@snoai/memory 1.5.1 → 1.7.0

package/dist/{retriever-CsBW6MZv.js → retriever-C54bXeDh.js}

@@ -1,4 +1,4 @@
-import { CHUNKING_VERSION, DEFAULT_BM25_WEIGHT, DEFAULT_RECALL_LIFECYCLE, DEFAULT_RERANK_TIMEOUT_MS, DEFAULT_VECTOR_WEIGHT, IMPORTANCE_WEIGHT_BASE, LIGHTWEIGHT_COSINE_WEIGHT, LIGHTWEIGHT_FUSION_WEIGHT, LIGHTWEIGHT_RERANK_PENALTY, MMR_LAMBDA, RECENCY_WEIGHT, RERANK_BLEND_CROSS, RERANK_BLEND_VECTOR, TIME_DECAY_FLOOR, experimentCandidatePoolOverride, experimentMmrDisabled } from "./internal/config/index.js";
+import { CHUNKING_VERSION, DEFAULT_BM25_WEIGHT, DEFAULT_RECALL_LIFECYCLE, DEFAULT_RERANK_TIMEOUT_MS, DEFAULT_VECTOR_WEIGHT, IMPORTANCE_WEIGHT_BASE, MMR_LAMBDA, RECENCY_WEIGHT, TIME_DECAY_FLOOR, experimentCandidatePoolOverride, experimentMmrDisabled } from "./internal/config/index.js";
 import "./internal/engine/shared/types.js";
 import { EmbeddingError, RetrievalError } from "./internal/engine/shared/errors.js";
 import { o as isMemoryExpired, s as parseInsightMetadata } from "./memory-metadata-codec-DL1sbJiN.js";
@@ -7,7 +7,7 @@ import { a as memoryOperationSignal } from "./operation-cancellation-QcwC4mDX.js
 import { appendQaTrace, computeConfigHash, isTraceEnabled } from "./internal/engine/eval/trace.js";
 import { computeEffectiveHalfLife, parseAccessMetadata } from "./internal/engine/retrieval/access-tracker.js";
 import { createLogger as createLogger$1, currentLogContext, privateLogReference, withLogContext } from "@snoai/utils/logger";
-import { createHash, randomUUID } from "node:crypto";
+import { randomUUID } from "node:crypto";
 //#region src/engine/retrieval/retrieval-trace.ts
 var StageLifecycle = class StageLifecycle {
 	name;
@@ -212,7 +212,7 @@ const DEFAULT_RETRIEVAL_CONFIG = {
 	vectorWeight: DEFAULT_VECTOR_WEIGHT,
 	bm25Weight: DEFAULT_BM25_WEIGHT,
 	minScore: 0,
-	rerank: "cross-encoder",
+	rerank: "lightweight",
 	candidatePoolSize: 64,
 	recencyHalfLifeDays: 14,
 	recencyWeight: RECENCY_WEIGHT,
@@ -222,10 +222,8 @@ const DEFAULT_RETRIEVAL_CONFIG = {
 	lengthNormAnchor: 0,
 	hardMinScore: 0,
 	timeDecayHalfLifeDays: 60,
-	rerankBlendVector: RERANK_BLEND_VECTOR,
-	rerankBlendCross: RERANK_BLEND_CROSS,
-	lightweightFusionWeight: LIGHTWEIGHT_FUSION_WEIGHT,
-	lightweightCosineWeight: LIGHTWEIGHT_COSINE_WEIGHT,
+	lightweightFusionWeight: 0,
+	lightweightCosineWeight: 1,
 	importanceWeightBase: IMPORTANCE_WEIGHT_BASE,
 	timeDecayFloor: TIME_DECAY_FLOOR,
 	mmrLambda: MMR_LAMBDA,
@@ -233,31 +231,6 @@ const DEFAULT_RETRIEVAL_CONFIG = {
 };
 //#endregion
 //#region src/engine/retrieval/retriever-core.ts
-/**
-* Enumerates every branch in `rerank()` that fell back to the pre-rerank
-* candidate ordering instead of applying the cross-encoder score. Aggregated
-* via {@link RetrievalStatsCollector} and surfaced as trace stage metadata
-* so callers can attribute degraded ranking without parsing log lines.
-*
-* - `missing_api_key`: cross-encoder configured but no rerankApiKey set;
-*   call routes through the lightweight cosine path instead.
-* - `no_endpoint`: no reranker address (`rerankEndpoint`) was set; rerank was skipped.
-* - `http_error`: the reranker responded with a non-OK status.
-* - `invalid_response`: the reranker answered HTTP 2xx but the response shape did not parse.
-* - `timeout`: the per-call AbortSignal fired before the reranker responded.
-* - `request_error`: any other thrown error inside the fetch path.
-* - `query_over_window`: the query alone fills the reranker's per-pair token window, so no
-*   document could be sent with it. Nothing was requested.
-*/
-const RERANK_FALLBACK_REASONS = [
-	"missing_api_key",
-	"no_endpoint",
-	"http_error",
-	"invalid_response",
-	"timeout",
-	"request_error",
-	"query_over_window"
-];
 var MemoryRetriever = class {
 	store;
 	embedder;
@@ -547,11 +520,7 @@ Object.assign(MemoryRetriever.prototype, {
 		});
 		fused = rerankOutcome.candidates;
 		this.throwIfAborted(context);
-		const rerankStageMetadata = rerankOutcome.fallback ? {
-			...rerankOutcome.stats,
-			rerankFallbackReason: rerankOutcome.fallback.reason
-		} : rerankOutcome.stats;
-		trace?.endStage(fused.map((r) => r.entry.id), fused.map((r) => r.score), rerankStageMetadata);
+		trace?.endStage(fused.map((r) => r.entry.id), fused.map((r) => r.score));
 		fused = this.applyScoringPipeline(fused, trace, context.limit, rerankOutcome.modelOrder);
 		trace?.startStage("min_score_filter", fused.map((result) => result.entry.id));
 		fused = fused.filter((result) => result.score >= (context.minScore ?? this.config.minScore));
@@ -620,424 +589,23 @@ Object.assign(MemoryRetriever.prototype, {
 	}
 });
 //#endregion
-//#region src/engine/retrieval/retrieval-rerank-provider.ts
-function buildRerankRequest(apiKey, query, candidates) {
-	return {
-		headers: {
-			"Content-Type": "application/json",
-			Authorization: `Bearer ${apiKey}`,
-			"X-Sno-Source": "sno-station",
-			"X-Sno-Workload": "station-rerank"
-		},
-		body: {
-			query,
-			texts: candidates,
-			raw_scores: true
-		}
-	};
-}
-/** Sno's reranker answers a list of `{index, score}`, bare or under `results` / `data`. */
-function parseRerankResponse(data) {
-	const parseItems = (items) => {
-		if (!Array.isArray(items)) return void 0;
-		const parsed = [];
-		for (const raw of items) {
-			const index = typeof raw?.index === "number" ? raw.index : Number(raw?.index);
-			if (!Number.isFinite(index)) continue;
-			let score;
-			for (const key of ["score", "relevance_score"]) {
-				const value = raw?.[key];
-				const n = typeof value === "number" ? value : Number(value);
-				if (Number.isFinite(n)) {
-					score = n;
-					break;
-				}
-			}
-			if (score === void 0) continue;
-			parsed.push({
-				index,
-				score
-			});
-		}
-		return parsed.length > 0 ? parsed : void 0;
-	};
-	const objectData = data && typeof data === "object" && !Array.isArray(data) ? data : void 0;
-	return parseItems(data) ?? parseItems(objectData?.results) ?? parseItems(objectData?.data);
-}
-//#endregion
 //#region src/engine/retrieval/retriever-rerank.ts
 /** @file retriever-rerank.ts
-* @purpose Runs remote and local reranking.
+* @purpose Runs local cosine ordering and local model reranking.
 * @boundary Prototype-mounted MemoryRetriever methods; no constructor state ownership.
 */
-const RERANK_TRANSIENT_ATTEMPTS = 3;
-const RERANK_TRANSIENT_BACKOFF_MS = 200;
-const RERANK_TRANSIENT_STATUSES = /* @__PURE__ */ new Set([
-	429,
-	502,
-	503,
-	504
-]);
-const RERANK_RETRY_AFTER_MAX_MS = 5e3;
-function isRerankTimeout(error) {
-	return error instanceof Error && (error.name === "TimeoutError" || error.name === "AbortError");
-}
-/**
-* The reranker's refusal when a pair is over its window, as the endpoint states it:
-* `{"detail":{"code":"input_token_limit_exceeded","max_input_tokens":512,
-* "candidates":[{"index":0,"input_tokens":728,"excess_tokens":216}]}}` (measured 2026-09-14).
-*
-* It names every offending text, so the batch can be repaired and sent again instead of
-* losing the scores of the 49 texts that were fine. The counts are the SERVER's, taken with
-* the ranking model's tokenizer — which is not the embedder's — so they are the only exact
-* measure available here of how far over the window a text is.
-*/
-const RERANK_OVER_WINDOW_CODE = "input_token_limit_exceeded";
-function parseOverWindowRefusal(body) {
-	let parsed;
-	try {
-		parsed = JSON.parse(body);
-	} catch {
-		return;
-	}
-	if (typeof parsed !== "object" || parsed === null) return void 0;
-	const detail = parsed.detail;
-	if (typeof detail !== "object" || detail === null) return void 0;
-	const shape = detail;
-	if (shape.code !== RERANK_OVER_WINDOW_CODE) return void 0;
-	if (typeof shape.max_input_tokens !== "number" || !Array.isArray(shape.candidates)) return;
-	const overLong = [];
-	for (const candidate of shape.candidates) {
-		if (typeof candidate !== "object" || candidate === null) continue;
-		const named = candidate;
-		if (typeof named.index !== "number" || typeof named.input_tokens !== "number") continue;
-		overLong.push({
-			index: named.index,
-			inputTokens: named.input_tokens
-		});
-	}
-	return overLong.length === 0 ? void 0 : {
-		maxInputTokens: shape.max_input_tokens,
-		overLong
-	};
-}
-function isTransientRerankError(error) {
-	if (isRerankTimeout(error)) return true;
-	return error instanceof TypeError && error.cause !== void 0;
-}
-async function fetchRerankWithRetry(send, stop) {
-	let lastError;
-	for (let attempt = 1; attempt <= RERANK_TRANSIENT_ATTEMPTS; attempt += 1) {
-		let retryDelayMs = RERANK_TRANSIENT_BACKOFF_MS * attempt;
-		try {
-			const reply = await send();
-			const { response } = reply;
-			if (response.ok) return reply;
-			if (!RERANK_TRANSIENT_STATUSES.has(response.status) || attempt === RERANK_TRANSIENT_ATTEMPTS) return { response };
-			if (response.status === 429) {
-				const retryAfter = response.headers.get("retry-after");
-				if (retryAfter !== null && /^\d+$/.test(retryAfter)) retryDelayMs = Math.min(Number.parseInt(retryAfter, 10) * 1e3, RERANK_RETRY_AFTER_MAX_MS);
-			}
-			await response.body?.cancel();
-			lastError = /* @__PURE__ */ new Error(`rerank API responded ${response.status}`);
-		} catch (error) {
-			if (!isTransientRerankError(error) || attempt === RERANK_TRANSIENT_ATTEMPTS || stop?.aborted) throw error;
-			lastError = error;
-		}
-		if (stop?.aborted) throw lastError;
-		await new Promise((resolve) => {
-			const onAbort = () => {
-				clearTimeout(timer);
-				resolve();
-			};
-			const timer = setTimeout(() => {
-				stop?.removeEventListener("abort", onAbort);
-				resolve();
-			}, retryDelayMs);
-			stop?.addEventListener("abort", onAbort, { once: true });
-		});
-		if (stop?.aborted) throw lastError;
-	}
-	throw lastError;
-}
 Object.assign(MemoryRetriever.prototype, {
 	async rerank(query, candidates, queryVector, options) {
-		if (candidates.length === 0) return { candidates };
-		if (this.config.rerank === "none") return { candidates };
-		const unranked = () => this.rerankLightweight(candidates, queryVector);
-		if (this.config.rerank === "cross-encoder" && !this.config.rerankApiKey) throw new RetrievalError("rerank \"cross-encoder\" requires retrieval.rerankApiKey; set the key, or set retrieval.rerank to \"lightweight\" to choose the local ranker");
-		if (this.config.rerank === "lightweight" || !this.config.rerankApiKey) return await this.rerankLocalModel(query, this.rerankLightweight(candidates, queryVector), options?.deadlineAt, options?.signal);
-		const rerankEndpoint = this.config.rerankEndpoint;
-		if (!rerankEndpoint) {
-			log.warn("rerank skipped: no reranker address", {}, {
-				event_name: "memory.retriever_rerank.diagnostic",
-				file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-				function: "rerank",
-				site_id: "retrieval.retriever-rerank.rerank.6fbc6b3a22"
-			});
-			return {
-				candidates: unranked(),
-				fallback: { reason: "no_endpoint" }
-			};
-		}
-		const rerankTimeoutMs = this.config.rerankTimeoutMs ?? 15e3;
-		const stepBudget = AbortSignal.timeout(Math.max(0, Math.floor(options?.deadlineAt === void 0 ? rerankTimeoutMs : options.deadlineAt - Date.now())));
-		const stepStop = options?.signal ? AbortSignal.any([options.signal, stepBudget]) : stepBudget;
-		const rerankApiKey = this.config.rerankApiKey;
-		const maxCandidates = this.config.rerankMaxCandidates;
-		const overCap = maxCandidates !== void 0 && candidates.length > maxCandidates;
-		const toRerank = overCap ? candidates.slice(0, maxCandidates) : candidates;
-		const beyondCap = overCap ? candidates.slice(maxCandidates) : [];
-		const batchSize = 50;
-		const queryTokens = this.embedder.countTokens(query);
-		const documentTokenBudget = 509 - queryTokens;
-		if (documentTokenBudget <= 0) {
-			log.warn("rerank skipped: the query alone fills the reranker token window", {
-				queryTokens,
-				window: 512,
-				templateTokens: 3
-			}, {
-				event_name: "memory.retriever_rerank.diagnostic",
-				file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-				function: "rerank",
-				site_id: "retrieval.retriever-rerank.rerank.query_over_window"
-			});
-			return {
-				candidates: unranked(),
-				fallback: { reason: "query_over_window" }
-			};
-		}
-		let truncatedCandidates = 0;
-		const rerankTexts = toRerank.map((c) => {
-			const bounded = this.embedder.truncateToTokens(c.entry.text, documentTokenBudget);
-			if (bounded !== c.entry.text) truncatedCandidates += 1;
-			return bounded;
-		});
-		if (truncatedCandidates > 0) log.warn("rerank candidates were cut to the document budget the query left", {
-			truncatedCandidates,
-			documentTokenBudget,
-			queryTokens
-		}, {
-			event_name: "memory.retriever_rerank.diagnostic",
-			file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-			function: "rerank",
-			site_id: "retrieval.retriever-rerank.rerank.candidate_over_ceiling"
-		});
-		try {
-			const runBatch = async (start) => {
-				let batchTexts = rerankTexts.slice(start, start + batchSize);
-				for (let pass = 0;; pass += 1) {
-					const { headers, body } = buildRerankRequest(rerankApiKey, query, batchTexts);
-					const sendAttempt = async () => {
-						const clientRequestId = randomUUID();
-						const attemptStarted = performance.now();
-						let outcome = "failed";
-						let statusCode;
-						try {
-							const attempt = await fetch(rerankEndpoint, {
-								method: "POST",
-								headers: {
-									...headers,
-									"X-Request-ID": clientRequestId
-								},
-								body: JSON.stringify(body),
-								signal: AbortSignal.any([AbortSignal.timeout(rerankTimeoutMs), stepStop])
-							});
-							statusCode = attempt.status;
-							if (!attempt.ok) {
-								outcome = "http_error";
-								return { response: attempt };
-							}
-							const data = await attempt.json();
-							outcome = "success";
-							return {
-								response: attempt,
-								data
-							};
-						} catch (error) {
-							outcome = isRerankTimeout(error) ? "timeout" : "failed";
-							throw error;
-						} finally {
-							log.info("Rerank request completed", {
-								outcome,
-								status_code: statusCode ?? "unavailable",
-								duration_ms: performance.now() - attemptStarted,
-								texts_sent: batchTexts.length,
-								client_request_id: clientRequestId
-							}, {
-								event_name: "rerank.request.completed",
-								file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-								function: "runBatch",
-								site_id: "rerank.request.completed"
-							});
-						}
-					};
-					const { response, data } = await fetchRerankWithRetry(sendAttempt, stepStop);
-					if (!response.ok) {
-						const retryAfter = response.headers.get("retry-after");
-						const errorBody = await response.text().catch(() => "<unreadable body>");
-						const refusal = pass === 0 ? parseOverWindowRefusal(errorBody) : void 0;
-						if (refusal) {
-							const allowed = refusal.maxInputTokens - 3;
-							batchTexts = batchTexts.map((text, index) => {
-								const named = refusal.overLong.find((c) => c.index === index);
-								if (!named) return text;
-								const localTokens = this.embedder.countTokens(text);
-								const localPairTokens = queryTokens + localTokens;
-								const target = Math.max(1, Math.floor(allowed * localPairTokens / named.inputTokens) - queryTokens);
-								return this.embedder.truncateToTokens(text, target);
-							});
-							log.warn("rerank batch refused over its token window; resending the named texts cut to fit", {
-								status: response.status,
-								overLongCount: refusal.overLong.length,
-								maxInputTokens: refusal.maxInputTokens,
-								sentToRerankCount: batchTexts.length,
-								batchStart: start
-							}, {
-								event_name: "memory.retriever_rerank.diagnostic",
-								file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-								function: "runBatch",
-								site_id: "retrieval.retriever-rerank.runBatch.over_window_repair"
-							});
-							continue;
-						}
-						log.warn("rerank API failed, using pre-rerank results", {
-							status: response.status,
-							endpoint_hash: createHash("sha256").update(rerankEndpoint).digest("hex"),
-							retryAfter,
-							sentToRerankCount: batchTexts.length,
-							batchStart: start,
-							body_length: errorBody.length
-						}, {
-							event_name: "memory.retriever_rerank.diagnostic",
-							file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-							function: "runBatch",
-							site_id: "retrieval.retriever-rerank.runBatch.76d7d1fac6"
-						});
-						return "http_error";
-					}
-					const parsed = parseRerankResponse(data);
-					if (!parsed) {
-						log.warn("rerank API returned unparseable response", {
-							endpoint_hash: createHash("sha256").update(rerankEndpoint).digest("hex"),
-							batchStart: start
-						}, {
-							event_name: "memory.retriever_rerank.diagnostic",
-							file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-							function: "runBatch",
-							site_id: "retrieval.retriever-rerank.runBatch.4429b8bd8a"
-						});
-						return "invalid_response";
-					}
-					const batchItems = [];
-					for (const item of parsed) {
-						if (item.index < 0 || item.index >= batchTexts.length) continue;
-						batchItems.push({
-							index: start + item.index,
-							score: item.score
-						});
-					}
-					return batchItems;
-				}
-			};
-			const batchStarts = [];
-			for (let start = 0; start < rerankTexts.length; start += batchSize) batchStarts.push(start);
-			const items = [];
-			for (let wave = 0; wave < batchStarts.length; wave += 4) {
-				const waveStarts = batchStarts.slice(wave, wave + 4);
-				const settled = await Promise.allSettled(waveStarts.map(runBatch));
-				for (const outcome of settled) if (outcome.status === "rejected" && outcome.reason instanceof RetrievalError) throw outcome.reason;
-				for (const outcome of settled) if (outcome.status === "rejected") throw outcome.reason;
-				for (const outcome of settled) if (outcome.status === "fulfilled" && typeof outcome.value === "string") return {
-					candidates: unranked(),
-					fallback: { reason: outcome.value }
-				};
-				for (const outcome of settled) if (outcome.status === "fulfilled" && typeof outcome.value !== "string") items.push(...outcome.value);
-			}
-			const blendCross = this.config.rerankBlendCross ?? .7;
-			const blendVector = this.config.rerankBlendVector ?? .3;
-			const returnedIndices = new Set(items.map((r) => r.index));
-			const reranked = [];
-			const seenIndices = /* @__PURE__ */ new Set();
-			for (const item of items) {
-				if (seenIndices.has(item.index)) continue;
-				seenIndices.add(item.index);
-				const candidate = toRerank[item.index];
-				if (!candidate) continue;
-				const sourceScore = this.getRerankSourceScore(candidate);
-				const crossScore = 1 / (1 + Math.exp(-item.score));
-				const blendedScore = clamp01(crossScore * blendCross + sourceScore * blendVector, 0);
-				reranked.push({
-					...candidate,
-					score: blendedScore,
-					sources: {
-						...candidate.sources,
-						reranked: { score: item.score }
-					}
-				});
-			}
-			const unreturned = [...toRerank.filter((_, idx) => !returnedIndices.has(idx)), ...beyondCap].map((r) => ({
-				...r,
-				score: clamp01(this.getRerankSourceScore(r) * LIGHTWEIGHT_RERANK_PENALTY, 0)
-			}));
-			const merged = [...reranked, ...unreturned].sort((a, b) => b.score - a.score);
-			log.debug("reranked", {
-				totalCandidateCount: candidates.length,
-				sentToRerankCount: toRerank.length,
-				returnedCount: reranked.length
-			}, {
-				event_name: "memory.retriever_rerank.diagnostic",
-				file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-				function: "rerank",
-				site_id: "retrieval.retriever-rerank.rerank.7847c6248a"
-			});
-			return {
-				candidates: merged.length > 0 ? merged : candidates,
-				stats: {
-					rerankSentCount: toRerank.length,
-					rerankReturnedCount: reranked.length,
-					rerankBeyondCapCount: beyondCap.length
-				}
-			};
-		} catch (error) {
-			if (error instanceof RetrievalError) throw error;
-			let reason;
-			if (isRerankTimeout(error)) {
-				reason = "timeout";
-				log.warn("rerank API timed out, using pre-rerank results", {
-					timeoutMs: rerankTimeoutMs,
-					attempts: RERANK_TRANSIENT_ATTEMPTS
-				}, {
-					event_name: "memory.retriever_rerank.diagnostic",
-					file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-					function: "rerank",
-					site_id: "retrieval.retriever-rerank.rerank.4af9c17c75"
-				});
-			} else {
-				reason = "request_error";
-				log.warn("rerank error, using pre-rerank results", {
-					error,
-					attempts: RERANK_TRANSIENT_ATTEMPTS
-				}, {
-					event_name: "memory.retriever_rerank.diagnostic",
-					file: "packages/memory/src/engine/retrieval/retriever-rerank.ts",
-					function: "rerank",
-					site_id: "retrieval.retriever-rerank.rerank.2c5fc67dbf"
-				});
-			}
-			return {
-				candidates: unranked(),
-				fallback: { reason }
-			};
-		}
+		if (candidates.length === 0 || this.config.rerank === "none") return { candidates };
+		return this.rerankLocalModel(query, this.rerankLightweight(candidates, queryVector), options?.deadlineAt, options?.signal);
 	},
 	rerankLightweight(candidates, queryVector) {
 		if (queryVector.length === 0) return candidates;
 		if (candidates.length === 0) return candidates;
 		const chunkIds = candidates.map((c) => c.chunkId).filter((id) => id !== void 0);
 		const vectorMap = chunkIds.length > 0 ? this.store.getVectorsByIds(chunkIds) : /* @__PURE__ */ new Map();
-		const wF = this.config.lightweightFusionWeight ?? .1;
-		const wC = this.config.lightweightCosineWeight ?? .9;
+		const wF = this.config.lightweightFusionWeight ?? 0;
+		const wC = this.config.lightweightCosineWeight ?? 1;
 		let blendedCount = 0;
 		let dimMismatchCount = 0;
 		const reranked = candidates.map((candidate) => {
@@ -1104,9 +672,10 @@ Object.assign(MemoryRetriever.prototype, {
 	*/
 	async rerankLocalModel(query, ordered, deadlineAt, signal) {
 		if (ordered.length < 2 || signal?.aborted || deadlineAt !== void 0 && Date.now() >= deadlineAt) return { candidates: ordered };
-		const head = ordered.slice(0, 15);
+		const head = ordered.slice(0, 32);
 		const started = performance.now();
-		const budgetMs = deadlineAt === void 0 ? this.config.rerankTimeoutMs ?? 15e3 : Math.max(0, Math.floor(deadlineAt - Date.now()));
+		const budgetMs = Math.max(0, Math.floor(Math.min(DEFAULT_RERANK_TIMEOUT_MS, this.config.rerankTimeoutMs ?? 2e3, deadlineAt === void 0 ? DEFAULT_RERANK_TIMEOUT_MS : deadlineAt - Date.now())));
+		if (budgetMs === 0) return { candidates: ordered };
 		const stop = signal ? AbortSignal.any([signal, AbortSignal.timeout(budgetMs)]) : AbortSignal.timeout(budgetMs);
 		const scoring = this.embedder.scoreRelevance(query, head.map((r) => r.entry.text), stop).catch((error) => {
 			log.warn("local reranker failed; recall keeps the local ranking order", { error }, {
@@ -1116,11 +685,23 @@ Object.assign(MemoryRetriever.prototype, {
 				site_id: "retrieval.retriever-rerank.rerankLocalModel.failed"
 			});
 		});
-		const logits = await Promise.race([scoring, new Promise((resolve) => stop.addEventListener("abort", () => resolve(void 0), { once: true }))]);
+		let onAbort = () => {};
+		const aborted = new Promise((resolve) => {
+			onAbort = () => resolve(void 0);
+			stop.addEventListener("abort", onAbort, { once: true });
+		});
+		let logits;
+		try {
+			logits = await Promise.race([scoring, aborted]);
+		} finally {
+			stop.removeEventListener("abort", onAbort);
+		}
 		if (!logits || signal?.aborted) return { candidates: ordered };
+		const minimum = Math.min(...logits);
+		const span = Math.max(1e-9, Math.max(...logits) - minimum);
 		const blended = head.map((candidate, index) => ({
 			candidate,
-			value: logits[index] ?? Number.NEGATIVE_INFINITY
+			value: .5 * (((logits[index] ?? minimum) - minimum) / span) + .5 * candidate.score
 		}));
 		blended.sort((a, b) => b.value - a.value || a.candidate.entry.id.localeCompare(b.candidate.entry.id));
 		const slots = head.map((candidate) => candidate.score).sort((a, b) => b - a);
@@ -1149,9 +730,6 @@ Object.assign(MemoryRetriever.prototype, {
 			candidates: [...rescored, ...ordered.slice(head.length)],
 			modelOrder: blended.map(({ candidate }) => candidate.entry.id)
 		};
-	},
-	getRerankSourceScore(result) {
-		return clamp01(Math.max(result.sources.bm25?.score ?? 0, result.sources.vector?.score ?? 0), 0);
 	}
 });
 //#endregion
@@ -1446,7 +1024,7 @@ Object.assign(MemoryRetriever.prototype, {
 	},
 	applyMmrDiversity(results) {
 		if (results.length === 0) return results;
-		const lambda = this.config.mmrLambda ?? .7;
+		const lambda = this.config.mmrLambda ?? .9;
 		const chunkIds = results.map((r) => r.chunkId).filter((id) => id !== void 0);
 		const vectorMap = chunkIds.length > 0 ? this.store.getVectorsByIds(chunkIds) : /* @__PURE__ */ new Map();
 		const sorted = [...results].sort((a, b) => {
@@ -1790,4 +1368,4 @@ Object.assign(MemoryRetriever.prototype, {
 	}
 });
 //#endregion
-export { DEFAULT_RETRIEVAL_CONFIG as a, createRetriever as i, MemoryRetriever as n, withServingValidityDefault as o, RERANK_FALLBACK_REASONS as r, collectParallelStage as t };
+export { withServingValidityDefault as a, DEFAULT_RETRIEVAL_CONFIG as i, MemoryRetriever as n, createRetriever as r, collectParallelStage as t };

package/dist/{runtime-audit-log-C-L0v2HU.d.ts → runtime-audit-log-D2Hqk_kl.d.ts}

@@ -1,4 +1,4 @@
-import "./paths-m8yncZgm.js";
+import "./paths-obFSsJv3.js";
 //#region src/engine/operations/runtime-audit-log.d.ts
 type AuditEvent = "tool_call" | "hook_trigger" | "kill_switch" | "startup" | "error" | "auto_recall" | "ambient_learning" | "session_summary" | "envelope_strip" | "storage_integrity" | "memory_injected" | "memory_searched" | "memory_read" | "memory_updated" | "memory_superseded" | "memory_deleted" | "memory_purged" | "rem_trigger_evaluated" | "rem_triggered" | "rem_completed" | "rem_failed";
 type AuditStatus = "ok" | "error" | "skipped" | "partial";

package/dist/{server-BPOT66a_.js → server-CByyXSvp.js}

@@ -1,5 +1,5 @@
-import { At as getAuditPath, Dt as appendAuditEntryStrict, K as REM_BUILT_OPERATION_TYPES, Ot as compactAuditLog, Pt as readJsonlLines, bt as ModelCallRefusedError, d as RemJobStore, f as parseRemJobStats, n as remNeedsHost, q as parseRemOperationType } from "./rem-trigger-BKr4Qmhn.js";
-import { C as getSnoStationMemStateDir, Fr as MEMORY_SKIN_HEADER, Lr as memoryMethod, Mr as MEMORY_ERROR_STATUS, Pr as MEMORY_ROUTES, S as getSidecarSocketPath, _ as readRemTestHoldMs, c as REM_SIDECAR_ORIGIN, d as getRemDiscoveryPath, f as getRemJobJournalPath, g as readRemOperationalConfig, i as REM_JOBS_PATH_PREFIX, jr as MEMORY_DEFAULT_SKIN_ID, l as REM_SOURCE, nr as MEMORY_SHUTDOWN_TIMEOUT_MS, r as REM_CORRELATION_ID_HEADER, s as REM_SIDECAR_HOST, u as getRemChassisJournalPath, v as SettingsUnavailableError, y as getDiscoveryPath } from "./config-BhlMin82.js";
+import { At as getAuditPath, Dt as appendAuditEntryStrict, K as REM_BUILT_OPERATION_TYPES, Ot as compactAuditLog, Pt as readJsonlLines, bt as ModelCallRefusedError, d as RemJobStore, f as parseRemJobStats, n as remNeedsHost, q as parseRemOperationType } from "./rem-trigger-pg_cCp5U.js";
+import { $n as MEMORY_SHUTDOWN_TIMEOUT_MS, Ar as MEMORY_SKIN_HEADER, C as getSnoStationMemStateDir, Dr as MEMORY_ERROR_STATUS, Er as MEMORY_DEFAULT_SKIN_ID, Mr as memoryMethod, S as getSidecarSocketPath, _ as readRemTestHoldMs, c as REM_SIDECAR_ORIGIN, d as getRemDiscoveryPath, f as getRemJobJournalPath, g as readRemOperationalConfig, i as REM_JOBS_PATH_PREFIX, kr as MEMORY_ROUTES, l as REM_SOURCE, r as REM_CORRELATION_ID_HEADER, s as REM_SIDECAR_HOST, u as getRemChassisJournalPath, v as SettingsUnavailableError, y as getDiscoveryPath } from "./config-BwUcq2zS.js";
 import { t as ContractError } from "./error-D-wb-rZw.js";
 import { setImmediate } from "node:timers/promises";
 import { createLogger, currentLogContext, effectiveLogLevel, emitDiagnostic } from "@snoai/utils/logger";
@@ -335,7 +335,7 @@ async function startOwnedRemSidecar() {
 		async open() {
 			if (currentMemory) return currentMemory;
 			if (settingsError) throw settingsError;
-			openingMemory ??= import("./memory-runtime-De1T9jTv.js").then((module) => module.MemoryRuntimePool.open()).then((pool) => {
+			openingMemory ??= import("./memory-runtime-B4A1t5DO.js").then((module) => module.MemoryRuntimePool.open()).then((pool) => {
 				currentMemory = pool;
 				if (stopping) pool.stopTimers();
 				return pool;
@@ -730,7 +730,7 @@ async function runChassisJob(store, journal, waveId, holdMs, runtime) {
 				cleanRefusalReasons.push(reason);
 				await appendChassisRefusal(journal, queued, operation, "refused", reason);
 			}
-			const { runRemProductionOrderedWave } = await import("./rem-batch-executor-D42YWCNg.js");
+			const { runRemProductionOrderedWave } = await import("./rem-batch-executor-BUgNyYLv.js");
 			const result = enabledOperations.length === 0 ? {
 				decision: "allow",
 				reasonCode: null,

package/dist/sidecar/main.js

@@ -1,4 +1,4 @@
-import { Nr as MEMORY_RECONNECT_INTERVAL_MS, m as isRemTraceEnabled, p as getRemTraceLogPath, v as SettingsUnavailableError } from "../config-BhlMin82.js";
+import { Or as MEMORY_RECONNECT_INTERVAL_MS, m as isRemTraceEnabled, p as getRemTraceLogPath, v as SettingsUnavailableError } from "../config-BwUcq2zS.js";
 import { setTimeout } from "node:timers/promises";
 import { addLogFileTarget, closeLogger, createLogger } from "@snoai/utils/logger";
 //#region src/sidecar/main.ts
@@ -6,7 +6,7 @@ import { addLogFileTarget, closeLogger, createLogger } from "@snoai/utils/logger
 * @purpose Boots the standalone local REM sidecar process.
 * @boundary Process environment, sidecar lifecycle, and local request-log destination.
 */
-const { startRemSidecar, DuplicateSidecarError } = await import("../server-BPOT66a_.js");
+const { startRemSidecar, DuplicateSidecarError } = await import("../server-CByyXSvp.js");
 let sidecar;
 for (let attempt = 1; !sidecar; attempt++) try {
 	sidecar = await startRemSidecar();

package/dist/sno-station-mem-runtime-dependencies-ByI3b-wB.d.ts

@@ -0,0 +1,31 @@
+import "./types-El7jViCz.js";
+import "./index-CJrMFlAU.js";
+import "./adapter-BFtbOOw3.js";
+import "./embedding-provider-client-CELWtPgt.js";
+import "./retriever-2RJYAvlD.js";
+import "./connection-D989FqGS.js";
+import "./sqlite-runtime-L_ywirTm.js";
+import "./store-Bz0y76cJ.js";
+import "./access-tracker-CFxRg7Bt.js";
+import "./memory-tier-promoter-m60-RlAj.js";
+import "./capture-policy-detector-3tDQkBi9.js";
+import "./memory-temporality-classifier-Wi6WWeAj.js";
+import "./runtime-audit-log-D2Hqk_kl.js";
+import "./atomic-memory-extraction-Bp7AWZqi.js";
+import "./llm-client-nWW1o1F6.js";
+import "./memory-snapshot-CKnQovB5.js";
+import "./observable-embedder-0FwxJMhJ.js";
+import "./observable-llm-client-DY7w1Fj1.js";
+import "./observable-memory-store-C25Q_hoW.js";
+import "./observable-retriever-Cs9Bwzq6.js";
+import "./reflection-embedded-generator-Cs59hWYr.js";
+import "./strategy-hook-runner-vbrWiDbK.js";
+import "./data-bootstrap-Bl3lthzE.js";
+import "@snoai/sqlite-crypto";
+//#region src/engine/bindings/sno-station-mem-runtime-dependencies.d.ts
+/** Legacy parameters carry only an unused logger slot; no host operations are exposed. */
+type SnoStationMemPluginApi = {
+  logger: unknown;
+};
+//#endregion
+export { SnoStationMemPluginApi as t };