured dedup-decision event (best-effort, never throws). */
+	record(
+		tier: DedupTier,
+		result: "deduped" | "new" | "mark_only",
+		reason: string | undefined,
+		latencyMs: number,
+	): void {
+		if (!this.eventsPath) return;
+		logDecision(this.eventsPath, {
+			ts: Date.now(),
+			tier,
+			result,
+			reason,
+			latencyMs: Math.round(latencyMs * 100) / 100,
+		});
+	}
 
-  /**
-   * Add a checkpoint. Dedup cascade:
-   *   1. regionHash exact match (legacy, backward-compat)
-   *   2. summaryHash exact match (new: catches same-topic incremental compactions)
-   *   3. content similarity ≥ dedupSim (catches near-identical summaries)
-   *   4. If none match → create new checkpoint
-   */
-  add(input: AddInput): AddResult {
-    const t0 = Date.now();
-    const sessionId = normalizeSessionId(input.sessionId);
-    const regionHash = computeRegionHash(input.regionText);
-    const all = listCheckpoints(sessionId, this.stateDir);
-    // Honest "tokens saved" base for this region. For a deduped add the whole
-    // original region is discarded (nothing new stored); for a new checkpoint
-    // we persist (orig − stored). Falls back to stored when orig is unknown.
-    const origTokens = input.originalTokenEstimate ?? input.tokenEstimate ?? 0;
-    const cfg = this.cfg;
-    // Live per-tier progress hook (Phase 1). Sync + optional; fired at each tier
-    // so the UI can paint "L0 ✓ → L1 ✓ → L2 0.91 → stored" during a compaction.
-    const onTier = input.onTier;
-    // Tracks whether a tier matched while in MARK_ONLY (record-but-don't-collapse),
-    // and which tier.
-    let markOnly: DedupTier | null = null;
+	/**
+	 * Add a checkpoint. Dedup cascade:
+	 *   1. regionHash exact match (legacy, backward-compat)
+	 *   2. summaryHash exact match (new: catches same-topic incremental compactions)
+	 *   3. content similarity ≥ dedupSim (catches near-identical summaries)
+	 *   4. If none match → create new checkpoint
+	 */
+	add(input: AddInput): AddResult {
+		const t0 = Date.now();
+		const sessionId = normalizeSessionId(input.sessionId);
+		const regionHash = computeRegionHash(input.regionText);
+		const all = listCheckpoints(sessionId, this.stateDir);
+		// Honest "tokens saved" base for this region. For a deduped add the whole
+		// original region is discarded (nothing new stored); for a new checkpoint
+		// we persist (orig − stored). Falls back to stored when orig is unknown.
+		const origTokens = input.originalTokenEstimate ?? input.tokenEstimate ?? 0;
+		const cfg = this.cfg;
+		// Live per-tier progress hook (Phase 1). Sync + optional; fired at each tier
+		// so the UI can paint "L0 ✓ → L1 ✓ → L2 0.91 → stored" during a compaction.
+		const onTier = input.onTier;
+		// Tracks whether a tier matched while in MARK_ONLY (record-but-don't-collapse),
+		// and which tier.
+		let markOnly: DedupTier | null = null;
 
-    // 0. L0 content-hash dedup (Sprint 9) — catches identical content arriving
-    //    under different regionText. Normalization handles case/whitespace/ANSI so
-    //    variants collapse to one row. Dual-hash guards a single-hash collision.
-    //    Sprint 10: bloom is the accelerator — a miss means "definitely new" and
-    //    skips the scan; a hit is only a candidate, confirmed against `all` below.
-    //    Gated by L0_ENABLED (Sprint 14). MARK_ONLY_L0 records the decision but
-    //    does not collapse — the new region is still stored.
-    onTier?.({ tier: "L0", status: "scanning" });
-    const digest = computeContentDigest(input.regionText);
-    const bloom = openBloom(this.stateDir);
-    if (cfg.L0_ENABLED && bloom.maybeHas(digest.contentHash)) {
-      const contentMatch = all.find(
-        (cp) =>
-          cp.contentHash === digest.contentHash &&
-          cp.contentHash2 === digest.contentHash2,
-      );
-      if (contentMatch) {
-        if (cfg.MARK_ONLY_L0) {
-          markOnly = "L0"; // Record-but-don't-collapse: fall through.
-        } else {
-          contentMatch.timestamp = input.timestamp;
-          upsertCheckpoint(contentMatch, this.stateDir);
-          bumpDedupStats(true, this.stateDir);
-          // Deduped: whole original region discarded, nothing new stored.
-          addTokensSaved(origTokens, this.stateDir);
-          const r = { checkpoint: contentMatch, deduped: true, reason: "contentHash" };
-          this.record("L0", "deduped", "contentHash", Date.now() - t0);
-          onTier?.({ tier: "L0", status: "deduped", detail: "contentHash" });
-          return r;
-        }
-      }
-    }
+		// 0. L0 content-hash dedup (Sprint 9) — catches identical content arriving
+		//    under different regionText. Normalization handles case/whitespace/ANSI so
+		//    variants collapse to one row. Dual-hash guards a single-hash collision.
+		//    Sprint 10: bloom is the accelerator — a miss means "definitely new" and
+		//    skips the scan; a hit is only a candidate, confirmed against `all` below.
+		//    Gated by L0_ENABLED (Sprint 14). MARK_ONLY_L0 records the decision but
+		//    does not collapse — the new region is still stored.
+		onTier?.({ tier: "L0", status: "scanning" });
+		const digest = computeContentDigest(input.regionText);
+		const bloom = openBloom(this.stateDir);
+		if (cfg.L0_ENABLED && bloom.maybeHas(digest.contentHash)) {
+			const contentMatch = all.find(
+				(cp) =>
+					cp.contentHash === digest.contentHash &&
+					cp.contentHash2 === digest.contentHash2,
+			);
+			if (contentMatch) {
+				if (cfg.MARK_ONLY_L0) {
+					markOnly = "L0"; // Record-but-don't-collapse: fall through.
+				} else {
+					contentMatch.timestamp = input.timestamp;
+					upsertCheckpoint(contentMatch, this.stateDir);
+					bumpDedupStats(true, this.stateDir);
+					// Deduped: whole original region discarded, nothing new stored.
+					addTokensSaved(origTokens, this.stateDir);
+					const r = {
+						checkpoint: contentMatch,
+						deduped: true,
+						reason: "contentHash",
+					};
+					this.record("L0", "deduped", "contentHash", Date.now() - t0);
+					onTier?.({ tier: "L0", status: "deduped", detail: "contentHash" });
+					return r;
+				}
+			}
+		}
 
-    // 1. Legacy regionHash dedup (backward-compat) — part of L0 tier gating.
-    if (cfg.L0_ENABLED) {
-      const regionMatch = all.find((cp) => cp.regionHash === regionHash);
-      if (regionMatch) {
-        if (cfg.MARK_ONLY_L0) {
-          markOnly = "L0"; // fall through
-        } else {
-          bumpDedupStats(true, this.stateDir);
-          // Deduped: whole original region discarded, nothing new stored.
-          addTokensSaved(origTokens, this.stateDir);
-          const r = { checkpoint: regionMatch, deduped: true, reason: "regionHash" };
-          this.record("L0", "deduped", "regionHash", Date.now() - t0);
-          onTier?.({ tier: "L0", status: "deduped", detail: "regionHash" });
-          return r;
-        }
-      }
-    }
+		// 1. Legacy regionHash dedup (backward-compat) — part of L0 tier gating.
+		if (cfg.L0_ENABLED) {
+			const regionMatch = all.find((cp) => cp.regionHash === regionHash);
+			if (regionMatch) {
+				if (cfg.MARK_ONLY_L0) {
+					markOnly = "L0"; // fall through
+				} else {
+					bumpDedupStats(true, this.stateDir);
+					// Deduped: whole original region discarded, nothing new stored.
+					addTokensSaved(origTokens, this.stateDir);
+					const r = {
+						checkpoint: regionMatch,
+						deduped: true,
+						reason: "regionHash",
+					};
+					this.record("L0", "deduped", "regionHash", Date.now() - t0);
+					onTier?.({ tier: "L0", status: "deduped", detail: "regionHash" });
+					return r;
+				}
+			}
+		}
 
-    // 2. SummaryHash dedup — catches same-topic incremental compactions.
-    //    Full 64-hex SHA-256 (was 16-hex in Sprint 8 — collision-prone).
-    const summaryHash = input.topicSummary
-      ? createHash("sha256").update(input.topicSummary).digest("hex")
-      : undefined;
-    if (summaryHash && cfg.L0_ENABLED) {
-      const summaryMatch = all.find((cp) => cp.summaryHash === summaryHash);
-      if (summaryMatch) {
-        if (cfg.MARK_ONLY_L0) {
-          markOnly = "L0"; // fall through
-        } else {
-          summaryMatch.timestamp = input.timestamp;
-          upsertCheckpoint(summaryMatch, this.stateDir);
-          bumpDedupStats(true, this.stateDir);
-          // Deduped: whole original region discarded, nothing new stored.
-          addTokensSaved(origTokens, this.stateDir);
-          const r = { checkpoint: summaryMatch, deduped: true, reason: "summaryHash" };
-          this.record("L0", "deduped", "summaryHash", Date.now() - t0);
-          onTier?.({ tier: "L0", status: "deduped", detail: "summaryHash" });
-          return r;
-        }
-      }
-    }
-    // L0 did not collapse this region.
-    onTier?.({ tier: "L0", status: "passed" });
+		// 2. SummaryHash dedup — catches same-topic incremental compactions.
+		//    Full 64-hex SHA-256 (was 16-hex in Sprint 8 — collision-prone).
+		const summaryHash = input.topicSummary
+			? createHash("sha256").update(input.topicSummary).digest("hex")
+			: undefined;
+		if (summaryHash && cfg.L0_ENABLED) {
+			const summaryMatch = all.find((cp) => cp.summaryHash === summaryHash);
+			if (summaryMatch) {
+				if (cfg.MARK_ONLY_L0) {
+					markOnly = "L0"; // fall through
+				} else {
+					summaryMatch.timestamp = input.timestamp;
+					upsertCheckpoint(summaryMatch, this.stateDir);
+					bumpDedupStats(true, this.stateDir);
+					// Deduped: whole original region discarded, nothing new stored.
+					addTokensSaved(origTokens, this.stateDir);
+					const r = {
+						checkpoint: summaryMatch,
+						deduped: true,
+						reason: "summaryHash",
+					};
+					this.record("L0", "deduped", "summaryHash", Date.now() - t0);
+					onTier?.({ tier: "L0", status: "deduped", detail: "summaryHash" });
+					return r;
+				}
+			}
+		}
+		// L0 did not collapse this region.
+		onTier?.({ tier: "L0", status: "passed" });
 
-    // 2b. L1 MinHash/LSH near-duplicate dedup (Sprint 11) — catches one-word
-    //     edits / rewordings that L0's exact hash misses. Cheap LSH bucket
-    //     retrieval → trigram verification (pg_trgm-equivalent) as the final gate.
-    //     Gated by L1_ENABLED (Sprint 14); MARK_ONLY_L1 records but doesn't collapse.
-    onTier?.({ tier: "L1", status: "scanning" });
-    if (cfg.L1_ENABLED) {
-      const l1 = this.findL1Duplicate(sessionId, input.regionText, all);
-      if (l1 && !cfg.MARK_ONLY_L1) {
-        l1.timestamp = input.timestamp;
-        upsertCheckpoint(l1, this.stateDir);
-        bumpDedupStats(true, this.stateDir);
-        const r = { checkpoint: l1, deduped: true, reason: "l1MinHash" };
-        this.record("L1", "deduped", "l1MinHash", Date.now() - t0);
-        onTier?.({ tier: "L1", status: "deduped", detail: "l1MinHash" });
-        return r;
-      }
-      if (l1 && cfg.MARK_ONLY_L1) markOnly = "L1";
-    }
-    onTier?.({ tier: "L1", status: "passed" });
+		// 2b. L1 MinHash/LSH near-duplicate dedup (Sprint 11) — catches one-word
+		//     edits / rewordings that L0's exact hash misses. Cheap LSH bucket
+		//     retrieval → trigram verification (pg_trgm-equivalent) as the final gate.
+		//     Gated by L1_ENABLED (Sprint 14); MARK_ONLY_L1 records but doesn't collapse.
+		onTier?.({ tier: "L1", status: "scanning" });
+		if (cfg.L1_ENABLED) {
+			const l1 = this.findL1Duplicate(sessionId, input.regionText, all);
+			if (l1 && !cfg.MARK_ONLY_L1) {
+				l1.timestamp = input.timestamp;
+				upsertCheckpoint(l1, this.stateDir);
+				bumpDedupStats(true, this.stateDir);
+				const r = { checkpoint: l1, deduped: true, reason: "l1MinHash" };
+				this.record("L1", "deduped", "l1MinHash", Date.now() - t0);
+				onTier?.({ tier: "L1", status: "deduped", detail: "l1MinHash" });
+				return r;
+			}
+			if (l1 && cfg.MARK_ONLY_L1) markOnly = "L1";
+		}
+		onTier?.({ tier: "L1", status: "passed" });
 
-    // 3. L2 semantic dedup — catches near-identical / semantically-similar regions
-    //    via cosine over the embedding. topicSummary is used for summaryHash dedup
-    //    (tier 2); the vector index is keyed on the original region for backward-
-    //    compat search semantics. Threshold from cfg (L2_COSINE trigram honest
-    //    firing point). QA #13 timeout guard: if the O(n) scan exceeds the budget,
-    //    degrade to "store without dedup this pass" so we never lose a checkpoint.
-    //    Gated by L2_ENABLED (Sprint 14); MARK_ONLY_L2 records but doesn't collapse.
-    const SIMILARITY_BUDGET_MS = cfg.SIMILARITY_BUDGET_MS;
-    const simThreshold = this.l2Threshold; // from cfg.L2_COSINE (default 0.85 trigram)
-    const embedding = this.embedder.embed(input.regionText);
-    onTier?.({ tier: "L2", status: "scanning" });
-    if (cfg.L2_ENABLED && all.length > 0) {
-      const start = Date.now();
-      let timedOut = false;
-      const nearest = all.reduce(
-        (best, cp) => {
-          if (!timedOut && Date.now() - start > SIMILARITY_BUDGET_MS) timedOut = true;
-          if (timedOut) return best;
-          const sim = cosineSimilarity(embedding, cp.embedding);
-          return sim > best.sim ? { checkpoint: cp, sim } : best;
-        },
-        { checkpoint: all[0], sim: -1 },
-      );
-      if (!timedOut && nearest.sim >= simThreshold) {
-        if (!cfg.MARK_ONLY_L2) {
-          // Near-identical — update timestamp on existing checkpoint
-          nearest.checkpoint.timestamp = input.timestamp;
-          upsertCheckpoint(nearest.checkpoint, this.stateDir);
-          bumpDedupStats(true, this.stateDir);
-          // Deduped: whole original region discarded, nothing new stored.
-          addTokensSaved(origTokens, this.stateDir);
-          const r = { checkpoint: nearest.checkpoint, deduped: true, reason: "contentSimilarity" };
-          this.record("L2", "deduped", "contentSimilarity", Date.now() - t0);
-          onTier?.({ tier: "L2", status: "deduped", detail: nearest.sim.toFixed(2) });
-          return r;
-        }
-        markOnly = "L2";
-      }
-      onTier?.({ tier: "L2", status: "passed", detail: `best ${nearest.sim.toFixed(2)}` });
-    }
+		// 3. L2 semantic dedup — catches near-identical / semantically-similar regions
+		//    via cosine over the embedding. topicSummary is used for summaryHash dedup
+		//    (tier 2); the vector index is keyed on the original region for backward-
+		//    compat search semantics. Threshold from cfg (L2_COSINE trigram honest
+		//    firing point). QA #13 timeout guard: if the O(n) scan exceeds the budget,
+		//    degrade to "store without dedup this pass" so we never lose a checkpoint.
+		//    Gated by L2_ENABLED (Sprint 14); MARK_ONLY_L2 records but doesn't collapse.
+		const SIMILARITY_BUDGET_MS = cfg.SIMILARITY_BUDGET_MS;
+		const simThreshold = this.l2Threshold; // from cfg.L2_COSINE (default 0.85 trigram)
+		const embedding = this.embedder.embed(input.regionText);
+		onTier?.({ tier: "L2", status: "scanning" });
+		if (cfg.L2_ENABLED && all.length > 0) {
+			const start = Date.now();
+			let timedOut = false;
+			const nearest = all.reduce(
+				(best, cp) => {
+					if (!timedOut && Date.now() - start > SIMILARITY_BUDGET_MS)
+						timedOut = true;
+					if (timedOut) return best;
+					const sim = cosineSimilarity(embedding, cp.embedding);
+					return sim > best.sim ? { checkpoint: cp, sim } : best;
+				},
+				{ checkpoint: all[0], sim: -1 },
+			);
+			if (!timedOut && nearest.sim >= simThreshold) {
+				if (!cfg.MARK_ONLY_L2) {
+					// Near-identical — update timestamp on existing checkpoint
+					nearest.checkpoint.timestamp = input.timestamp;
+					upsertCheckpoint(nearest.checkpoint, this.stateDir);
+					bumpDedupStats(true, this.stateDir);
+					// Deduped: whole original region discarded, nothing new stored.
+					addTokensSaved(origTokens, this.stateDir);
+					const r = {
+						checkpoint: nearest.checkpoint,
+						deduped: true,
+						reason: "contentSimilarity",
+					};
+					this.record("L2", "deduped", "contentSimilarity", Date.now() - t0);
+					onTier?.({
+						tier: "L2",
+						status: "deduped",
+						detail: nearest.sim.toFixed(2),
+					});
+					return r;
+				}
+				markOnly = "L2";
+			}
+			onTier?.({
+				tier: "L2",
+				status: "passed",
+				detail: `best ${nearest.sim.toFixed(2)}`,
+			});
+		}
 
-    // 4. Genuinely new — create checkpoint
-    const checkpointId = nextCheckpointId(sessionId, this.stateDir);
-    const checkpoint: StoredCheckpoint = {
-      checkpointId,
-      sessionId,
-      repoId: this.repoId,
-      summary: input.summary,
-      topicSummary: input.topicSummary,
-      summaryHash,
-      keyDecisions: input.keyDecisions ?? [],
-      nextSteps: input.nextSteps ?? [],
-      filesModified: input.filesModified ?? [],
-      tokenEstimate: input.tokenEstimate ?? 0,
-      originalTokenEstimate: input.originalTokenEstimate,
-      regionHash,
-      contentHash: digest.contentHash,
-      contentHash2: digest.contentHash2,
-      contentHashVersion: digest.contentHashVersion,
-      normalizedText: digest.normalizedText,
-      compressedOriginal: compressSmart(
-        Buffer.from(input.regionText, "utf-8"),
-        input.compressionPressure,
-      ),
-      embedding,
-      timestamp: input.timestamp,
-    };
-    // Persistence is SQLite (store/sqlite.ts). upsertCheckpoint keeps the
-    // idempotent-by-id semantics the old JSON append implied.
-    upsertCheckpoint(checkpoint, this.stateDir);
-    // Cumulative "tokens saved" counter (per-repo SQLite meta). For a NEW
-    // checkpoint the saved amount is (original − stored); for a deduped add the
-    // whole original region is discarded (handled in the deduped return paths
-    // below). Survives sessions and travels with the repo.
-    const stored = input.tokenEstimate ?? 0;
-    addTokensSaved(Math.max(0, origTokens - stored), this.stateDir);
-    // L1: persist this checkpoint's MinHash signature + LSH buckets so future
-    // near-duplicate inserts can find it. Deterministic given the seed.
-    const sig = minhashSignature(input.regionText);
-    upsertMinhashSignature(checkpointId, sessionId, SIGNATURE_VERSION, sig, this.stateDir);
-    insertLshBuckets(
-      checkpointId,
-      sessionId,
-      SIGNATURE_VERSION,
-      lshBands(sig, sessionId, SIGNATURE_VERSION),
-      this.stateDir,
-    );
-    // Bloom accelerator: record the new content_hash so a future add() can short-
-    // circuit the scan on a hit (still confirmed by the SELECT-based `all` above).
-    bloom.add(digest.contentHash);
-    saveBloom(this.stateDir);
+		// 4. Genuinely new — create checkpoint
+		const checkpointId = nextCheckpointId(sessionId, this.stateDir);
+		const checkpoint: StoredCheckpoint = {
+			checkpointId,
+			sessionId,
+			repoId: this.repoId,
+			summary: input.summary,
+			topicSummary: input.topicSummary,
+			summaryHash,
+			keyDecisions: input.keyDecisions ?? [],
+			nextSteps: input.nextSteps ?? [],
+			filesModified: input.filesModified ?? [],
+			tokenEstimate: input.tokenEstimate ?? 0,
+			originalTokenEstimate: input.originalTokenEstimate,
+			regionHash,
+			contentHash: digest.contentHash,
+			contentHash2: digest.contentHash2,
+			contentHashVersion: digest.contentHashVersion,
+			normalizedText: digest.normalizedText,
+			compressedOriginal: compressSmart(
+				Buffer.from(input.regionText, "utf-8"),
+				input.compressionPressure,
+			),
+			embedding,
+			timestamp: input.timestamp,
+		};
+		// Persistence is SQLite (store/sqlite.ts). upsertCheckpoint keeps the
+		// idempotent-by-id semantics the old JSON append implied.
+		upsertCheckpoint(checkpoint, this.stateDir);
+		// Cumulative "tokens saved" counter (per-repo SQLite meta). For a NEW
+		// checkpoint the saved amount is (original − stored); for a deduped add the
+		// whole original region is discarded (handled in the deduped return paths
+		// below). Survives sessions and travels with the repo.
+		const stored = input.tokenEstimate ?? 0;
+		addTokensSaved(Math.max(0, origTokens - stored), this.stateDir);
+		// L1: persist this checkpoint's MinHash signature + LSH buckets so future
+		// near-duplicate inserts can find it. Deterministic given the seed.
+		const sig = minhashSignature(input.regionText);
+		upsertMinhashSignature(
+			checkpointId,
+			sessionId,
+			SIGNATURE_VERSION,
+			sig,
+			this.stateDir,
+		);
+		insertLshBuckets(
+			checkpointId,
+			sessionId,
+			SIGNATURE_VERSION,
+			lshBands(sig, sessionId, SIGNATURE_VERSION),
+			this.stateDir,
+		);
+		// Bloom accelerator: record the new content_hash so a future add() can short-
+		// circuit the scan on a hit (still confirmed by the SELECT-based `all` above).
+		bloom.add(digest.contentHash);
+		saveBloom(this.stateDir);
 
-    // Track the region hash in session state for fast sentinel checks.
-    const state = loadSessionState(sessionId, this.stateDir);
-    if (!state.storedRegionHashes.includes(regionHash)) {
-      state.storedRegionHashes.push(regionHash);
-      saveSessionState(sessionId, state, this.stateDir);
-    }
-    // A new checkpoint. If a tier matched while MARK_ONLY, record that (the
-    // decision fired but we intentionally did not collapse).
-    if (markOnly) {
-      this.record(markOnly, "mark_only", "mark_only", Date.now() - t0);
-    } else {
-      this.record("L0", "new", undefined, Date.now() - t0);
-    }
-    // Cumulative store-wide dedup accounting (attempt, not collapsed).
-    bumpDedupStats(false, this.stateDir);
-    onTier?.({ tier: "new", status: "stored" });
-    return { checkpoint, deduped: false };
-  }
+		// Track the region hash in session state for fast sentinel checks.
+		const state = loadSessionState(sessionId, this.stateDir);
+		if (!state.storedRegionHashes.includes(regionHash)) {
+			state.storedRegionHashes.push(regionHash);
+			saveSessionState(sessionId, state, this.stateDir);
+		}
+		// A new checkpoint. If a tier matched while MARK_ONLY, record that (the
+		// decision fired but we intentionally did not collapse).
+		if (markOnly) {
+			this.record(markOnly, "mark_only", "mark_only", Date.now() - t0);
+		} else {
+			this.record("L0", "new", undefined, Date.now() - t0);
+		}
+		// Cumulative store-wide dedup accounting (attempt, not collapsed).
+		bumpDedupStats(false, this.stateDir);
+		onTier?.({ tier: "new", status: "stored" });
+		return { checkpoint, deduped: false };
+	}
 
-  /**
-   * L1 near-duplicate lookup: MinHash → LSH candidate retrieval → trigram verify.
-   * Returns the matching checkpoint or undefined. Bounded by a 100-candidate cap
-   * and a 20ms verify budget (QA #7/#15) so it never hangs a large session.
-   */
-  private findL1Duplicate(
-    sessionId: string,
-    regionText: string,
-    all: StoredCheckpoint[],
-  ): StoredCheckpoint | undefined {
-    if (all.length === 0) return undefined;
-    const sig = minhashSignature(regionText);
-    if (sig.length !== NUM_HASHES) return undefined;
-    const bands = lshBands(sig, sessionId, SIGNATURE_VERSION);
-    // Cheap candidate retrieval (single query, capped). Exclude nothing yet —
-    // the new checkpoint has no id, so pass a sentinel that never matches.
-    const candidateIds = lshCandidateChunks(
-      bands,
-      sessionId,
-      "__new__",
-      this.stateDir,
-      100,
-    );
-    if (candidateIds.length === 0) return undefined;
-    const byId = new Map(all.map((cp) => [cp.checkpointId, cp]));
-    const VERIFY_BUDGET_MS = 20;
-    const start = Date.now();
-    for (const id of candidateIds) {
-      if (Date.now() - start > VERIFY_BUDGET_MS) break; // QA #15: abort → "not dup"
-      const cand = byId.get(id);
-      if (!cand) continue;
-      const candText = cand.normalizedText ?? cand.summary ?? "";
-      if (isNearDuplicate(regionText, candText)) return cand;
-    }
-    return undefined;
-  }
+	/**
+	 * L1 near-duplicate lookup: MinHash → LSH candidate retrieval → trigram verify.
+	 * Returns the matching checkpoint or undefined. Bounded by a 100-candidate cap
+	 * and a 20ms verify budget (QA #7/#15) so it never hangs a large session.
+	 */
+	private findL1Duplicate(
+		sessionId: string,
+		regionText: string,
+		all: StoredCheckpoint[],
+	): StoredCheckpoint | undefined {
+		if (all.length === 0) return undefined;
+		const sig = minhashSignature(regionText);
+		if (sig.length !== NUM_HASHES) return undefined;
+		const bands = lshBands(sig, sessionId, SIGNATURE_VERSION);
+		// Cheap candidate retrieval (single query, capped). Exclude nothing yet —
+		// the new checkpoint has no id, so pass a sentinel that never matches.
+		const candidateIds = lshCandidateChunks(
+			bands,
+			sessionId,
+			"__new__",
+			this.stateDir,
+			100,
+		);
+		if (candidateIds.length === 0) return undefined;
+		const byId = new Map(all.map((cp) => [cp.checkpointId, cp]));
+		const VERIFY_BUDGET_MS = 20;
+		const start = Date.now();
+		for (const id of candidateIds) {
+			if (Date.now() - start > VERIFY_BUDGET_MS) break; // QA #15: abort → "not dup"
+			const cand = byId.get(id);
+			if (!cand) continue;
+			const candText = cand.normalizedText ?? cand.summary ?? "";
+			if (isNearDuplicate(regionText, candText)) return cand;
+		}
+		return undefined;
+	}
 }
 
 // Re-exports (back-compat): existing call sites keep importing from "./vectorStore.js"
 export {
-  vectorSemDedup,
-  vectorDedupe,
-  vectorMarkInjected,
-  vectorWasInjected,
-  vectorSimilarity,
-  vectorList,
-  vectorTopSimilar,
-  vectorStats,
-  vectorRepoStats,
-  vectorDataInvariant,
+	vectorSemDedup,
+	vectorDedupe,
+	vectorMarkInjected,
+	vectorWasInjected,
+	vectorSimilarity,
+	vectorList,
+	vectorTopSimilar,
+	vectorStats,
+	vectorRepoStats,
+	vectorDataInvariant,
 } from "./vector-read.js";
 
 // Re-exports: vectorSearch / vectorSearchAsync (moved to vector-search.ts)