@remnic/bench 9.3.613 → 9.3.615

package/dist/index.d.ts

@@ -389,6 +389,8 @@ interface RunBenchmarkOptions {
     outputDir?: string;
     limit?: number;
     seed?: number;
+    /** Override the number of full-mode benchmark iterations. Quick mode remains single-run. */
+    iterations?: number;
     adapterMode?: string;
     runtimeProfile?: BenchRuntimeProfile | null;
     system: BenchMemoryAdapter;

package/dist/index.js

@@ -7316,6 +7316,7 @@ var SECRET_KEY_SEGMENT_SUFFIXES = /* @__PURE__ */ new Set([
   "bearertoken",
   "clientsecret",
   "secretkey",
+  "secretaccesskey",
   "privatekey"
 ]);
 var SECRET_MATERIAL_DESCRIPTORS = /* @__PURE__ */ new Set([
@@ -7343,7 +7344,7 @@ function isSecretSegments(segments) {
     return true;
   }
   for (let width = 2; width <= Math.min(3, segments.length); width += 1) {
-    const candidate = segments.slice(-width).join("");
+    const candidate = segments.slice(segments.length - width).join("");
     if (SECRET_KEY_SEGMENT_SUFFIXES.has(candidate)) {
       return true;
     }
@@ -10639,6 +10640,7 @@ function rougeL(predicted, expected) {
   return 2 * precision * recall / (precision + recall);
 }
 function recallAtK(retrieved, relevant, k) {
+  if (!Number.isInteger(k) || k <= 0) return 0;
   if (relevant.length === 0) return 1;
   const topK = retrieved.slice(0, k).map(normalizeText);
   const relevantSet = new Set(relevant.map(normalizeText));
@@ -19924,7 +19926,7 @@ function resolveTrialLimit2(raw) {
   if (raw === void 0 || raw === null) {
     return void 0;
   }
-  const parsed = typeof raw === "number" ? raw : Number(raw);
+  const parsed = typeof raw === "number" ? raw : typeof raw === "string" && raw.trim().length > 0 && /^[0-9]+$/.test(raw.trim()) ? Number(raw.trim()) : Number.NaN;
   if (!Number.isInteger(parsed) || parsed < 0) {
     throw new Error(
       "MemoryAgentBench benchmarkOptions.trialLimit must be a non-negative integer."
@@ -26584,12 +26586,16 @@ import path29 from "path";
 
 // src/run-seeds.ts
 function buildBenchmarkRunSeeds(runCount, baseSeed) {
-  if (!Number.isInteger(runCount) || runCount <= 0) {
-    throw new Error("benchmark run count must be a positive integer");
+  if (!Number.isSafeInteger(runCount) || runCount <= 0) {
+    throw new Error("benchmark run count must be a positive integer within JavaScript safe integer range");
   }
   const firstSeed = baseSeed ?? 0;
-  if (!Number.isInteger(firstSeed) || firstSeed < 0) {
-    throw new Error("benchmark seed must be a non-negative integer");
+  if (!Number.isSafeInteger(firstSeed) || firstSeed < 0) {
+    throw new Error("benchmark seed must be a non-negative integer within JavaScript safe integer range");
+  }
+  const maxOffset = Number.MAX_SAFE_INTEGER - firstSeed;
+  if (runCount - 1 > maxOffset) {
+    throw new Error("benchmark seed sequence must stay within JavaScript safe integer range");
   }
   return Array.from({ length: runCount }, (_, index) => firstSeed + index);
 }
@@ -28498,15 +28504,6 @@ async function runContradictionDetectionBenchmark(options) {
   const remnicVersion = await getRemnicVersion();
   const totalLatencyMs = tasks.reduce((sum, task) => sum + task.latencyMs, 0);
   const meanQueryLatencyMs = tasks.length > 0 ? totalLatencyMs / tasks.length : 0;
-  tasks.push({
-    taskId: "_aggregate_verdict_metrics",
-    question: "Per-verdict precision/recall/F1",
-    expected: "see scores",
-    actual: "see scores",
-    scores: verdictScores,
-    latencyMs: 0,
-    tokens: { input: 0, output: 0 }
-  });
   return {
     meta: {
       id: randomUUID29(),
@@ -28536,7 +28533,10 @@ async function runContradictionDetectionBenchmark(options) {
     },
     results: {
       tasks,
-      aggregates: aggregateTaskScores(tasks.map((task) => task.scores))
+      aggregates: {
+        ...aggregateTaskScores(tasks.map((task) => task.scores)),
+        ...aggregateTaskScores([verdictScores])
+      }
     },
     environment: {
       os: process.platform,
@@ -30004,7 +30004,7 @@ async function runCustomBenchmark(spec, options) {
       `Custom benchmark "${spec.name}" uses llm_judge scoring but no judge provider is configured.`
     );
   }
-  const runCount = resolveBenchmarkRunCount(options.mode);
+  const runCount = resolveBenchmarkRunCount(options.mode, options.iterations);
   const tasksPerRun = selectTasks(spec, options.limit);
   const totalTaskCount = runCount * tasksPerRun.length;
   let completedTaskCount = 0;
@@ -30021,7 +30021,7 @@ async function runCustomBenchmark(spec, options) {
         options.onTaskComplete?.(taskResult, completedTaskCount, totalTaskCount);
       }
     ),
-    void 0,
+    options.iterations,
     options.seed
   );
   const tasks = runs.flat();
@@ -32286,10 +32286,11 @@ function createMitigatedTarget(config) {
   const timestamps = [];
   function recordAndCheck(now) {
     const cutoff = now - budgetWindowMs;
-    while (timestamps.length > 0 && timestamps[0].ts <= cutoff) {
+    while (timestamps.length > 0 && timestamps[0].ts < cutoff) {
       timestamps.shift();
     }
-    if (timestamps.length >= budgetHardLimit) {
+    const projected = timestamps.length + 1;
+    if (projected >= budgetHardLimit) {
       return false;
     }
     timestamps.push({ ts: now });

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/bench",
-  "version": "9.3.613",
+  "version": "9.3.615",
   "description": "Retrieval latency ladder benchmarks + CI regression gates for @remnic/core",
   "type": "module",
   "main": "./dist/index.js",
@@ -34,7 +34,7 @@
   "dependencies": {
     "hyparquet": "^1.25.7",
     "yaml": "^2.4.2",
-    "@remnic/core": "^9.3.613"
+    "@remnic/core": "^9.3.615"
   },
   "devDependencies": {
     "tsup": "^8.5.1",