@remnic/bench 9.52.1 → 9.53.0

package/dist/index.d.ts

@@ -4589,6 +4589,8 @@ type LoadedJudgeCalibrationState = BenchmarkArtifactJudgeCalibration & Partial<J
     sourceResultId?: string;
     /** Stable hash of the exact question/predicted/expected triples judged. */
     answerSetHash?: string;
+    /** Stable hash of the full ordered source question-id list. */
+    orderedQuestionIdsHash?: string;
     /** Question ids in the pinned calibration slice, in verdict order. */
     sliceQuestionIds?: readonly string[];
     /** Full sanitized provider configuration hashes bound to this calibration. */
@@ -4732,14 +4734,16 @@ declare function hashOrderedQuestionIds(questionIds: readonly string[]): string;
  * then-rename (cursor review: a direct writeFile can leave truncated JSON on
  * crash, which loadJudgeCalibrationState would silently drop as a miss).
  *
- * Only the artifact-relevant subset (`BenchmarkArtifactJudgeCalibration`) is
- * persisted — never the per-question verdicts or answer text (repo ethics +
- * rule 10: nothing interpolated into shell). Optional `identities` record the
- * judge pair that produced the kappa so a later run can refuse a stale kappa
- * for a different pair (codex P2 review); omitted on pre-binding state files.
+ * The artifact-relevant subset (`BenchmarkArtifactJudgeCalibration`) plus
+ * private source provenance is persisted — never the per-question verdicts
+ * or answer text (repo ethics + rule 10: nothing interpolated into shell).
+ * Optional `identities` record the judge pair that produced the kappa so a
+ * later run can refuse a stale kappa for a different pair (codex P2 review);
+ * omitted on pre-binding state files.
  */
 declare function writeJudgeCalibrationState(result: JudgeCalibrationResult, calibrationDir: string, identities?: JudgeCalibrationIdentities, provenance?: {
     sourceResultId: string;
+    orderedQuestionIdsHash?: string;
     localJudgeConfigHash?: string;
     frontierJudgeConfigHash?: string;
 }): Promise<string>;

package/dist/index.js

@@ -40672,11 +40672,15 @@ async function loadJudgeCalibrationState(benchmarkId, calibrationDir) {
   }
   const sourceResultId = record.sourceResultId;
   const answerSetHash = record.answerSetHash;
+  const orderedQuestionIdsHash = record.orderedQuestionIdsHash;
   if (typeof sourceResultId === "string" && sourceResultId.length > 0 && typeof answerSetHash === "string" && /^[0-9a-f]{64}$/.test(answerSetHash) && Array.isArray(record.sliceQuestionIds) && record.sliceQuestionIds.length > 0 && record.sliceQuestionIds.length <= CALIBRATION_SLICE_SIZE && record.sliceQuestionIds.length === sampleSize && record.sliceQuestionIds.every((id) => typeof id === "string" && id.length > 0) && new Set(record.sliceQuestionIds).size === record.sliceQuestionIds.length) {
     loaded.sourceResultId = sourceResultId;
     loaded.answerSetHash = answerSetHash;
     loaded.sliceQuestionIds = record.sliceQuestionIds;
   }
+  if (typeof orderedQuestionIdsHash === "string" && /^[0-9a-f]{64}$/.test(orderedQuestionIdsHash)) {
+    loaded.orderedQuestionIdsHash = orderedQuestionIdsHash;
+  }
   if (typeof record.localJudgeConfigHash === "string" && /^[0-9a-f]{64}$/.test(record.localJudgeConfigHash) && typeof record.frontierJudgeConfigHash === "string" && /^[0-9a-f]{64}$/.test(record.frontierJudgeConfigHash)) {
     loaded.localJudgeConfigHash = record.localJudgeConfigHash;
     loaded.frontierJudgeConfigHash = record.frontierJudgeConfigHash;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@remnic/bench",
-  "version": "9.52.1",
+  "version": "9.53.0",
   "description": "Retrieval latency ladder benchmarks + CI regression gates for @remnic/core",
   "type": "module",
   "main": "./dist/index.js",
@@ -41,8 +41,8 @@
     "hyparquet": "^1.25.7",
     "yaml": "^2.4.2",
     "zod": "^3.24.0",
-    "@remnic/coding-graph": "^9.52.1",
-    "@remnic/core": "^9.52.1"
+    "@remnic/coding-graph": "^9.53.0",
+    "@remnic/core": "^9.53.0"
   },
   "devDependencies": {
     "tsup": "^8.5.1",