@remnic/bench 9.3.675 → 9.3.677

package/dist/index.js

@@ -10562,12 +10562,315 @@ function isPlainObject(value) {
 
 // src/benchmark.ts
 import fs2 from "fs";
-import path31 from "path";
+import path32 from "path";
+import { createHash as createHash9 } from "crypto";
+import { expandTildePath as expandTildePath3 } from "@remnic/core";
+
+// src/judges/judge-cache.ts
+import { createHash as createHash6, randomBytes as randomBytes2 } from "crypto";
+import {
+  mkdir as mkdir8,
+  readFile as readFile9,
+  rename as rename2,
+  rm as rm3,
+  writeFile as writeFile8
+} from "fs/promises";
+import path11 from "path";
+var JUDGE_CACHE_PROTOCOL_VERSION = "judge-protocol-v1";
+function stableStringify2(value) {
+  if (Array.isArray(value)) {
+    return `[${value.map((item) => stableStringify2(item)).join(",")}]`;
+  }
+  if (value !== null && typeof value === "object") {
+    const record = value;
+    const keys = Object.keys(record).sort();
+    const body = keys.map((key) => `${JSON.stringify(key)}:${stableStringify2(record[key])}`).join(",");
+    return `{${body}}`;
+  }
+  return JSON.stringify(value) ?? "null";
+}
+var JudgeCache = class {
+  dir;
+  // Per-key write serialization so concurrent writers never race a temp-file
+  // rename into place for the same key. Cached entries are read straight from
+  // disk, so reads remain lock-free.
+  writeQueues = /* @__PURE__ */ new Map();
+  // PR #1591 round-8 (cursor thread): in-memory layer for fire-and-forget
+  // writes. putSafely chains cache.put onto a pendingWrites promise
+  // without awaiting, so a second benchmark iteration can call get()
+  // before the disk rename lands. The inflight map is populated
+  // synchronously inside put() (before the first await) and cleared in
+  // the finally after the write settles — closing the gap without
+  // changing the byte-identical baseline (a fresh process has no
+  // inflight entries).
+  inflight = /* @__PURE__ */ new Map();
+  cachedDirExists = false;
+  constructor(options) {
+    this.dir = path11.resolve(options.dir);
+  }
+  /** Compute the sha256-hex key for a set of parts. Pure, sync, side-effect-free. */
+  computeKey(parts) {
+    const fieldDigest = (value) => createHash6("sha256").update(value).digest();
+    return createHash6("sha256").update(fieldDigest(parts.benchmarkId)).update(fieldDigest(parts.datasetVersion)).update(fieldDigest(parts.questionId)).update(fieldDigest(parts.answerText)).update(fieldDigest(parts.judgePromptHash)).update(fieldDigest(parts.judgeModelId)).update(fieldDigest(parts.judgeParamsHash)).digest("hex");
+  }
+  /**
+   * Read a previously-stored verdict. Returns `undefined` on miss, corrupted
+   * entry, missing required field, or read error — never throws, never
+   * fabricates.
+   */
+  async get(parts) {
+    const key = this.computeKey(parts);
+    const inflightHit = this.inflight.get(key);
+    if (inflightHit !== void 0) {
+      return {
+        cacheHit: true,
+        verdict: inflightHit.verdict,
+        storedAt: inflightHit.storedAt
+      };
+    }
+    const filePath = this.entryPath(key);
+    let raw;
+    try {
+      raw = await readFile9(filePath, "utf8");
+    } catch {
+      return void 0;
+    }
+    const envelope = parseEnvelope(raw);
+    if (envelope === void 0) return void 0;
+    return {
+      cacheHit: true,
+      verdict: envelope.verdict,
+      storedAt: envelope.storedAt
+    };
+  }
+  /**
+   * Persist a verdict atomically: write to a temp file then rename into
+   * place. Concurrent writes for the same key serialize via an in-memory
+   * chain so the temp-file never lands on top of a sibling rename.
+   */
+  async put(parts, verdict) {
+    const key = this.computeKey(parts);
+    const envelope = {
+      storedAt: (/* @__PURE__ */ new Date()).toISOString(),
+      key,
+      verdict
+    };
+    this.inflight.set(key, envelope);
+    const prior = this.writeQueues.get(key) ?? Promise.resolve();
+    const next = prior.then(() => this.writeOne(key, envelope));
+    const tracked = next.catch(() => void 0);
+    this.writeQueues.set(key, tracked);
+    try {
+      await next;
+    } finally {
+      if (this.writeQueues.get(key) === tracked) {
+        this.writeQueues.delete(key);
+      }
+      if (this.inflight.get(key) === envelope) {
+        this.inflight.delete(key);
+      }
+    }
+  }
+  /** Number of in-flight per-key write chains (diagnostic/test seam). */
+  pendingWriteCount() {
+    return this.writeQueues.size;
+  }
+  async writeOne(key, envelope) {
+    if (!this.cachedDirExists) {
+      await mkdir8(this.dir, { recursive: true });
+      this.cachedDirExists = true;
+    }
+    const filePath = this.entryPath(key);
+    const tempPath = path11.join(
+      this.dir,
+      `.${key}.${randomBytes2(6).toString("hex")}.tmp`
+    );
+    await writeFile8(tempPath, `${JSON.stringify(envelope)}
+`, "utf8");
+    try {
+      await rename2(tempPath, filePath);
+    } catch (error) {
+      await rm3(tempPath, { force: true }).catch(() => void 0);
+      throw error;
+    }
+  }
+  entryPath(key) {
+    return path11.join(this.dir, `${key}.json`);
+  }
+};
+function runJudgeWithCache(options) {
+  const { judge, cache } = options;
+  const keyExtras = options.keyExtras ?? {};
+  const counters = {
+    modelCalls: 0,
+    cacheHits: 0,
+    cacheMisses: 0,
+    cacheWriteFailures: 0
+  };
+  let pendingWrites = Promise.resolve();
+  const putSafely = (parts, verdict, control) => {
+    if (!cache) return;
+    if (control?.signal?.aborted) return;
+    const write = cache.put(parts, verdict).catch(() => {
+      counters.cacheWriteFailures += 1;
+    });
+    pendingWrites = pendingWrites.then(() => write);
+  };
+  const CACHE_READ_BUDGET_MS = 250;
+  async function readCacheWithAbort(cache2, parts, control) {
+    if (control?.signal?.aborted) return void 0;
+    const read = cache2.get(parts);
+    const readBudget = new Promise((resolveBudget) => {
+      setTimeout(() => {
+        resolveBudget(void 0);
+      }, CACHE_READ_BUDGET_MS);
+    });
+    return Promise.race([read, readBudget]);
+  }
+  const cachedVerdict = (stored) => ({
+    score: stored.score,
+    tokens: { input: 0, output: 0 },
+    latencyMs: 0,
+    ...stored.model !== void 0 ? { model: stored.model } : {}
+  });
+  const wrapper = {
+    counters,
+    cache,
+    drainPendingWrites: () => pendingWrites,
+    async score(question, predicted, expected, control) {
+      const detailed = await wrapper.scoreWithMetrics(
+        question,
+        predicted,
+        expected,
+        control
+      );
+      return detailed.score;
+    },
+    async scoreWithMetrics(question, predicted, expected, control) {
+      const answerText = `${predicted}
${expected}`;
+      const parts = {
+        benchmarkId: keyExtras.benchmarkId ?? "unknown-benchmark",
+        datasetVersion: keyExtras.datasetVersion ?? "unknown-version",
+        questionId: question,
+        answerText,
+        judgePromptHash: keyExtras.judgePromptHash ?? "unknown-prompt",
+        judgeModelId: keyExtras.judgeModelId ?? "unknown-judge",
+        judgeParamsHash: keyExtras.judgeParamsHash ?? "unknown-params"
+      };
+      if (cache) {
+        let hit;
+        try {
+          hit = await readCacheWithAbort(cache, parts, control);
+        } catch {
+          hit = void 0;
+        }
+        if (hit) {
+          counters.cacheHits += 1;
+          return cachedVerdict(hit.verdict);
+        }
+        counters.cacheMisses += 1;
+      }
+      if (!judge.scoreWithMetrics) {
+        counters.modelCalls += 1;
+        const scoreStartedAt = Date.now();
+        const scoreValue = judge.score ? await judge.score(question, predicted, expected, control) : 0;
+        const synthesized = {
+          score: scoreValue,
+          tokens: { input: 0, output: 0 },
+          latencyMs: Date.now() - scoreStartedAt,
+          model: keyExtras.judgeModelId ?? void 0
+        };
+        putSafely(parts, synthesized, control);
+        return synthesized;
+      }
+      counters.modelCalls += 1;
+      const fresh = await judge.scoreWithMetrics(
+        question,
+        predicted,
+        expected,
+        control
+      );
+      putSafely(parts, fresh, control);
+      return fresh;
+    }
+  };
+  if (typeof judge.scoreBinaryPrompt === "function") {
+    Object.defineProperty(wrapper, "scoreBinaryPrompt", {
+      configurable: true,
+      enumerable: true,
+      writable: false,
+      value: async function scoreBinaryPrompt(prompt, control) {
+        const parts = {
+          benchmarkId: keyExtras.benchmarkId ?? "unknown-benchmark",
+          datasetVersion: keyExtras.datasetVersion ?? "unknown-version",
+          // Binary prompts are content-sensitive: two distinct prompts of
+          // the same character length would collide on the previous
+          // `binary:N` key, so key on a sha256 prefix of the prompt body.
+          questionId: `binary:${createHash6("sha256").update(prompt).digest("hex").slice(0, 16)}`,
+          answerText: prompt,
+          judgePromptHash: keyExtras.judgePromptHash ?? "unknown-prompt",
+          judgeModelId: keyExtras.judgeModelId ?? "unknown-judge",
+          judgeParamsHash: keyExtras.judgeParamsHash ?? "unknown-params"
+        };
+        if (cache) {
+          let hit;
+          try {
+            hit = await readCacheWithAbort(cache, parts, control);
+          } catch {
+            hit = void 0;
+          }
+          if (hit) {
+            counters.cacheHits += 1;
+            return cachedVerdict(hit.verdict);
+          }
+          counters.cacheMisses += 1;
+        }
+        counters.modelCalls += 1;
+        const fresh = await judge.scoreBinaryPrompt(prompt, control);
+        putSafely(parts, fresh, control);
+        return fresh;
+      }
+    });
+  }
+  return wrapper;
+}
+function parseEnvelope(raw) {
+  let parsed;
+  try {
+    parsed = JSON.parse(raw);
+  } catch {
+    return void 0;
+  }
+  if (parsed === null || typeof parsed !== "object" || Array.isArray(parsed)) {
+    return void 0;
+  }
+  const candidate = parsed;
+  if (typeof candidate.storedAt !== "string") return void 0;
+  if (typeof candidate.key !== "string") return void 0;
+  if (!isBenchJudgeResult(candidate.verdict)) return void 0;
+  return candidate;
+}
+function isBenchJudgeResult(value) {
+  if (value === null || typeof value !== "object" || Array.isArray(value)) {
+    return false;
+  }
+  const v = value;
+  if (typeof v.score !== "number" || !Number.isFinite(v.score)) return false;
+  if (v.tokens === null || typeof v.tokens !== "object" || Array.isArray(v.tokens)) {
+    return false;
+  }
+  const tokens = v.tokens;
+  if (typeof tokens.input !== "number" || !Number.isFinite(tokens.input)) return false;
+  if (typeof tokens.output !== "number" || !Number.isFinite(tokens.output)) return false;
+  if (typeof v.latencyMs !== "number" || !Number.isFinite(v.latencyMs)) return false;
+  if (v.model !== void 0 && typeof v.model !== "string") return false;
+  return true;
+}
 
 // src/benchmarks/published/ama-bench/runner.ts
 import { randomUUID as randomUUID2 } from "crypto";
-import { readFile as readFile9 } from "fs/promises";
-import path11 from "path";
+import { readFile as readFile10 } from "fs/promises";
+import path12 from "path";
 
 // src/benchmarks/published/ama-bench/fixture.ts
 var AMA_BENCH_SMOKE_FIXTURE = [
@@ -11142,10 +11445,10 @@ async function loadDataset(mode, datasetDir, limit) {
     return episodes;
   };
   if (datasetDir) {
-    const filePath = path11.join(datasetDir, "open_end_qa_set.jsonl");
+    const filePath = path12.join(datasetDir, "open_end_qa_set.jsonl");
     let raw;
     try {
-      raw = await readFile9(filePath, "utf8");
+      raw = await readFile10(filePath, "utf8");
     } catch (error) {
       throw new Error(
         `AMA-Bench dataset not found at ${filePath}: ${error instanceof Error ? error.message : String(error)}`
@@ -11437,8 +11740,8 @@ function isValidQaPairs(value) {
 
 // src/benchmarks/published/amemgym/runner.ts
 import { randomUUID as randomUUID3 } from "crypto";
-import { readFile as readFile10 } from "fs/promises";
-import path12 from "path";
+import { readFile as readFile11 } from "fs/promises";
+import path13 from "path";
 
 // src/benchmarks/published/amemgym/fixture.ts
 var AMEMGYM_SMOKE_FIXTURE = [
@@ -11967,7 +12270,7 @@ async function loadDataset2(mode, datasetDir, limit) {
     const datasetErrors = [];
     for (const filename of DATASET_FILENAMES) {
       try {
-        const raw = await readFile10(path12.join(datasetDir, filename), "utf8");
+        const raw = await readFile11(path13.join(datasetDir, filename), "utf8");
         const parsed = parseDataset(raw, filename, normalizedLimit);
         return ensureDatasetProfiles(parsed);
       } catch (error) {
@@ -12141,8 +12444,8 @@ function normalizeRole(role) {
 
 // src/benchmarks/published/memory-arena/runner.ts
 import { randomUUID as randomUUID4 } from "crypto";
-import { readFile as readFile11, readdir as readdir5, stat as stat3 } from "fs/promises";
-import path13 from "path";
+import { readFile as readFile12, readdir as readdir5, stat as stat3 } from "fs/promises";
+import path14 from "path";
 import { expandTildePath as expandTildePath2 } from "@remnic/core";
 
 // src/benchmarks/published/memory-arena/fixture.ts
@@ -12469,7 +12772,7 @@ async function loadDataset3(mode, datasetDir, limit) {
       if (remainingLimit2 === 0) {
         break;
       }
-      const raw = await readFile11(path13.join(datasetDir, filename), "utf8");
+      const raw = await readFile12(path14.join(datasetDir, filename), "utf8");
       const parsedTasks = [];
       raw.split("\n").forEach((line, lineIndex) => {
         if (line.trim().length === 0) {
@@ -12805,7 +13108,7 @@ async function loadMemoryArenaWebshopProductCatalog(datasetDir) {
       `MemoryArena WebShop product sidecar is ${sourceStat.size} bytes; provide a compact JSON/JSONL sidecar smaller than ${MEMORY_ARENA_WEBSHOP_PRODUCTS_MAX_BYTES} bytes instead of the full WebShop catalog.`
     );
   }
-  const raw = await readFile11(sourcePath, "utf8");
+  const raw = await readFile12(sourcePath, "utf8");
   const records = parseMemoryArenaWebshopSidecarRecords(raw, sourcePath);
   const byAsin = /* @__PURE__ */ new Map();
   for (const record of records) {
@@ -12825,14 +13128,14 @@ async function loadMemoryArenaWebshopProductCatalog(datasetDir) {
 async function resolveMemoryArenaWebshopProductCatalogPath(datasetDir) {
   const configuredPath = process.env[MEMORY_ARENA_WEBSHOP_PRODUCTS_ENV]?.trim();
   if (configuredPath && configuredPath.length > 0) {
-    return path13.resolve(expandTildePath2(configuredPath));
+    return path14.resolve(expandTildePath2(configuredPath));
   }
   if (datasetDir === void 0) {
     return void 0;
   }
   const candidatePaths = [
     ...MEMORY_ARENA_WEBSHOP_PRODUCT_SIDECAR_FILENAMES
-  ].map((filename) => path13.join(datasetDir, filename));
+  ].map((filename) => path14.join(datasetDir, filename));
   for (const candidatePath of candidatePaths) {
     try {
       const candidateStat = await stat3(candidatePath);
@@ -14254,8 +14557,8 @@ function scoreSubtaskSuccess(scores) {
 import { collectTemporalLexicalCues } from "@remnic/core";
 
 // src/benchmarks/published/dataset-loader.ts
-import { readFile as readFile12 } from "fs/promises";
-import path14 from "path";
+import { readFile as readFile13 } from "fs/promises";
+import path15 from "path";
 
 // src/benchmarks/published/longmemeval/fixture.ts
 var LONG_MEM_EVAL_SMOKE_FIXTURE = [
@@ -14358,10 +14661,10 @@ async function loadDataset4(options) {
   const errors = [];
   if (options.datasetDir) {
     for (const filename of options.filenames) {
-      const abs = path14.join(options.datasetDir, filename);
+      const abs = path15.join(options.datasetDir, filename);
       let raw;
       try {
-        raw = await readFile12(abs, "utf8");
+        raw = await readFile13(abs, "utf8");
       } catch (error) {
         errors.push(
           `${filename}: ${error instanceof Error ? error.message : String(error)}`
@@ -16098,7 +16401,7 @@ function normalizeQaArray(value, location) {
 import { randomUUID as randomUUID6 } from "crypto";
 import { createReadStream as createReadStream2 } from "fs";
 import { readdir as readdir6 } from "fs/promises";
-import path15 from "path";
+import path16 from "path";
 import { createInterface } from "readline/promises";
 import {
   asyncBufferFromFile,
@@ -16569,8 +16872,8 @@ async function listBeamDatasetFiles(datasetDir) {
     return directFiles;
   }
   try {
-    const nestedFilenames = await readdir6(path15.join(datasetDir, "data"));
-    return nestedFilenames.filter((filename) => isBeamDatasetFilename(filename)).map((filename) => path15.join("data", filename));
+    const nestedFilenames = await readdir6(path16.join(datasetDir, "data"));
+    return nestedFilenames.filter((filename) => isBeamDatasetFilename(filename)).map((filename) => path16.join("data", filename));
   } catch {
     return [];
   }
@@ -16597,7 +16900,7 @@ async function* iterateDatasetFiles(datasetDir, datasetFiles, limit) {
   let remainingLimit = limit;
   for (const filename of datasetFiles) {
     const scale = inferScaleFromFilename(filename);
-    const filePath = path15.join(datasetDir, filename);
+    const filePath = path16.join(datasetDir, filename);
     const conversations = filename.endsWith(".jsonl") ? streamJsonlDataset(filePath, filename, remainingLimit) : filename.endsWith(".parquet") ? streamParquetDataset(filePath, filename, remainingLimit) : streamJsonDataset(filePath, filename, remainingLimit);
     for await (const conversation of conversations) {
       yield {
@@ -17608,9 +17911,9 @@ var StructuredLiteralParser = class {
 };
 
 // src/benchmarks/published/personamem/runner.ts
-import { createHash as createHash6, randomUUID as randomUUID7 } from "crypto";
-import { readFile as readFile13, realpath as realpath4 } from "fs/promises";
-import path16 from "path";
+import { createHash as createHash7, randomUUID as randomUUID7 } from "crypto";
+import { readFile as readFile14, realpath as realpath4 } from "fs/promises";
+import path17 from "path";
 
 // src/benchmarks/published/personamem/fixture.ts
 var PERSONAMEM_SMOKE_FIXTURE = [
@@ -17886,10 +18189,10 @@ async function loadDataset8(mode, datasetDir, limit) {
   if (datasetDir) {
     const datasetErrors = [];
     for (const relativePath of DATASET_FILE_CANDIDATES) {
-      const datasetPath = path16.join(datasetDir, relativePath);
+      const datasetPath = path17.join(datasetDir, relativePath);
       let raw;
       try {
-        raw = await readFile13(datasetPath, "utf8");
+        raw = await readFile14(datasetPath, "utf8");
       } catch (error) {
         datasetErrors.push(
           `${relativePath}: ${error instanceof Error ? error.message : String(error)}`
@@ -17947,7 +18250,7 @@ async function hydrateSample(row, datasetRoot) {
     datasetRoot,
     row.chat_history_32k_link
   );
-  const chatHistoryRaw = await readFile13(chatHistoryPath, "utf8");
+  const chatHistoryRaw = await readFile14(chatHistoryPath, "utf8");
   const chatHistory = parseChatHistory(
     chatHistoryRaw,
     row.chat_history_32k_link
@@ -18080,12 +18383,12 @@ function parseCsv(raw, limit) {
   return rows;
 }
 async function resolveDatasetFilePath(datasetRoot, relativePath) {
-  const rootPath = path16.resolve(datasetRoot);
+  const rootPath = path17.resolve(datasetRoot);
   const rootRealPath = await realpath4(rootPath);
-  const candidatePath = path16.resolve(rootPath, relativePath);
+  const candidatePath = path17.resolve(rootPath, relativePath);
   const candidateRealPath = await realpath4(candidatePath);
-  const relativeToRoot = path16.relative(rootRealPath, candidateRealPath);
-  if (relativeToRoot.startsWith("..") || path16.isAbsolute(relativeToRoot)) {
+  const relativeToRoot = path17.relative(rootRealPath, candidateRealPath);
+  if (relativeToRoot.startsWith("..") || path17.isAbsolute(relativeToRoot)) {
     throw new Error(
       `PersonaMem-v2 dataset file reference "${relativePath}" must stay within datasetDir.`
     );
@@ -18213,7 +18516,7 @@ function buildMcqPrompt(sample, seed) {
 function deterministicShuffle(values, seedMaterial) {
   return values.map((value, index) => ({
     value,
-    key: createHash6("sha256").update(`${seedMaterial}:${index}:${value}`).digest("hex"),
+    key: createHash7("sha256").update(`${seedMaterial}:${index}:${value}`).digest("hex"),
     index
   })).sort((left, right) => {
     const byKey = left.key.localeCompare(right.key);
@@ -18413,8 +18716,8 @@ function applyLimit6(items, limit) {
 
 // src/benchmarks/published/membench/runner.ts
 import { randomUUID as randomUUID8 } from "crypto";
-import { readFile as readFile14, readdir as readdir7 } from "fs/promises";
-import path17 from "path";
+import { readFile as readFile15, readdir as readdir7 } from "fs/promises";
+import path18 from "path";
 
 // src/benchmarks/published/membench/fixture.ts
 var MEMBENCH_SMOKE_FIXTURE = [
@@ -18675,7 +18978,7 @@ async function loadDataset9(mode, datasetDir, limit) {
     let remainingLimit = normalizedLimit;
     for (const filename of filenames) {
       try {
-        const raw = await readFile14(path17.join(datasetDir, filename), "utf8");
+        const raw = await readFile15(path18.join(datasetDir, filename), "utf8");
         const parsed = filename.endsWith(".jsonl") ? parseJsonlDataset(raw, filename) : parseJsonDataset(raw, filename);
         const limitedCases = remainingLimit === 0 ? [] : applyLimit7(parsed, remainingLimit);
         if (limitedCases.length > 0) {
@@ -19542,8 +19845,8 @@ function isPlainObject2(value) {
 
 // src/benchmarks/published/memoryagentbench/runner.ts
 import { randomUUID as randomUUID9 } from "crypto";
-import { access, readFile as readFile15 } from "fs/promises";
-import path18 from "path";
+import { access, readFile as readFile16 } from "fs/promises";
+import path19 from "path";
 
 // src/benchmarks/published/memoryagentbench/fixture.ts
 var MEMORY_AGENT_BENCH_SMOKE_FIXTURE = [
@@ -20565,7 +20868,7 @@ async function loadRecSysEntityMapping(datasetDir) {
     }
     let parsed;
     try {
-      parsed = JSON.parse(await readFile15(candidate, "utf8"));
+      parsed = JSON.parse(await readFile16(candidate, "utf8"));
     } catch (error) {
       console.error(
         `  [WARN] MemoryAgentBench ReDial entity mapping ${candidate} is invalid JSON; trying the next candidate: ${error instanceof Error ? error.message : String(error)}`
@@ -20622,21 +20925,21 @@ function recsysEntityMappingCandidates(datasetDir) {
   if (!datasetDir) {
     return [];
   }
-  const absoluteDatasetDir = path18.resolve(datasetDir);
+  const absoluteDatasetDir = path19.resolve(datasetDir);
   const roots = [
     absoluteDatasetDir,
-    path18.dirname(absoluteDatasetDir)
+    path19.dirname(absoluteDatasetDir)
   ];
   const canonicalSuffixes = [
-    path18.join("processed_data", "Recsys_Redial", "entity2id.json"),
-    path18.join("Recsys_Redial", "entity2id.json")
+    path19.join("processed_data", "Recsys_Redial", "entity2id.json"),
+    path19.join("Recsys_Redial", "entity2id.json")
   ];
   const looseSuffixes = ["entity2id.json"];
   return [
     ...roots.flatMap(
-      (root) => canonicalSuffixes.map((suffix) => path18.join(root, suffix))
+      (root) => canonicalSuffixes.map((suffix) => path19.join(root, suffix))
     ),
-    ...looseSuffixes.map((suffix) => path18.join(absoluteDatasetDir, suffix))
+    ...looseSuffixes.map((suffix) => path19.join(absoluteDatasetDir, suffix))
   ];
 }
 async function fileExists(filePath) {
@@ -20673,7 +20976,7 @@ async function loadDataset10(mode, datasetDir, limit) {
     const datasetErrors = [];
     for (const filename of DATASET_BUNDLE_CANDIDATES) {
       const parsed = await tryReadDatasetFile(
-        path18.join(datasetDir, filename),
+        path19.join(datasetDir, filename),
         filename,
         datasetErrors
       );
@@ -20690,7 +20993,7 @@ async function loadDataset10(mode, datasetDir, limit) {
       let splitData;
       for (const filename of splitConfig.candidates) {
         try {
-          splitData = await readDatasetFile(path18.join(datasetDir, filename), filename);
+          splitData = await readDatasetFile(path19.join(datasetDir, filename), filename);
           break;
         } catch (error) {
           if (!isFileNotFoundError2(error)) {
@@ -20728,7 +21031,7 @@ async function loadDataset10(mode, datasetDir, limit) {
   return ensureDatasetItems(applyLimit8(MEMORY_AGENT_BENCH_SMOKE_FIXTURE, normalizedLimit));
 }
 async function readDatasetFile(filePath, filename) {
-  const raw = await readFile15(filePath, "utf8");
+  const raw = await readFile16(filePath, "utf8");
   const parsed = filename.endsWith(".jsonl") ? parseJsonLines(raw, filename) : parseJsonArray(raw, filename);
   return parsed.map(
     (item, index) => parseMemoryAgentBenchItem(item, `${filename} item ${index + 1}`)
@@ -21339,7 +21642,7 @@ function loadCases(mode, limit) {
 // src/benchmarks/remnic/extraction-judge-calibration/runner.ts
 import { randomUUID as randomUUID11 } from "crypto";
 import os4 from "os";
-import path19 from "path";
+import path20 from "path";
 import {
   createVerdictCache,
   judgeFactDurability,
@@ -21449,8 +21752,8 @@ var extractionJudgeCalibrationDefinition = {
 async function runExtractionJudgeCalibrationBenchmark(options) {
   const cases = loadCases2(options.mode, options.limit);
   const config = parseConfig2({
-    memoryDir: path19.join(os4.tmpdir(), "remnic-bench-extraction-judge"),
-    workspaceDir: path19.join(os4.tmpdir(), "remnic-bench-extraction-judge-workspace"),
+    memoryDir: path20.join(os4.tmpdir(), "remnic-bench-extraction-judge"),
+    workspaceDir: path20.join(os4.tmpdir(), "remnic-bench-extraction-judge-workspace"),
     openaiApiKey: "bench-test-key",
     extractionJudgeEnabled: true,
     extractionJudgeBatchSize: 4,
@@ -21999,9 +22302,9 @@ function constantAggregate2(value) {
 
 // src/benchmarks/remnic/entity-consolidation/runner.ts
 import os5 from "os";
-import path20 from "path";
+import path21 from "path";
 import { randomUUID as randomUUID13 } from "crypto";
-import { mkdtemp as mkdtemp3, rm as rm3 } from "fs/promises";
+import { mkdtemp as mkdtemp3, rm as rm4 } from "fs/promises";
 import { StorageManager } from "@remnic/core";
 
 // src/benchmarks/remnic/entity-consolidation/fixture.ts
@@ -22162,7 +22465,7 @@ function loadCases4(mode, limit) {
   return limited;
 }
 async function executeCase(sample) {
-  const tmpDir = await mkdtemp3(path20.join(os5.tmpdir(), "remnic-bench-entity-consolidation-"));
+  const tmpDir = await mkdtemp3(path21.join(os5.tmpdir(), "remnic-bench-entity-consolidation-"));
   try {
     const storage = new StorageManager(tmpDir);
     await storage.ensureDirectories();
@@ -22170,7 +22473,7 @@ async function executeCase(sample) {
     const rawEntity = await storage.readEntity(canonicalName);
     return summarizeEntity(rawEntity, canonicalName);
   } finally {
-    await rm3(tmpDir, { recursive: true, force: true });
+    await rm4(tmpDir, { recursive: true, force: true });
   }
 }
 async function applyScenario(storage, sample) {
@@ -22341,9 +22644,9 @@ function parseNonNegativeInt(rawValue) {
 
 // src/benchmarks/remnic/page-versioning/runner.ts
 import { randomUUID as randomUUID14 } from "crypto";
-import { mkdir as mkdir8, mkdtemp as mkdtemp4, readFile as readFile16, rm as rm4, writeFile as writeFile8 } from "fs/promises";
+import { mkdir as mkdir9, mkdtemp as mkdtemp4, readFile as readFile17, rm as rm5, writeFile as writeFile9 } from "fs/promises";
 import os6 from "os";
-import path21 from "path";
+import path22 from "path";
 import {
   createVersion,
   diffVersions,
@@ -22507,21 +22810,21 @@ function loadCases5(mode, limit) {
   return limited;
 }
 async function executeCase2(sample, dependencies) {
-  const tmpDir = await mkdtemp4(path21.join(os6.tmpdir(), "remnic-bench-page-versioning-"));
+  const tmpDir = await mkdtemp4(path22.join(os6.tmpdir(), "remnic-bench-page-versioning-"));
   try {
-    const factsDir = path21.join(tmpDir, "facts");
-    const pagePath = path21.join(factsDir, `${sample.id}.md`);
-    await mkdir8(factsDir, { recursive: true });
+    const factsDir = path22.join(tmpDir, "facts");
+    const pagePath = path22.join(factsDir, `${sample.id}.md`);
+    await mkdir9(factsDir, { recursive: true });
     const config = versioningConfig();
     switch (sample.scenario) {
       case "revert-flow": {
-        await writeFile8(pagePath, "original content", "utf-8");
+        await writeFile9(pagePath, "original content", "utf-8");
         await dependencies.createVersion(pagePath, "original content", "write", config, void 0, void 0, tmpDir);
-        await writeFile8(pagePath, "modified content", "utf-8");
+        await writeFile9(pagePath, "modified content", "utf-8");
         await dependencies.createVersion(pagePath, "modified content", "write", config, void 0, void 0, tmpDir);
         await dependencies.revertToVersion(pagePath, "1", config, void 0, tmpDir);
         const history = await dependencies.listVersions(pagePath, config, tmpDir);
-        const pageContent = await readFile16(pagePath, "utf-8");
+        const pageContent = await readFile17(pagePath, "utf-8");
         const observed = await dependencies.getVersion(pagePath, "3", config, tmpDir);
         return {
           versionIds: history.versions.map((version) => version.versionId),
@@ -22534,11 +22837,11 @@ async function executeCase2(sample, dependencies) {
         const pruningConfig = versioningConfig({ maxVersionsPerPage: 2 });
         for (let index = 1; index <= 4; index += 1) {
           const content = `content v${index}`;
-          await writeFile8(pagePath, content, "utf-8");
+          await writeFile9(pagePath, content, "utf-8");
           await dependencies.createVersion(pagePath, content, "write", pruningConfig, void 0, void 0, tmpDir);
         }
         const history = await dependencies.listVersions(pagePath, pruningConfig, tmpDir);
-        const pageContent = await readFile16(pagePath, "utf-8");
+        const pageContent = await readFile17(pagePath, "utf-8");
         const prunedIds = [];
         for (const versionId of ["1", "2"]) {
           try {
@@ -22558,7 +22861,7 @@ async function executeCase2(sample, dependencies) {
         };
       }
       case "diff-output": {
-        await writeFile8(pagePath, "line 1\nline 2\nline 3", "utf-8");
+        await writeFile9(pagePath, "line 1\nline 2\nline 3", "utf-8");
         await dependencies.createVersion(
           pagePath,
           "line 1\nline 2\nline 3",
@@ -22568,7 +22871,7 @@ async function executeCase2(sample, dependencies) {
           void 0,
           tmpDir
         );
-        await writeFile8(pagePath, "line 1\nline 2 changed\nline 3\nline 4", "utf-8");
+        await writeFile9(pagePath, "line 1\nline 2 changed\nline 3\nline 4", "utf-8");
         await dependencies.createVersion(
           pagePath,
           "line 1\nline 2 changed\nline 3\nline 4",
@@ -22579,7 +22882,7 @@ async function executeCase2(sample, dependencies) {
           tmpDir
         );
         const history = await dependencies.listVersions(pagePath, config, tmpDir);
-        const pageContent = await readFile16(pagePath, "utf-8");
+        const pageContent = await readFile17(pagePath, "utf-8");
         const diff = await dependencies.diffVersions(pagePath, "1", "2", config, tmpDir);
         const observedLines = normalizeDiffChangedLines(diff);
         return {
@@ -22591,7 +22894,7 @@ async function executeCase2(sample, dependencies) {
       }
     }
   } finally {
-    await rm4(tmpDir, { recursive: true, force: true });
+    await rm5(tmpDir, { recursive: true, force: true });
   }
 }
 function isMissingPageVersionError(error, pagePath, versionId) {
@@ -24864,9 +25167,9 @@ function loadCases9(mode, limit) {
 
 // src/benchmarks/remnic/procedural-recall/runner.ts
 import { randomUUID as randomUUID21 } from "crypto";
-import { mkdtemp as mkdtemp5, rm as rm5 } from "fs/promises";
+import { mkdtemp as mkdtemp5, rm as rm6 } from "fs/promises";
 import os7 from "os";
-import path22 from "path";
+import path23 from "path";
 import {
   StorageManager as StorageManager2,
   parseConfig as parseConfig3,
@@ -24996,7 +25299,7 @@ async function runProceduralRecallBenchmark(options) {
   }
   for (const sample of e2eCases) {
     const startedAt = performance.now();
-    const dir = await mkdtemp5(path22.join(os7.tmpdir(), "remnic-bench-procedural-recall-"));
+    const dir = await mkdtemp5(path23.join(os7.tmpdir(), "remnic-bench-procedural-recall-"));
     let section = null;
     try {
       const storage = new StorageManager2(dir);
@@ -25011,7 +25314,7 @@ ${body}`,
       );
       const config = parseConfig3({
         memoryDir: dir,
-        workspaceDir: path22.join(dir, "ws"),
+        workspaceDir: path23.join(dir, "ws"),
         openaiApiKey: "bench-key",
         procedural: {
           enabled: sample.proceduralEnabled !== false,
@@ -25020,7 +25323,7 @@ ${body}`,
       });
       section = await buildProcedureRecallSection(storage, sample.prompt, config);
     } finally {
-      await rm5(dir, { recursive: true, force: true });
+      await rm6(dir, { recursive: true, force: true });
     }
     const latencyMs = Math.round(performance.now() - startedAt);
     const nonNull = section !== null && section.length > 0;
@@ -25081,9 +25384,9 @@ ${body}`,
 
 // src/benchmarks/remnic/ingestion-entity-recall/runner.ts
 import { randomUUID as randomUUID22 } from "crypto";
-import { mkdtemp as mkdtemp6, writeFile as writeFile9, rm as rm6, mkdir as mkdir9, realpath as realpath5 } from "fs/promises";
+import { mkdtemp as mkdtemp6, writeFile as writeFile10, rm as rm7, mkdir as mkdir10, realpath as realpath5 } from "fs/promises";
 import { tmpdir as tmpdir2 } from "os";
-import path23 from "path";
+import path24 from "path";
 
 // src/ingestion-scorer.ts
 function normalize(value) {
@@ -25585,13 +25888,13 @@ async function runIngestionEntityRecallBenchmark(options) {
     throw new Error("ingestionAdapter is required for ingestion benchmarks");
   }
   const fixture = emailFixture.generate();
-  const fixtureDir = await mkdtemp6(path23.join(tmpdir2(), "bench-email-"));
+  const fixtureDir = await mkdtemp6(path24.join(tmpdir2(), "bench-email-"));
   try {
     await options.ingestionAdapter.reset();
     for (const file of fixture.files) {
-      const filePath = path23.join(fixtureDir, file.relativePath);
-      await mkdir9(path23.dirname(filePath), { recursive: true });
-      await writeFile9(filePath, file.content, "utf8");
+      const filePath = path24.join(fixtureDir, file.relativePath);
+      await mkdir10(path24.dirname(filePath), { recursive: true });
+      await writeFile10(filePath, file.content, "utf8");
     }
     const { result: ingestionLog, durationMs } = await timed(
       async () => options.ingestionAdapter.ingest(await realpath5(fixtureDir))
@@ -25672,7 +25975,7 @@ async function runIngestionEntityRecallBenchmark(options) {
     ];
     return buildResult(options, tasks, durationMs);
   } finally {
-    await rm6(fixtureDir, { recursive: true, force: true });
+    await rm7(fixtureDir, { recursive: true, force: true });
   }
 }
 async function buildResult(options, tasks, totalLatencyMs) {
@@ -25718,9 +26021,9 @@ async function buildResult(options, tasks, totalLatencyMs) {
 
 // src/benchmarks/remnic/ingestion-schema-completeness/runner.ts
 import { randomUUID as randomUUID23 } from "crypto";
-import { mkdtemp as mkdtemp7, writeFile as writeFile10, rm as rm7, mkdir as mkdir10, realpath as realpath6 } from "fs/promises";
+import { mkdtemp as mkdtemp7, writeFile as writeFile11, rm as rm8, mkdir as mkdir11, realpath as realpath6 } from "fs/promises";
 import { tmpdir as tmpdir3 } from "os";
-import path24 from "path";
+import path25 from "path";
 var ingestionSchemaCompletenessDefinition = {
   id: "ingestion-schema-completeness",
   title: "Ingestion: Schema Completeness",
@@ -25739,13 +26042,13 @@ async function runIngestionSchemaCompletenessBenchmark(options) {
     throw new Error("ingestionAdapter is required for ingestion benchmarks");
   }
   const fixture = emailFixture.generate();
-  const fixtureDir = await mkdtemp7(path24.join(tmpdir3(), "bench-email-"));
+  const fixtureDir = await mkdtemp7(path25.join(tmpdir3(), "bench-email-"));
   try {
     await options.ingestionAdapter.reset();
     for (const file of fixture.files) {
-      const filePath = path24.join(fixtureDir, file.relativePath);
-      await mkdir10(path24.dirname(filePath), { recursive: true });
-      await writeFile10(filePath, file.content, "utf8");
+      const filePath = path25.join(fixtureDir, file.relativePath);
+      await mkdir11(path25.dirname(filePath), { recursive: true });
+      await writeFile11(filePath, file.content, "utf8");
     }
     const { result: ingestionLog, durationMs } = await timed(
       async () => options.ingestionAdapter.ingest(await realpath6(fixtureDir))
@@ -25885,15 +26188,15 @@ async function runIngestionSchemaCompletenessBenchmark(options) {
       }
     };
   } finally {
-    await rm7(fixtureDir, { recursive: true, force: true });
+    await rm8(fixtureDir, { recursive: true, force: true });
   }
 }
 
 // src/benchmarks/remnic/ingestion-backlink-f1/runner.ts
 import { randomUUID as randomUUID24 } from "crypto";
-import { mkdtemp as mkdtemp8, writeFile as writeFile11, rm as rm8, mkdir as mkdir11, realpath as realpath7 } from "fs/promises";
+import { mkdtemp as mkdtemp8, writeFile as writeFile12, rm as rm9, mkdir as mkdir12, realpath as realpath7 } from "fs/promises";
 import { tmpdir as tmpdir4 } from "os";
-import path25 from "path";
+import path26 from "path";
 var ingestionBacklinkF1Definition = {
   id: "ingestion-backlink-f1",
   title: "Ingestion: Backlink F1",
@@ -25912,13 +26215,13 @@ async function runIngestionBacklinkF1Benchmark(options) {
     throw new Error("ingestionAdapter is required for ingestion benchmarks");
   }
   const fixture = emailFixture.generate();
-  const fixtureDir = await mkdtemp8(path25.join(tmpdir4(), "bench-email-"));
+  const fixtureDir = await mkdtemp8(path26.join(tmpdir4(), "bench-email-"));
   try {
     await options.ingestionAdapter.reset();
     for (const file of fixture.files) {
-      const filePath = path25.join(fixtureDir, file.relativePath);
-      await mkdir11(path25.dirname(filePath), { recursive: true });
-      await writeFile11(filePath, file.content, "utf8");
+      const filePath = path26.join(fixtureDir, file.relativePath);
+      await mkdir12(path26.dirname(filePath), { recursive: true });
+      await writeFile12(filePath, file.content, "utf8");
     }
     const { result: ingestionLog, durationMs } = await timed(
       async () => options.ingestionAdapter.ingest(await realpath7(fixtureDir))
@@ -25986,15 +26289,15 @@ async function runIngestionBacklinkF1Benchmark(options) {
       }
     };
   } finally {
-    await rm8(fixtureDir, { recursive: true, force: true });
+    await rm9(fixtureDir, { recursive: true, force: true });
   }
 }
 
 // src/benchmarks/remnic/ingestion-setup-friction/runner.ts
 import { randomUUID as randomUUID25 } from "crypto";
-import { mkdtemp as mkdtemp9, writeFile as writeFile12, rm as rm9, mkdir as mkdir12, realpath as realpath8 } from "fs/promises";
+import { mkdtemp as mkdtemp9, writeFile as writeFile13, rm as rm10, mkdir as mkdir13, realpath as realpath8 } from "fs/promises";
 import { tmpdir as tmpdir5 } from "os";
-import path26 from "path";
+import path27 from "path";
 var INGESTION_SETUP_FRICTION_LOWER_IS_BETTER = /* @__PURE__ */ new Set(["setup_friction", "commands_count", "prompts_count", "errors_count"]);
 var ingestionSetupFrictionDefinition = {
   id: "ingestion-setup-friction",
@@ -26014,13 +26317,13 @@ async function runIngestionSetupFrictionBenchmark(options) {
     throw new Error("ingestionAdapter is required for ingestion benchmarks");
   }
   const fixture = emailFixture.generate();
-  const fixtureDir = await mkdtemp9(path26.join(tmpdir5(), "bench-friction-"));
+  const fixtureDir = await mkdtemp9(path27.join(tmpdir5(), "bench-friction-"));
   try {
     await options.ingestionAdapter.reset();
     for (const file of fixture.files) {
-      const filePath = path26.join(fixtureDir, file.relativePath);
-      await mkdir12(path26.dirname(filePath), { recursive: true });
-      await writeFile12(filePath, file.content, "utf8");
+      const filePath = path27.join(fixtureDir, file.relativePath);
+      await mkdir13(path27.dirname(filePath), { recursive: true });
+      await writeFile13(filePath, file.content, "utf8");
     }
     const { result: ingestionLog, durationMs } = await timed(
       async () => options.ingestionAdapter.ingest(await realpath8(fixtureDir))
@@ -26092,15 +26395,15 @@ async function runIngestionSetupFrictionBenchmark(options) {
       }
     };
   } finally {
-    await rm9(fixtureDir, { recursive: true, force: true });
+    await rm10(fixtureDir, { recursive: true, force: true });
   }
 }
 
 // src/benchmarks/remnic/ingestion-citation-accuracy/runner.ts
 import { randomUUID as randomUUID26 } from "crypto";
-import { mkdtemp as mkdtemp10, writeFile as writeFile13, rm as rm10, mkdir as mkdir13, realpath as realpath9 } from "fs/promises";
+import { mkdtemp as mkdtemp10, writeFile as writeFile14, rm as rm11, mkdir as mkdir14, realpath as realpath9 } from "fs/promises";
 import { tmpdir as tmpdir6 } from "os";
-import path27 from "path";
+import path28 from "path";
 var CITATION_SUPPORT_THRESHOLD = 0.72;
 var ingestionCitationAccuracyDefinition = {
   id: "ingestion-citation-accuracy",
@@ -26159,10 +26462,10 @@ function resolveCitedSources(sourceRefs, seeAlso, pageRef, sourceContentMap) {
     return "";
   }
   for (const ref of normalizedRefs) {
-    const refBase = path27.basename(ref).toLowerCase();
+    const refBase = path28.basename(ref).toLowerCase();
     let matched = false;
     for (const [relativePath, content] of sourceContentMap) {
-      if (relativePath === ref || relativePath.endsWith(ref) || path27.basename(relativePath).toLowerCase() === refBase) {
+      if (relativePath === ref || relativePath.endsWith(ref) || path28.basename(relativePath).toLowerCase() === refBase) {
         resolved.push(content);
         matched = true;
         break;
@@ -26178,9 +26481,9 @@ function resolveCitedSources(sourceRefs, seeAlso, pageRef, sourceContentMap) {
   if (normalizedRefs.length > 0) {
     return "";
   }
-  const pageBase = path27.basename(pageRef).toLowerCase();
+  const pageBase = path28.basename(pageRef).toLowerCase();
   for (const [relativePath, content] of sourceContentMap) {
-    if (path27.basename(relativePath).toLowerCase() === pageBase) {
+    if (path28.basename(relativePath).toLowerCase() === pageBase) {
       return content;
     }
   }
@@ -26191,13 +26494,13 @@ async function runIngestionCitationAccuracyBenchmark(options) {
     throw new Error("ingestionAdapter is required for ingestion benchmarks");
   }
   const fixture = emailFixture.generate();
-  const fixtureDir = await mkdtemp10(path27.join(tmpdir6(), "bench-citation-"));
+  const fixtureDir = await mkdtemp10(path28.join(tmpdir6(), "bench-citation-"));
   try {
     await options.ingestionAdapter.reset();
     for (const file of fixture.files) {
-      const filePath = path27.join(fixtureDir, file.relativePath);
-      await mkdir13(path27.dirname(filePath), { recursive: true });
-      await writeFile13(filePath, file.content, "utf8");
+      const filePath = path28.join(fixtureDir, file.relativePath);
+      await mkdir14(path28.dirname(filePath), { recursive: true });
+      await writeFile14(filePath, file.content, "utf8");
     }
     const benchmarkStart = performance.now();
     const { result: ingestionLog, durationMs: ingestionDurationMs } = await timed(
@@ -26394,7 +26697,7 @@ async function runIngestionCitationAccuracyBenchmark(options) {
       }
     };
   } finally {
-    await rm10(fixtureDir, { recursive: true, force: true });
+    await rm11(fixtureDir, { recursive: true, force: true });
   }
 }
 function citationSupportScore(claim, citedSources) {
@@ -26582,7 +26885,7 @@ var ASSISTANT_MORNING_BRIEF_SMOKE_SCENARIOS = ASSISTANT_MORNING_BRIEF_SCENARIOS.
 
 // src/benchmarks/remnic/_assistant-common/runner.ts
 import { randomUUID as randomUUID27 } from "crypto";
-import path29 from "path";
+import path30 from "path";
 
 // src/run-seeds.ts
 function buildBenchmarkRunSeeds(runCount, baseSeed) {
@@ -26672,9 +26975,9 @@ function pairedDeltaConfidenceInterval(candidateValues, baselineValues, options
 }
 
 // src/judges/sealed-rubric.ts
-import { createHash as createHash7 } from "crypto";
+import { createHash as createHash8 } from "crypto";
 import { appendFileSync, mkdirSync } from "fs";
-import path28 from "path";
+import path29 from "path";
 
 // src/judges/sealed-prompts/assistant-rubric-v1.ts
 var ASSISTANT_RUBRIC_V1 = `# Assistant rubric v1 (sealed)
@@ -26781,7 +27084,7 @@ function loadSealedRubric(id = DEFAULT_ASSISTANT_RUBRIC_ID, options = {}) {
   if (typeof prompt !== "string" || prompt.length === 0) {
     throw new Error(`sealed rubric not found in registry: ${id}`);
   }
-  const sha256 = createHash7("sha256").update(prompt, "utf8").digest("hex");
+  const sha256 = createHash8("sha256").update(prompt, "utf8").digest("hex");
   const version = parseVersionFromId(id);
   return { id, version, prompt, sha256 };
 }
@@ -26954,7 +27257,7 @@ function createSpotCheckFileLogger(options) {
     return { log() {
     } };
   }
-  const logPath = path28.join(directory, `${runId}.jsonl`);
+  const logPath = path29.join(directory, `${runId}.jsonl`);
   let written = 0;
   let warnedOnWriteFailure = false;
   const cap = typeof sampleSize === "number" && sampleSize > 0 ? sampleSize : 5;
@@ -27043,7 +27346,7 @@ async function runAssistantBenchmark(definition, scenarios, resolved, runnerOpti
   const runId = buildRunId(definition.id);
   const spotCheckLogger = createSpotCheckFileLogger({
     runId,
-    directory: runnerOptions.spotCheckDir ?? path29.join(process.cwd(), "benchmarks", "results", "spot-checks"),
+    directory: runnerOptions.spotCheckDir ?? path30.join(process.cwd(), "benchmarks", "results", "spot-checks"),
     sampleRate: 0.35,
     sampleSize: 5
   });
@@ -27692,9 +27995,9 @@ async function runAssistantSynthesisBenchmark(options) {
 
 // src/benchmarks/remnic/buffer-surprise-trigger/runner.ts
 import { randomUUID as randomUUID28 } from "crypto";
-import path30 from "path";
+import path31 from "path";
 import os8 from "os";
-import { mkdir as mkdir14, rm as rm11 } from "fs/promises";
+import { mkdir as mkdir15, rm as rm12 } from "fs/promises";
 import {
   SmartBuffer,
   computeSurprise,
@@ -27923,11 +28226,11 @@ function hasExplicitTopicPivotCue(text) {
 }
 async function runBufferSurpriseTriggerBenchmark(options) {
   const cases = loadCases10(options.mode, options.limit);
-  const tmpRoot = path30.join(
+  const tmpRoot = path31.join(
     os8.tmpdir(),
     `remnic-bench-buffer-surprise-${randomUUID28()}`
   );
-  await mkdir14(tmpRoot, { recursive: true });
+  await mkdir15(tmpRoot, { recursive: true });
   const tasks = [];
   const startedAt = performance.now();
   try {
@@ -27945,7 +28248,7 @@ async function runBufferSurpriseTriggerBenchmark(options) {
       tasks.push(buildTaskResult(caseDef, control, candidate));
     }
   } finally {
-    await rm11(tmpRoot, { recursive: true, force: true });
+    await rm12(tmpRoot, { recursive: true, force: true });
   }
   const totalLatencyMs = Math.round(performance.now() - startedAt);
   const aggregates = buildAggregates2(tasks);
@@ -27992,12 +28295,12 @@ async function runBufferSurpriseTriggerBenchmark(options) {
   };
 }
 async function runSingleCase(caseDef, options) {
-  const memoryDir = path30.join(
+  const memoryDir = path31.join(
     options.tmpRoot,
     `${caseDef.id}-${options.label}`
   );
-  const workspaceDir = path30.join(memoryDir, "workspace");
-  await mkdir14(workspaceDir, { recursive: true });
+  const workspaceDir = path31.join(memoryDir, "workspace");
+  await mkdir15(workspaceDir, { recursive: true });
   const config = parseConfig4({
     memoryDir,
     workspaceDir,
@@ -29151,8 +29454,8 @@ function finalizeBenchmarkResultConfig(result, options) {
 }
 
 // src/benchmark.ts
-var DEFAULT_BASELINE_PATH = path31.join(process.cwd(), "benchmarks", "baseline.json");
-var DEFAULT_REPORT_PATH = path31.join(process.cwd(), "benchmarks", "report.json");
+var DEFAULT_BASELINE_PATH = path32.join(process.cwd(), "benchmarks", "baseline.json");
+var DEFAULT_REPORT_PATH = path32.join(process.cwd(), "benchmarks", "report.json");
 var BASELINE_VERSION = 1;
 var DEFAULT_TOLERANCE = 10;
 var DEFAULT_FULL_RUN_COUNT = 5;
@@ -29214,7 +29517,33 @@ async function runBenchmark(benchmarkId, options) {
   const log = (message) => {
     console.error(`  ${message}`);
   };
-  const system = !shouldGuardSystem ? options.system : createTimeoutGuardedAdapter(options.system, {
+  const originalSystemJudge = options.system.judge;
+  let systemJudgeMutatedInPlace = false;
+  let judgeCacheCounters;
+  let cachedCrossJudge;
+  let crossJudgeCacheCounters;
+  let primaryDrainPendingWrites;
+  let crossDrainPendingWrites;
+  const cacheWiring = (() => {
+    if (options.noJudgeCache) {
+      return void 0;
+    }
+    const willWrapPrimary = options.system.judge !== void 0 && (options.judgeProvider ?? null) !== null;
+    const willWrapCross = options.amaBenchCrossJudge !== void 0 && (options.amaBenchCrossJudgeProvider ?? null) !== null;
+    if (!willWrapPrimary && !willWrapCross) {
+      return void 0;
+    }
+    const cacheDir = options.judgeCacheDir ? path32.resolve(expandTildePath3(options.judgeCacheDir)) : options.outputDir ? path32.join(path32.resolve(expandTildePath3(options.outputDir)), "judge-cache") : void 0;
+    if (cacheDir === void 0) {
+      return void 0;
+    }
+    return {
+      cache: new JudgeCache({ dir: cacheDir }),
+      willWrapPrimary,
+      willWrapCross
+    };
+  })();
+  let system = !shouldGuardSystem ? options.system : createTimeoutGuardedAdapter(options.system, {
     benchmarkId,
     ...timeoutMs !== void 0 ? { timeoutMs } : {},
     ...options.drainTimeoutMs !== void 0 ? { drainTimeoutMs: options.drainTimeoutMs } : {},
@@ -29245,18 +29574,118 @@ async function runBenchmark(benchmarkId, options) {
   }) : rawIngestionAdapter;
   let result;
   try {
+    if (cacheWiring?.willWrapPrimary && system.judge !== void 0) {
+      const primary = wrapJudgeWithCache({
+        role: "primary",
+        judge: system.judge,
+        benchmarkId,
+        datasetVersion: definition.meta.version,
+        amaBenchJudgeProtocol: options.amaBenchJudgeProtocol ?? "default",
+        provider: options.judgeProvider ?? null,
+        cache: cacheWiring.cache
+      });
+      judgeCacheCounters = primary.counters;
+      primaryDrainPendingWrites = primary.drainPendingWrites;
+      try {
+        system.judge = primary.judge;
+        systemJudgeMutatedInPlace = system === options.system;
+      } catch {
+        system = createJudgeOverrideProxy(system, primary.judge);
+      }
+    }
+    if (cacheWiring?.willWrapCross) {
+      const wrapped = wrapJudgeWithCache({
+        role: "cross",
+        judge: options.amaBenchCrossJudge,
+        benchmarkId,
+        datasetVersion: definition.meta.version,
+        amaBenchJudgeProtocol: options.amaBenchJudgeProtocol ?? "default",
+        provider: options.amaBenchCrossJudgeProvider ?? null,
+        cache: cacheWiring.cache
+      });
+      cachedCrossJudge = wrapped.judge;
+      crossJudgeCacheCounters = wrapped.counters;
+      crossDrainPendingWrites = wrapped.drainPendingWrites;
+    }
     result = await registeredBenchmark.run({
       ...options,
       system,
+      // PR #1591 P2 (thread #10): when caching is on AND a cross judge is
+      // configured, hand the cached cross judge to the runner so AMA-Bench
+      // cross-judge calls participate in the same content-keyed cache as
+      // the primary system judge. Without this override, the runner kept
+      // calling the unwrapped cross judge on every iteration.
+      ...cachedCrossJudge ? { amaBenchCrossJudge: cachedCrossJudge } : {},
       ...ingestionAdapter ? { ingestionAdapter } : {},
       mode: options.mode ?? "quick",
       benchmark: definition
     });
   } finally {
-    await destroyOwnedIngestionAdapter();
+    try {
+      await destroyOwnedIngestionAdapter();
+    } finally {
+      if (systemJudgeMutatedInPlace) {
+        options.system.judge = originalSystemJudge;
+      }
+    }
+    if (primaryDrainPendingWrites) {
+      await primaryDrainPendingWrites();
+    }
+    if (crossDrainPendingWrites) {
+      await crossDrainPendingWrites();
+    }
+  }
+  const primaryCalls = judgeCacheCounters?.modelCalls ?? 0;
+  const crossCalls = crossJudgeCacheCounters?.modelCalls ?? 0;
+  if (judgeCacheCounters !== void 0 || crossJudgeCacheCounters !== void 0) {
+    result.cost.judgeModelCalls = primaryCalls + crossCalls;
   }
   return finalizeBenchmarkResultConfig(result, options);
 }
+function wrapJudgeWithCache(args) {
+  const crossJudgeIdSuffix = args.role === "cross" ? "-cross" : "";
+  const wrapped = runJudgeWithCache({
+    judge: args.judge,
+    cache: args.cache,
+    keyExtras: {
+      benchmarkId: `${args.benchmarkId}${crossJudgeIdSuffix}`,
+      datasetVersion: args.datasetVersion,
+      // Protocol identity: bench judge protocol version + the selected
+      // judge protocol variant, suffixed by role so primary vs cross
+      // differentiator is part of the prompt hash. Bumping
+      // JUDGE_CACHE_PROTOCOL_VERSION invalidates verdicts when judge
+      // prompt/parse semantics change (PR #1591, High).
+      judgePromptHash: createHash9("sha256").update(JUDGE_CACHE_PROTOCOL_VERSION).update("
").update(args.amaBenchJudgeProtocol).update("
").update(args.role).digest("hex"),
+      judgeModelId: args.provider?.model !== void 0 && args.provider.model.length > 0 ? `${args.provider.model}${crossJudgeIdSuffix}` : `unknown-${args.role}-judge`,
+      // Full judge configuration, deterministically serialized (sorted
+      // keys) so provider/base-url/retry changes produce fresh cache
+      // keys. `role` is included so primary and cross judges never
+      // share a paramsHash.
+      judgeParamsHash: createHash9("sha256").update(
+        stableStringify2({
+          role: args.role,
+          provider: args.provider
+        })
+      ).digest("hex")
+    }
+  });
+  return {
+    judge: wrapped,
+    counters: wrapped.counters,
+    drainPendingWrites: wrapped.drainPendingWrites
+  };
+}
+function createJudgeOverrideProxy(adapter, judge) {
+  return new Proxy(adapter, {
+    get(target, prop) {
+      if (prop === "judge") {
+        return judge;
+      }
+      const value = Reflect.get(target, prop, target);
+      return typeof value === "function" ? value.bind(target) : value;
+    }
+  });
+}
 function benchmarkDefinition(id) {
   const definition = getBenchmark(id);
   if (!definition) {
@@ -29293,7 +29722,7 @@ function loadBaseline(baselinePath) {
   return raw;
 }
 function saveBaseline(baselinePath, baseline) {
-  fs2.mkdirSync(path31.dirname(baselinePath), { recursive: true });
+  fs2.mkdirSync(path32.dirname(baselinePath), { recursive: true });
   fs2.writeFileSync(baselinePath, `${JSON.stringify(baseline, null, 2)}
 `);
 }
@@ -29523,7 +29952,7 @@ function generateReport(results, reportPath) {
     totalDurationMs: results.reduce((sum, result) => sum + result.totalDurationMs, 0)
   };
   if (reportPath) {
-    fs2.mkdirSync(path31.dirname(reportPath), { recursive: true });
+    fs2.mkdirSync(path32.dirname(reportPath), { recursive: true });
     fs2.writeFileSync(reportPath, `${JSON.stringify(report, null, 2)}
 `);
   }
@@ -29670,7 +30099,7 @@ function getBenchmarkLowerIsBetter(benchmarkId) {
 }
 
 // src/integrity/sealed-qrels.ts
-import { readFile as readFile17 } from "fs/promises";
+import { readFile as readFile18 } from "fs/promises";
 function isSealedQrelsArtifact(value) {
   if (!value || typeof value !== "object") {
     return false;
@@ -29740,7 +30169,7 @@ function parseSealedQrels(raw, options = {}) {
   };
 }
 async function loadSealedQrels(filePath, options = {}) {
-  const raw = await readFile17(filePath, "utf8");
+  const raw = await readFile18(filePath, "utf8");
   return parseSealedQrels(raw, options);
 }
 function serializeSealedQrels(artifact) {
@@ -29860,7 +30289,7 @@ function selectFixtureVariant(variants, seed) {
 }
 
 // src/benchmarks/custom/loader.ts
-import { readFile as readFile18 } from "fs/promises";
+import { readFile as readFile19 } from "fs/promises";
 import { parse as parseYaml } from "yaml";
 var CUSTOM_SCORING_VALUES = /* @__PURE__ */ new Set([
   "exact_match",
@@ -29880,7 +30309,7 @@ function parseCustomBenchmark(source) {
 async function loadCustomBenchmarkFile(filePath) {
   let source;
   try {
-    source = await readFile18(filePath, "utf8");
+    source = await readFile19(filePath, "utf8");
   } catch (error) {
     throw new Error(
       `Failed to read custom benchmark file ${filePath}: ${formatError(error)}`
@@ -29988,15 +30417,60 @@ function formatError(error) {
 
 // src/benchmarks/custom/runner.ts
 import { randomUUID as randomUUID31 } from "crypto";
-import path32 from "path";
+import path33 from "path";
+import { expandTildePath as expandTildePath4 } from "@remnic/core";
 async function runCustomBenchmarkFile(filePath, options) {
   const spec = await loadCustomBenchmarkFile(filePath);
   const benchmark = createCustomBenchmarkDefinition(spec, filePath);
-  return runCustomBenchmark(spec, {
+  const runOptions = {
     ...options,
     mode: options.mode ?? "quick",
     benchmark
-  });
+  };
+  let cacheRestore;
+  let cacheCounters;
+  if (spec.scoring === "llm_judge" && runOptions.system.judge !== void 0 && !runOptions.noJudgeCache && (runOptions.judgeProvider ?? null) !== null) {
+    const cacheDir = runOptions.judgeCacheDir ? path33.resolve(expandTildePath4(runOptions.judgeCacheDir)) : runOptions.outputDir ? path33.join(path33.resolve(expandTildePath4(runOptions.outputDir)), "judge-cache") : void 0;
+    if (cacheDir !== void 0) {
+      const originalJudge = runOptions.system.judge;
+      const wrapped = wrapJudgeWithCache({
+        role: "primary",
+        judge: originalJudge,
+        benchmarkId: benchmark.id,
+        datasetVersion: benchmark.meta.version,
+        amaBenchJudgeProtocol: runOptions.amaBenchJudgeProtocol ?? "default",
+        provider: runOptions.judgeProvider ?? null,
+        cache: new JudgeCache({ dir: cacheDir })
+      });
+      let systemJudgeMutatedInPlace = false;
+      try {
+        runOptions.system.judge = wrapped.judge;
+        systemJudgeMutatedInPlace = true;
+      } catch {
+        runOptions.system = createJudgeOverrideProxy(runOptions.system, wrapped.judge);
+      }
+      cacheCounters = wrapped.counters;
+      const needRestore = systemJudgeMutatedInPlace;
+      cacheRestore = async () => {
+        if (needRestore) {
+          runOptions.system.judge = originalJudge;
+        }
+        await wrapped.drainPendingWrites();
+      };
+    }
+  }
+  let result;
+  try {
+    result = await runCustomBenchmark(spec, runOptions);
+  } finally {
+    if (cacheRestore) {
+      await cacheRestore();
+    }
+  }
+  if (cacheCounters) {
+    result.cost.judgeModelCalls = cacheCounters.modelCalls;
+  }
+  return result;
 }
 async function runCustomBenchmark(spec, options) {
   if (spec.scoring === "llm_judge" && !options.system.judge) {
@@ -30157,7 +30631,7 @@ async function scoreTask(scoring, options, question, actual, expected) {
   }
 }
 function createCustomBenchmarkDefinition(benchmark, filePath) {
-  const id = `custom:${slugify(path32.basename(filePath, path32.extname(filePath)) || benchmark.name)}`;
+  const id = `custom:${slugify(path33.basename(filePath, path33.extname(filePath)) || benchmark.name)}`;
   return {
     id,
     title: benchmark.name,
@@ -31027,9 +31501,9 @@ var chatFixture = {
 };
 
 // src/benchmarks/remnic/procedural-recall/ablation.ts
-import { mkdir as mkdir15, mkdtemp as mkdtemp11, rm as rm12, writeFile as writeFile14, readFile as readFile19 } from "fs/promises";
+import { mkdir as mkdir16, mkdtemp as mkdtemp11, rm as rm13, writeFile as writeFile15, readFile as readFile20 } from "fs/promises";
 import os9 from "os";
-import path33 from "path";
+import path34 from "path";
 import {
   StorageManager as StorageManager3,
   parseConfig as parseConfig5,
@@ -31060,7 +31534,7 @@ async function runSide(scenarios, proceduralEnabled) {
   const observed = [];
   for (const scenario of scenarios) {
     const dir = await mkdtemp11(
-      path33.join(os9.tmpdir(), "remnic-bench-proc-ablation-")
+      path34.join(os9.tmpdir(), "remnic-bench-proc-ablation-")
     );
     try {
       const storage = new StorageManager3(dir);
@@ -31075,7 +31549,7 @@ ${body}`,
       );
       const config = parseConfig5({
         memoryDir: dir,
-        workspaceDir: path33.join(dir, "ws"),
+        workspaceDir: path34.join(dir, "ws"),
         openaiApiKey: "bench-key",
         procedural: {
           enabled: proceduralEnabled,
@@ -31089,7 +31563,7 @@ ${body}`,
       );
       observed.push(section !== null && section.length > 0);
     } finally {
-      await rm12(dir, { recursive: true, force: true });
+      await rm13(dir, { recursive: true, force: true });
     }
   }
   return observed;
@@ -31150,7 +31624,7 @@ async function runProceduralAblation(options) {
   };
 }
 async function loadAblationFixture(fixturePath) {
-  const raw = await readFile19(fixturePath, "utf8");
+  const raw = await readFile20(fixturePath, "utf8");
   let parsed;
   try {
     parsed = JSON.parse(raw);
@@ -31246,9 +31720,9 @@ async function runProceduralAblationCli(args) {
     random: args.random,
     seed: args.seed
   });
-  const outDir = path33.dirname(path33.resolve(args.outPath));
-  await mkdir15(outDir, { recursive: true });
-  await writeFile14(args.outPath, JSON.stringify(artifact, null, 2) + "\n", "utf8");
+  const outDir = path34.dirname(path34.resolve(args.outPath));
+  await mkdir16(outDir, { recursive: true });
+  await writeFile15(args.outPath, JSON.stringify(artifact, null, 2) + "\n", "utf8");
   return artifact;
 }