hippo-memory 1.56.0 → 1.58.0

package/dist/db.js

@@ -6,6 +6,7 @@ import { createPhysicsTable } from './physics-state.js';
 import { cleanupArchivedMirrors } from './raw-archive-mirror-cleanup.js';
 import { PACKAGE_VERSION, compareSemver } from './version.js';
 import { deriveOriginProject, originFromSource, isGlobalStoreRoot } from './project-identity.js';
+import { log } from './log.js';
 const require = createRequire(import.meta.url);
 // SAFETY: node:sqlite's DatabaseSync constructor genuinely has this shape at
 // runtime (Node's built-in synchronous SQLite module); there are no bundled
@@ -2607,7 +2608,7 @@ export function isSqliteBusy(error) {
 // busy_timeout covers neither of this file's two contended statements: SQLite
 // skips the busy handler for `PRAGMA journal_mode` and for a write that upgrades
 // a deferred read snapshot. Both need an explicit wait instead.
-function execWithBusyRetry(db, sql, timeoutMs = 30000) {
+export function execWithBusyRetry(db, sql, timeoutMs = 30000) {
     const deadline = Date.now() + timeoutMs;
     const idle = new Int32Array(new SharedArrayBuffer(4));
     for (;;) {
@@ -2622,8 +2623,66 @@ function execWithBusyRetry(db, sql, timeoutMs = 30000) {
         }
     }
 }
+// Hook commands run on every prompt, so inside withSharedStoreHandles each store pays its pragmas, migration check and mirror cleanup once.
+const sharedHandles = new Map();
+const sharedSet = new WeakSet();
+let shareDepth = 0;
+let shareBusyWaitMs;
+/** Lock wait for hook commands: under the 5 s prompt-hook budget even after a few skipped writes, and far above a normal write's hold. */
+export const HOOK_DB_WAIT_MS = 1000;
+/** A busy store made a command skip work: warn once per process (the holder is usually `hippo sleep`). */
+export function noteStoreBusy(skipped) {
+    log.once('store-busy', 'warn', `store busy (another hippo process holds the write lock); ${skipped}`);
+    // A lock held past one full wait belongs to a long transaction, so the hook's later writes skip at once.
+    for (const db of sharedHandles.values()) {
+        if (db.isOpen !== false)
+            db.exec('PRAGMA busy_timeout = 0');
+    }
+}
+function closeSharedStoreHandles() {
+    for (const db of sharedHandles.values()) {
+        sharedSet.delete(db);
+        if (db.isOpen !== false)
+            db.close();
+    }
+    sharedHandles.clear();
+}
+/** Runs `fn` with one handle per store: openHippoDb reuses it and closeHippoDb leaves it open until `fn` settles or the process exits.
+ *  `busyWaitMs` is the lock wait of every open inside `fn` that does not pass its own. */
+export async function withSharedStoreHandles(fn, opts) {
+    if (shareDepth++ === 0) {
+        process.once('exit', closeSharedStoreHandles);
+        shareBusyWaitMs = opts?.busyWaitMs;
+    }
+    try {
+        return await fn();
+    }
+    finally {
+        if (--shareDepth === 0) {
+            process.off('exit', closeSharedStoreHandles);
+            closeSharedStoreHandles();
+            shareBusyWaitMs = undefined;
+        }
+    }
+}
 /** `busyWaitMs` shortens every lock wait of this open, for a hook that must finish inside its own timeout. */
 export function openHippoDb(hippoRoot, opts) {
+    if (shareDepth === 0)
+        return openOwnHippoDb(hippoRoot, opts);
+    const busyWaitMs = opts?.busyWaitMs ?? shareBusyWaitMs;
+    const key = `${path.resolve(getHippoDbPath(hippoRoot))}\0${busyWaitMs ?? ''}`;
+    const shared = sharedHandles.get(key);
+    if (shared?.isOpen && !shared.isTransaction)
+        return shared;
+    const db = openOwnHippoDb(hippoRoot, { busyWaitMs });
+    // An open nested inside a transaction gets its own connection, as it did before sharing.
+    if (!shared?.isOpen) {
+        sharedHandles.set(key, db);
+        sharedSet.add(db);
+    }
+    return db;
+}
+function openOwnHippoDb(hippoRoot, opts) {
     fs.mkdirSync(hippoRoot, { recursive: true });
     const db = new DatabaseSync(getHippoDbPath(hippoRoot));
     const busyWaitMs = opts?.busyWaitMs;
@@ -2937,6 +2996,8 @@ function backfillFtsIndex(db) {
   `);
 }
 export function closeHippoDb(db) {
+    if (sharedSet.has(db))
+        return;
     db.close();
 }
 export function getMeta(db, key, fallback = '') {

package/dist/decisions.js

@@ -23,8 +23,10 @@
  * 'write_entry' (store.ts:1196) via the afterWrite hook, so a failure in any
  * step rolls all of them back. Pattern matches savePrediction (predictions.ts).
  */
+import { BadRequestError, ConflictError, NotFoundError } from './api-errors.js';
 import { openHippoDb, closeHippoDb } from './db.js';
-import { writeEntry, assertTenantId } from './store.js';
+import { writeEntry } from './store.js';
+import { assertTenantId } from './tenant.js';
 import { markGraphDirty, removeGraphEntitiesForObject } from './graph.js';
 import { createMemory, Layer } from './memory.js';
 import { appendAuditEvent } from './audit.js';
@@ -72,7 +74,7 @@ const DECISION_COLS = `
 export function saveDecision(hippoRoot, tenantId, opts, actor = 'cli') {
     assertTenantId('saveDecision', tenantId);
     if (!opts.decisionText)
-        throw new Error('saveDecision: decisionText is required');
+        throw new BadRequestError('saveDecision: decisionText is required');
     const now = new Date().toISOString();
     const content = opts.context
         ? `${opts.decisionText}\n\nContext: ${opts.context}`
@@ -102,10 +104,10 @@ export function saveDecision(hippoRoot, tenantId, opts, actor = 'cli') {
                 // SAFETY: row shape matches the single `status` column named in the SELECT below.
                 const pred = db.prepare(`SELECT status FROM decisions WHERE id = ? AND tenant_id = ?`).get(opts.supersedesDecisionId, tenantId);
                 if (!pred) {
-                    throw new Error(`saveDecision: decision ${opts.supersedesDecisionId} to supersede not found for tenant ${tenantId}`);
+                    throw new NotFoundError(`saveDecision: decision ${opts.supersedesDecisionId} to supersede not found for tenant ${tenantId}`);
                 }
                 if (pred.status !== 'active') {
-                    throw new Error(`saveDecision: decision ${opts.supersedesDecisionId} is not active (status='${pred.status}'); only active decisions can be superseded.`);
+                    throw new ConflictError(`saveDecision: decision ${opts.supersedesDecisionId} is not active (status='${pred.status}'); only active decisions can be superseded.`);
                 }
             }
             const result = db.prepare(`
@@ -126,7 +128,7 @@ export function saveDecision(hippoRoot, tenantId, opts, actor = 'cli') {
           WHERE id = ? AND tenant_id = ? AND status = 'active' AND id != ?
         `).run(decisionId, now, opts.supersedesDecisionId, tenantId, decisionId);
                 if (sup.changes === 0) {
-                    throw new Error(`saveDecision: decision ${opts.supersedesDecisionId} could not be superseded (no longer active or self-reference).`);
+                    throw new BadRequestError(`saveDecision: decision ${opts.supersedesDecisionId} could not be superseded (no longer active or self-reference).`);
                 }
                 appendAuditEvent(db, {
                     tenantId,
@@ -190,15 +192,15 @@ export function closeDecision(hippoRoot, tenantId, id, actor = 'cli') {
                 // SAFETY: row shape matches the single `status` column named in the SELECT above.
                 const existing = db.prepare(`SELECT status FROM decisions WHERE id = ? AND tenant_id = ?`).get(id, tenantId);
                 if (!existing) {
-                    throw new Error(`closeDecision: decision ${id} not found for tenant ${tenantId}`);
+                    throw new NotFoundError(`closeDecision: decision ${id} not found for tenant ${tenantId}`);
                 }
-                throw new Error(`closeDecision: decision ${id} is not active (status='${existing.status}'); only active decisions can be closed.`);
+                throw new ConflictError(`closeDecision: decision ${id} is not active (status='${existing.status}'); only active decisions can be closed.`);
             }
             // SAFETY: row's shape matches the columns named in DECISION_COLS above.
             const row = db.prepare(`SELECT ${DECISION_COLS} FROM decisions WHERE id = ? AND tenant_id = ?`)
                 .get(id, tenantId);
             if (!row)
-                throw new Error(`closeDecision: decision ${id} not found after UPDATE`);
+                throw new NotFoundError(`closeDecision: decision ${id} not found after UPDATE`);
             appendAuditEvent(db, {
                 tenantId,
                 actor,
@@ -255,7 +257,7 @@ export function loadDecisions(hippoRoot, tenantId, opts = {}) {
         let rows;
         if (opts.status) {
             if (!VALID_DECISION_STATES.has(opts.status)) {
-                throw new Error(`loadDecisions: status must be one of ${Array.from(VALID_DECISION_STATES).join('|')}; got ${opts.status}`);
+                throw new BadRequestError(`loadDecisions: status must be one of ${Array.from(VALID_DECISION_STATES).join('|')}; got ${opts.status}`);
             }
             // SAFETY: rows' shape matches the columns named in DECISION_COLS above.
             rows = db.prepare(`

package/dist/doctor.js

@@ -10,6 +10,7 @@ import * as path from 'node:path';
 import { findHippoStoreDir } from './project-identity.js';
 import { getGlobalRoot } from './shared.js';
 import { isInitialized } from './store.js';
+import { loadConfig } from './config.js';
 import { openHippoDbReadOnly, closeHippoDb, getSchemaVersion, getCurrentSchemaVersion, countTableRows, IncompatibleBinaryError } from './db.js';
 import { REPLAY_AFTER_MS, TRANSCRIPT_FILL_WINDOW_MS } from './compaction-record.js';
 import { isEmbeddingAvailable } from './embeddings.js';
@@ -208,6 +209,10 @@ export function runDoctor(opts) {
                 closeHippoDb(db);
         }
     }
+    const holdoutRateBp = store === null ? 0 : loadConfig(store).pilot.holdoutRateBp;
+    if (holdoutRateBp > 0) {
+        checks.push({ id: 'pilot', status: 'info', detail: `pilot holdout on: about ${holdoutRateBp / 100}% of sessions get no memories pushed by hippo (pilot.holdoutRateBp=${holdoutRateBp})` });
+    }
     const claudeDir = path.join(home, '.claude');
     if (fs.existsSync(claudeDir)) {
         const settings = readJson(path.join(claudeDir, 'settings.json'));

package/dist/embedding-provider.js

@@ -33,6 +33,7 @@
 import { getEmbedding, isEmbeddingAvailable, resolveEmbeddingModel, DEFAULT_EMBEDDING_MODEL, } from './embeddings.js';
 import { loadConfig } from './config.js';
 import { redactSecrets } from './secret-detect.js';
+import { fetchWithRetry } from './http-retry.js';
 export const API_PROVIDER_KINDS = ['openai', 'voyage', 'cohere'];
 function isApiProviderKind(x) {
     return x === 'openai' || x === 'voyage' || x === 'cohere';
@@ -184,15 +185,14 @@ class ApiEmbeddingProvider {
         const url = `${this.baseUrl.replace(/\/$/, '')}/${spec.path}`;
         let resp;
         try {
-            resp = await fetch(url, {
+            resp = await fetchWithRetry(url, {
                 method: 'POST',
                 headers: {
                     'content-type': 'application/json',
                     authorization: `Bearer ${key}`,
                 },
                 body: JSON.stringify(spec.buildBody(this.model, chunk.map(redactSecrets), role)),
-                signal: AbortSignal.timeout(REQUEST_TIMEOUT_MS),
-            });
+            }, { timeoutMs: REQUEST_TIMEOUT_MS });
         }
         catch (err) {
             const msg = err instanceof Error ? err.message : String(err);

package/dist/embeddings.d.ts

@@ -4,6 +4,7 @@
  * Falls back silently if the library is not installed.
  */
 import { MemoryEntry } from './memory.js';
+import { type EmbeddingProvider } from './embedding-provider.js';
 export declare const DEFAULT_EMBEDDING_MODEL = "Xenova/all-MiniLM-L6-v2";
 export declare const EMBEDDING_MODEL_META_KEY = "embedding_model";
 /**
@@ -114,8 +115,7 @@ export declare function getEmbedding(text: string, model?: string, role?: Embedd
  */
 export declare function cosineSimilarity(a: number[], b: number[]): number;
 /**
- * Load the cached embedding index from disk.
- * Returns an empty object if the file doesn't exist or is corrupt.
+ * Load the cached embedding index; `{}` when the file is missing. A corrupt file is moved aside and rebuilt on the next embed; any other read error throws, so nothing saves over an index it could not read.
  */
 export declare function loadEmbeddingIndex(hippoRoot: string): Record<string, number[]>;
 /**
@@ -129,7 +129,7 @@ export declare function embedMemory(hippoRoot: string, entry: MemoryEntry, model
 /**
  * Embed all entries in hippoRoot that don't already have cached vectors.
  * Prunes orphaned embeddings for memories that no longer exist.
- * Returns the count of newly embedded entries.
+ * Returns the count of newly embedded entries. `provider` defaults to the store's configured one.
  */
-export declare function embedAll(hippoRoot: string, model?: string): Promise<number>;
+export declare function embedAll(hippoRoot: string, model?: string, provider?: EmbeddingProvider): Promise<number>;
 //# sourceMappingURL=embeddings.d.ts.map

package/dist/embeddings.js

@@ -13,6 +13,7 @@ import { initializeParticle, savePhysicsState, loadPhysicsState, resetAllPhysics
 import { loadConfig } from './config.js';
 import { resolveEmbeddingProvider } from './embedding-provider.js';
 import { redactSecretsStrict } from './secret-detect.js';
+import { log } from './log.js';
 // Use createRequire for synchronous module resolution check in ESM
 const _require = createRequire(import.meta.url);
 // Cached availability check
@@ -270,6 +271,9 @@ async function rebuildEmbeddingIndex(entries, provider) {
         if (vec && vec.length > 0) {
             rebuilt[entries[i].id] = vec;
         }
+        else {
+            noteSkippedEmbedding(entries[i].id);
+        }
     }
     return rebuilt;
 }
@@ -283,10 +287,15 @@ function resetPhysicsFromIndex(hippoRoot, entries, index) {
             closeHippoDb(db);
         }
     }
-    catch {
-        // Physics reset is best-effort; retrieval will still fall back gracefully.
+    catch (err) {
+        // Best effort: retrieval still falls back without physics state.
+        log.warn(`physics reset after reindex failed: ${err instanceof Error ? err.message : String(err)}`);
     }
 }
+/** A provider's `[]` row is a swallowed per-item failure; name the memory so the gap can be traced. */
+function noteSkippedEmbedding(id) {
+    log.warn('memory not embedded; the next embed run retries it', { id });
+}
 /**
  * Get an embedding vector for a piece of text.
  * Returns an empty array if transformers is not available or fails.
@@ -314,7 +323,9 @@ export async function getEmbedding(text, model = DEFAULT_EMBEDDING_MODEL, role)
         // pipeline's documented tensor output shape.
         return Array.from(output.data);
     }
-    catch {
+    catch (err) {
+        // The caller sees `[]` and names the memory; the reason only shows at debug.
+        log.debug(`local embedding failed: ${err instanceof Error ? err.message : String(err)}`);
         return [];
     }
 }
@@ -340,23 +351,66 @@ export function cosineSimilarity(a, b) {
     return Math.min(1, Math.max(-1, dot / denom));
 }
 const EMBEDDINGS_FILE = 'embeddings.json';
+// Never equal to a real index identity, so the next embed run treats it as a model change and rebuilds every vector.
+const QUARANTINED_INDEX_IDENTITY = 'quarantined-corrupt-index';
+function isErrnoCode(err, code) {
+    return err instanceof Error && 'code' in err && err.code === code;
+}
+function parseEmbeddingIndex(raw) {
+    try {
+        const parsed = JSON.parse(raw);
+        // SAFETY: saveEmbeddingIndex is the only writer and always writes this shape; anything that is not an object is corrupt.
+        return parsed instanceof Object && !Array.isArray(parsed) ? parsed : null;
+    }
+    catch {
+        return null;
+    }
+}
+/** Move a corrupt index aside and flag a full rebuild, so no later save can write over the only copy of those bytes. */
+function quarantineCorruptIndex(hippoRoot, fp) {
+    const aside = `${fp}.corrupt-${new Date().toISOString().replace(/[:.]/g, '-')}-${process.pid}`;
+    try {
+        fs.renameSync(fp, aside);
+    }
+    catch (err) {
+        if (isErrnoCode(err, 'ENOENT'))
+            return;
+        // A rename blocked by an open handle (Windows) still gets a copy kept; if the copy fails too, the throw stops the save.
+        fs.copyFileSync(fp, aside, fs.constants.COPYFILE_EXCL);
+    }
+    log.error(`${EMBEDDINGS_FILE} could not be parsed; kept it as ${path.basename(aside)} and the next embed rebuilds the index`, { hippoRoot });
+    try {
+        const db = openHippoDb(hippoRoot);
+        try {
+            setMeta(db, EMBEDDING_MODEL_META_KEY, QUARANTINED_INDEX_IDENTITY);
+        }
+        finally {
+            closeHippoDb(db);
+        }
+    }
+    catch (err) {
+        log.warn(`could not flag the embedding index for rebuild; run 'hippo embed' to restore vectors (${err instanceof Error ? err.message : String(err)})`, { hippoRoot });
+    }
+}
 /**
- * Load the cached embedding index from disk.
- * Returns an empty object if the file doesn't exist or is corrupt.
+ * Load the cached embedding index; `{}` when the file is missing. A corrupt file is moved aside and rebuilt on the next embed; any other read error throws, so nothing saves over an index it could not read.
  */
 export function loadEmbeddingIndex(hippoRoot) {
     const fp = path.join(hippoRoot, EMBEDDINGS_FILE);
-    if (!fs.existsSync(fp))
-        return {};
+    let raw;
     try {
-        // SAFETY: embeddings.json is written exclusively by saveEmbeddingIndex with
-        // this exact shape; a corrupt or foreign file is caught by the try/catch
-        // below and treated as an empty index.
-        return JSON.parse(fs.readFileSync(fp, 'utf8'));
+        raw = fs.readFileSync(fp, 'utf8');
     }
-    catch {
-        return {};
+    catch (err) {
+        if (isErrnoCode(err, 'ENOENT'))
+            return {};
+        throw err;
     }
+    const index = parseEmbeddingIndex(raw);
+    if (index)
+        return index;
+    quarantineCorruptIndex(hippoRoot, fp);
+    return {};
 }
 /**
  * Save the embedding index to disk.
@@ -543,10 +597,9 @@ export async function embedMemory(hippoRoot, entry, model) {
 /**
  * Embed all entries in hippoRoot that don't already have cached vectors.
  * Prunes orphaned embeddings for memories that no longer exist.
- * Returns the count of newly embedded entries.
+ * Returns the count of newly embedded entries. `provider` defaults to the store's configured one.
  */
-export async function embedAll(hippoRoot, model) {
-    const provider = resolveEmbeddingProvider(hippoRoot, { model });
+export async function embedAll(hippoRoot, model, provider = resolveEmbeddingProvider(hippoRoot, { model })) {
     if (!provider.isAvailable()) {
         // A configured (non-disabled) API provider with a missing key is a
         // misconfiguration, not a no-op: surface it so programmatic callers of the
@@ -618,6 +671,9 @@ export async function embedAll(hippoRoot, model) {
                     dirty = true;
                     chunkDirty = true;
                 }
+                else {
+                    noteSkippedEmbedding(chunk[j].id);
+                }
             }
             if (chunkDirty)
                 saveEmbeddingIndex(hippoRoot, index);

package/dist/eval-stats.d.ts

@@ -120,4 +120,62 @@ export declare function passAtK(runsByTask: boolean[][], k: number): number;
  * NaN when no task has k runs.
  */
 export declare function passHatK(runsByTask: boolean[][], k: number): number;
+/** All units of one family (tasks by seeds), resampled as one block so seeds stay together. */
+export type Family<T> = readonly T[];
+/** One repository's families; a task with no lesson is a family of one. */
+export type Repository<T> = readonly Family<T>[];
+/** An {@link Estimate} with a two-sided p-value from the same resamples. */
+export interface TestedEstimate extends Estimate {
+    /** 2 x min(share <= null, share >= null), capped at 1; NaN with fewer than two repositories. */
+    readonly p: number;
+    /** Non-finite resamples, dropped; `iterations + dropped` is the number requested. */
+    readonly dropped: number;
+    /** The null the p-value tested against; {@link verdict} reads its sides from here. */
+    readonly nullValue: number;
+}
+/** Options for {@link twoLevelBootstrap}. */
+export interface TwoLevelOpts extends BootstrapOpts {
+    /** Resamples. Default 10,000. */
+    readonly iterations?: number;
+    /** Value the p-value tests against: 0 for differences, 1 for ratios. Default 0. */
+    readonly nullValue?: number;
+}
+/** Resamples repositories, then families inside each; only the repository draw carries a shared-store fault.
+ * The statistic never uses the PRNG, so a second call on one seed draws the same resamples. */
+export declare function twoLevelBootstrap<T>(repos: readonly Repository<T>[], statistic: (units: readonly T[]) => number, opts?: TwoLevelOpts): TestedEstimate;
+/** Holm step-down in input order; the family size is `ps.length`, as preregistered.
+ * A NaN stays NaN but ranks as 1, so it never loosens the others; a finite p outside [0, 1] throws. */
+export declare function holmAdjust(ps: readonly number[]): number[];
+/** One of the four mutually exclusive outcomes the preregistration allows per hypothesis. */
+export type Verdict = 'loss' | 'win' | 'tie' | 'inconclusive';
+/** What a hypothesis needs to be read; see {@link verdict}. */
+export interface VerdictSpec {
+    /** Direction that favours the treatment arm. */
+    readonly helpful: 'lower' | 'higher';
+    /** Inclusive band the interval must sit inside for a tie. */
+    readonly tieBand: readonly [number, number];
+    /** An estimate on this value or beyond it, on the helpful side, reaches the minimum effect. */
+    readonly minimumEffectAt: number;
+    /** Default 0.05. */
+    readonly alpha?: number;
+}
+/** A verdict, plus whether a win reaches the minimum effect (a win below it is a small win). */
+export interface VerdictResult {
+    readonly verdict: Verdict;
+    readonly reachesMinimum: boolean;
+}
+/** Checks in the preregistered order; the null comes from `e.nullValue`, so p and sides cannot disagree.
+ * A NaN estimate, p or bound is inconclusive, since a win or loss cannot be ruled out. */
+export declare function verdict(e: TestedEstimate, adjustedP: number, spec: VerdictSpec): VerdictResult;
+/** A win must hold under both codings of not-applicable, while a loss under either is reported. */
+export declare function combineCodings(a: VerdictResult, b: VerdictResult): VerdictResult;
+/** Outcome of the harm gate; see {@link harmGate}. */
+export interface HarmGate {
+    readonly pass: boolean;
+    readonly costOk: boolean;
+    readonly resolveOk: boolean;
+}
+/** Cost ratio's upper bound below 1.10, resolve difference's lower bound (a fraction) above -0.05.
+ * Both estimates must use alpha 0.05, the conservative reading of "upper 95% bound"; a NaN bound fails. */
+export declare function harmGate(costRatio: Estimate, resolveDiff: Estimate): HarmGate;
 //# sourceMappingURL=eval-stats.d.ts.map

package/dist/eval-stats.js

@@ -184,4 +184,115 @@ export function passHatK(runsByTask, k) {
         return Number.NaN;
     return eligible.filter((r) => r.slice(0, k).every(Boolean)).length / eligible.length;
 }
+function notANumber(nullValue, dropped) {
+    const nan = Number.NaN;
+    return { estimate: nan, low: nan, high: nan, p: nan, iterations: 0, dropped, nullValue };
+}
+function resampleUnits(repos, rand) {
+    const units = [];
+    for (let i = 0; i < repos.length; i++) {
+        const repo = repos[Math.floor(rand() * repos.length)];
+        for (let j = 0; j < repo.length; j++) {
+            for (const unit of repo[Math.floor(rand() * repo.length)])
+                units.push(unit);
+        }
+    }
+    return units;
+}
+// Inclusive on both sides, as the p-value is worded; float noise around the null breaks a tie.
+function twoSidedP(samples, nullValue) {
+    let below = 0;
+    let above = 0;
+    for (const s of samples) {
+        if (s <= nullValue)
+            below++;
+        if (s >= nullValue)
+            above++;
+    }
+    return Math.min(1, (2 * Math.min(below, above)) / samples.length);
+}
+/** Resamples repositories, then families inside each; only the repository draw carries a shared-store fault.
+ * The statistic never uses the PRNG, so a second call on one seed draws the same resamples. */
+export function twoLevelBootstrap(repos, statistic, opts = {}) {
+    const requested = opts.iterations ?? 10_000;
+    if (!Number.isInteger(requested) || requested <= 0) {
+        throw new RangeError(`iterations must be a positive integer, got ${requested}`);
+    }
+    const nullValue = opts.nullValue ?? 0;
+    const kept = repos.map((r) => r.filter((f) => f.length > 0)).filter((r) => r.length > 0);
+    if (kept.length === 0)
+        return notANumber(nullValue, 0);
+    const rand = seededRandom(opts.seed ?? 1);
+    const samples = [];
+    for (let b = 0; b < requested; b++) {
+        const value = statistic(resampleUnits(kept, rand));
+        if (Number.isFinite(value))
+            samples.push(value);
+    }
+    const dropped = requested - samples.length;
+    if (samples.length === 0)
+        return notANumber(nullValue, dropped);
+    const estimate = statistic(kept.flatMap((r) => r.flat()));
+    const p = kept.length < 2 ? Number.NaN : twoSidedP(samples, nullValue);
+    const interval = percentileInterval(samples, opts.alpha ?? 0.05);
+    return { estimate, ...interval, p, iterations: samples.length, dropped, nullValue };
+}
+/** Holm step-down in input order; the family size is `ps.length`, as preregistered.
+ * A NaN stays NaN but ranks as 1, so it never loosens the others; a finite p outside [0, 1] throws. */
+export function holmAdjust(ps) {
+    for (const p of ps) {
+        if (!Number.isNaN(p) && !(p >= 0 && p <= 1))
+            throw new RangeError(`p-value out of range: ${p}`);
+    }
+    const ranked = ps
+        .map((p, index) => ({ index, key: Number.isNaN(p) ? 1 : p }))
+        .sort((a, b) => a.key - b.key || a.index - b.index);
+    const adjusted = Array.from({ length: ps.length }, () => Number.NaN);
+    let running = 0;
+    ranked.forEach(({ index, key }, rank) => {
+        running = Math.max(running, (ps.length - rank) * key);
+        if (!Number.isNaN(ps[index]))
+            adjusted[index] = Math.min(1, running);
+    });
+    return adjusted;
+}
+/** Checks in the preregistered order; the null comes from `e.nullValue`, so p and sides cannot disagree.
+ * A NaN estimate, p or bound is inconclusive, since a win or loss cannot be ruled out. */
+export function verdict(e, adjustedP, spec) {
+    const inconclusive = { verdict: 'inconclusive', reachesMinimum: false };
+    if ([adjustedP, e.estimate, e.low, e.high].some(Number.isNaN))
+        return inconclusive;
+    const nullValue = e.nullValue;
+    const lowerIsHelpful = spec.helpful === 'lower';
+    if (adjustedP < (spec.alpha ?? 0.05)) {
+        const helpfulSide = lowerIsHelpful ? e.estimate < nullValue : e.estimate > nullValue;
+        const harmfulSide = lowerIsHelpful ? e.estimate > nullValue : e.estimate < nullValue;
+        if (harmfulSide)
+            return { verdict: 'loss', reachesMinimum: false };
+        if (helpfulSide) {
+            const reaches = lowerIsHelpful ? e.estimate <= spec.minimumEffectAt : e.estimate >= spec.minimumEffectAt;
+            return { verdict: 'win', reachesMinimum: reaches };
+        }
+    }
+    const insideBand = e.low >= spec.tieBand[0] && e.high <= spec.tieBand[1];
+    return insideBand ? { verdict: 'tie', reachesMinimum: false } : inconclusive;
+}
+/** A win must hold under both codings of not-applicable, while a loss under either is reported. */
+export function combineCodings(a, b) {
+    if (a.verdict === 'loss' || b.verdict === 'loss')
+        return { verdict: 'loss', reachesMinimum: false };
+    if (a.verdict === 'win' && b.verdict === 'win') {
+        return { verdict: 'win', reachesMinimum: a.reachesMinimum && b.reachesMinimum };
+    }
+    if (a.verdict === 'tie' && b.verdict === 'tie')
+        return { verdict: 'tie', reachesMinimum: false };
+    return { verdict: 'inconclusive', reachesMinimum: false };
+}
+/** Cost ratio's upper bound below 1.10, resolve difference's lower bound (a fraction) above -0.05.
+ * Both estimates must use alpha 0.05, the conservative reading of "upper 95% bound"; a NaN bound fails. */
+export function harmGate(costRatio, resolveDiff) {
+    const costOk = costRatio.high < 1.1;
+    const resolveOk = resolveDiff.low > -0.05;
+    return { pass: costOk && resolveOk, costOk, resolveOk };
+}
 //# sourceMappingURL=eval-stats.js.map

package/dist/extract.js

@@ -3,6 +3,7 @@ import { writeEntry } from './store.js';
 import { loadConfig } from './config.js';
 import { RejectedValueError } from './rejection.js';
 import { redactSecrets } from './secret-detect.js';
+import { fetchWithRetry, llmTimeoutMs } from './http-retry.js';
 import { neverAutoShareTags } from './shared.js';
 function isJsonString(value) {
     return typeof value === 'string';
@@ -24,7 +25,7 @@ export async function extractFacts(text, opts) {
     const fetchFn = opts.fetcher ?? fetch;
     let res;
     try {
-        res = await fetchFn('https://api.anthropic.com/v1/messages', {
+        res = await fetchWithRetry('https://api.anthropic.com/v1/messages', {
             method: 'POST',
             headers: {
                 'content-type': 'application/json',
@@ -36,7 +37,7 @@ export async function extractFacts(text, opts) {
                 max_tokens: 1200,
                 messages: [{ role: 'user', content: EXTRACTION_PROMPT + redactSecrets(text) }],
             }),
-        });
+        }, { timeoutMs: llmTimeoutMs(), fetchFn });
     }
     catch (err) {
         opts.onError?.(`request failed: ${err instanceof Error ? err.message : String(err)}`);