@m4trix/evals 0.25.0 → 0.25.1

package/dist/index.cjs

@@ -184,9 +184,7 @@ var data_mock_default = {
             { name: "contract_match", score: 100 },
             { name: "arg_validity", score: 100 }
           ],
-          checks: [
-            { name: "tool_calls", passed: true, detail: "0 unexpected" }
-          ],
+          checks: [{ name: "tool_calls", passed: true, detail: "0 unexpected" }],
           failures: [],
           meta: {
             model: "gpt-4o-mini",
@@ -209,9 +207,21 @@ var data_mock_default = {
     }
   ],
   evaluators: [
-    { id: "json-schema-validator", name: "JSON Schema Validator", configPreview: "strict=true" },
-    { id: "tool-call-contract-checker", name: "Tool-call Contract Checker", configPreview: "unexpectedCalls=error" },
-    { id: "rubric-judge", name: "Rubric Judge (LLM)", configPreview: "model=gpt-4o-mini; scale=0-100" },
+    {
+      id: "json-schema-validator",
+      name: "JSON Schema Validator",
+      configPreview: "strict=true"
+    },
+    {
+      id: "tool-call-contract-checker",
+      name: "Tool-call Contract Checker",
+      configPreview: "unexpectedCalls=error"
+    },
+    {
+      id: "rubric-judge",
+      name: "Rubric Judge (LLM)",
+      configPreview: "model=gpt-4o-mini; scale=0-100"
+    },
     { id: "pii-leak-detector", name: "PII Leak Detector", configPreview: "redact=false" }
   ]
 };
@@ -291,9 +301,7 @@ async function loadRunnerData(runner) {
   const memSnapshots = runner.getAllRunSnapshots();
   const seen = new Set(memSnapshots.map((s) => s.runId));
   const fromDisk = diskSnapshots.filter((s) => !seen.has(s.runId));
-  const snapshots = [...memSnapshots, ...fromDisk].sort(
-    (a, b) => b.queuedAt - a.queuedAt
-  );
+  const snapshots = [...memSnapshots, ...fromDisk].sort((a, b) => b.queuedAt - a.queuedAt);
   if (datasets.length === 0 && evaluators.length === 0) {
     return loadMockData();
   }
@@ -447,9 +455,7 @@ var Evaluator = class _Evaluator {
     return this._config.passCriterion;
   }
   async resolveContext() {
-    const parts = await Promise.all(
-      this._config.middlewares.map((mw) => mw.resolve())
-    );
+    const parts = await Promise.all(this._config.middlewares.map((mw) => mw.resolve()));
     return Object.assign({}, ...parts);
   }
 };
@@ -553,10 +559,7 @@ var ScoreAggregate = {
       const count = values.length || 1;
       const result = {};
       for (const field of fields) {
-        result[field] = values.reduce(
-          (s, v) => s + (v[field] ?? 0),
-          0
-        ) / count;
+        result[field] = values.reduce((s, v) => s + (v[field] ?? 0), 0) / count;
       }
       return result;
     };
@@ -590,13 +593,10 @@ var ScoreAggregate = {
           (s, v) => s + (v[valueField] ?? 0),
           0
         );
-        const sumSq = values.reduce(
-          (s, v) => {
-            const value = v[valueField] ?? 0;
-            return s + value * value;
-          },
-          0
-        );
+        const sumSq = values.reduce((s, v) => {
+          const value = v[valueField] ?? 0;
+          return s + value * value;
+        }, 0);
         const mean = sum / count;
         const variance = (sumSq - count * mean * mean) / (count - 1);
         stdDev = variance > 0 ? Math.sqrt(variance) : 0;
@@ -791,16 +791,8 @@ function createDiffString(expected, actual, diffOptions) {
   const expectedProcessed = preprocessForDiff(expected, diffOptions);
   const actualProcessed = preprocessForDiff(actual, diffOptions);
   if (diffOptions?.keysOnly) {
-    const expectedKeys = JSON.stringify(
-      extractKeys(expectedProcessed),
-      null,
-      2
-    );
-    const actualKeys = JSON.stringify(
-      extractKeys(actualProcessed),
-      null,
-      2
-    );
+    const expectedKeys = JSON.stringify(extractKeys(expectedProcessed), null, 2);
+    const actualKeys = JSON.stringify(extractKeys(actualProcessed), null, 2);
     const parts2 = diff.diffLines(expectedKeys, actualKeys);
     return formatDiffParts(parts2);
   }
@@ -811,9 +803,7 @@ function createDiffString(expected, actual, diffOptions) {
   }
   const parts = diff.diffLines(expectedStr, actualStr);
   if (diffOptions?.outputNewOnly) {
-    const filtered = parts.filter(
-      (p) => p.added === true
-    );
+    const filtered = parts.filter((p) => p.added === true);
     return formatDiffParts(filtered);
   }
   return formatDiffParts(parts);
@@ -896,18 +886,8 @@ var defaultRunnerConfig = {
   discovery: {
     rootDir: process.cwd(),
     datasetSuffixes: [".dataset.ts", ".dataset.tsx", ".dataset.js", ".dataset.mjs"],
-    evaluatorSuffixes: [
-      ".evaluator.ts",
-      ".evaluator.tsx",
-      ".evaluator.js",
-      ".evaluator.mjs"
-    ],
-    testCaseSuffixes: [
-      ".test-case.ts",
-      ".test-case.tsx",
-      ".test-case.js",
-      ".test-case.mjs"
-    ],
+    evaluatorSuffixes: [".evaluator.ts", ".evaluator.tsx", ".evaluator.js", ".evaluator.mjs"],
+    testCaseSuffixes: [".test-case.ts", ".test-case.tsx", ".test-case.js", ".test-case.mjs"],
     excludeDirectories: ["node_modules", "dist", ".next", ".git", ".pnpm-store"]
   },
   artifactDirectory: ".eval-results",
@@ -974,14 +954,15 @@ function getJitiLoader() {
   }
   const createJiti2 = jitiModule__namespace.createJiti ?? jitiModule__namespace.default;
   if (typeof createJiti2 !== "function") {
-    throw new Error(
-      "Failed to initialize jiti for m4trix eval config loading."
-    );
+    throw new Error("Failed to initialize jiti for m4trix eval config loading.");
   }
-  cachedLoader = createJiti2((typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('out.js', document.baseURI).href)), {
-    interopDefault: true,
-    moduleCache: true
-  });
+  cachedLoader = createJiti2(
+    (typeof document === 'undefined' ? require('u' + 'rl').pathToFileURL(__filename).href : (_documentCurrentScript && _documentCurrentScript.tagName.toUpperCase() === 'SCRIPT' && _documentCurrentScript.src || new URL('out.js', document.baseURI).href)),
+    {
+      interopDefault: true,
+      moduleCache: true
+    }
+  );
   return cachedLoader;
 }
 function resolveConfigModuleExport(loadedModule) {
@@ -1085,9 +1066,7 @@ async function loadModuleExports(filePath) {
 }
 async function collectDatasetsFromFiles(config) {
   const files = await walkDirectory(config.rootDir, config.excludeDirectories);
-  const matched = files.filter(
-    (filePath) => hasOneSuffix(filePath, config.datasetSuffixes)
-  );
+  const matched = files.filter((filePath) => hasOneSuffix(filePath, config.datasetSuffixes));
   const found = await Promise.all(
     matched.map(async (absolutePath) => {
       const exports = await loadModuleExports(absolutePath);
@@ -1104,9 +1083,7 @@ async function collectDatasetsFromFiles(config) {
 }
 async function collectEvaluatorsFromFiles(config) {
   const files = await walkDirectory(config.rootDir, config.excludeDirectories);
-  const matched = files.filter(
-    (filePath) => hasOneSuffix(filePath, config.evaluatorSuffixes)
-  );
+  const matched = files.filter((filePath) => hasOneSuffix(filePath, config.evaluatorSuffixes));
   const found = await Promise.all(
     matched.map(async (absolutePath) => {
       const exports = await loadModuleExports(absolutePath);
@@ -1123,9 +1100,7 @@ async function collectEvaluatorsFromFiles(config) {
 }
 async function collectTestCasesFromFiles(config) {
   const files = await walkDirectory(config.rootDir, config.excludeDirectories);
-  const matched = files.filter(
-    (filePath) => hasOneSuffix(filePath, config.testCaseSuffixes)
-  );
+  const matched = files.filter((filePath) => hasOneSuffix(filePath, config.testCaseSuffixes));
   const found = await Promise.all(
     matched.map(async (absolutePath) => {
       const exports = await loadModuleExports(absolutePath);
@@ -1233,20 +1208,14 @@ function nowIsoForFile() {
   return (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-");
 }
 function createArtifactPath(artifactDirectory, datasetId, runId) {
-  return path.join(
-    artifactDirectory,
-    `${datasetId}_${runId}_${nowIsoForFile()}.jsonl`
-  );
+  return path.join(artifactDirectory, `${datasetId}_${runId}_${nowIsoForFile()}.jsonl`);
 }
 function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persistenceQueue, updateSnapshot, startedRef, completedRef, passedRef, failedRef, testCaseResultsRef) {
   const { testCaseItem, rerunIndex, rerunTotal } = unit;
   return effect.Effect.gen(function* () {
     const evaluatorRunId = `run-${crypto.randomUUID()}`;
     const started = Date.now();
-    const startedEvaluations = yield* effect.Ref.modify(startedRef, (n) => [
-      n + 1,
-      n + 1
-    ]);
+    const startedEvaluations = yield* effect.Ref.modify(startedRef, (n) => [n + 1, n + 1]);
     yield* publishEvent({
       type: "TestCaseStarted",
       runId: task.runId,
@@ -1279,9 +1248,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
         return error;
       };
       try {
-        const ctx = yield* effect.Effect.promise(
-          () => Promise.resolve(evaluator.resolveContext())
-        );
+        const ctx = yield* effect.Effect.promise(() => Promise.resolve(evaluator.resolveContext()));
         const result = yield* effect.Effect.promise(
           () => Promise.resolve().then(
             () => evaluateFn({
@@ -1336,10 +1303,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
       }
     }
     const rerunPassedThis = evaluatorScores.every((s) => s.passed);
-    const completedEvaluations = yield* effect.Ref.modify(completedRef, (n) => [
-      n + 1,
-      n + 1
-    ]);
+    const completedEvaluations = yield* effect.Ref.modify(completedRef, (n) => [n + 1, n + 1]);
     const progressEvent = {
       type: "TestCaseProgress",
       runId: task.runId,
@@ -1388,10 +1352,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
       } else {
         yield* effect.Ref.update(failedRef, (n) => n + 1);
       }
-      const [passed, failed] = yield* effect.Effect.all([
-        effect.Ref.get(passedRef),
-        effect.Ref.get(failedRef)
-      ]);
+      const [passed, failed] = yield* effect.Effect.all([effect.Ref.get(passedRef), effect.Ref.get(failedRef)]);
       yield* updateSnapshot(task.runId, (snapshot) => ({
         ...snapshot,
         passedTestCases: passed,
@@ -1711,15 +1672,11 @@ var EffectRunner = class {
     this.persistenceQueue = effect.Effect.runSync(
       effect.Queue.unbounded()
     );
-    this.snapshotsRef = effect.Effect.runSync(
-      effect.Ref.make(/* @__PURE__ */ new Map())
-    );
+    this.snapshotsRef = effect.Effect.runSync(effect.Ref.make(/* @__PURE__ */ new Map()));
     this.listeners = /* @__PURE__ */ new Set();
     this.datasetsById = /* @__PURE__ */ new Map();
     this.evaluatorsById = /* @__PURE__ */ new Map();
-    this.schedulerFiber = effect.Effect.runFork(
-      this.createSchedulerEffect()
-    );
+    this.schedulerFiber = effect.Effect.runFork(this.createSchedulerEffect());
     this.persistenceFiber = effect.Effect.runFork(
       createPersistenceWorker(this.persistenceQueue)
     );
@@ -1866,9 +1823,9 @@ var EffectRunner = class {
     return effect.Effect.runSync(effect.Ref.get(this.snapshotsRef)).get(runId);
   }
   getAllRunSnapshots() {
-    return Array.from(
-      effect.Effect.runSync(effect.Ref.get(this.snapshotsRef)).values()
-    ).sort((a, b) => b.queuedAt - a.queuedAt);
+    return Array.from(effect.Effect.runSync(effect.Ref.get(this.snapshotsRef)).values()).sort(
+      (a, b) => b.queuedAt - a.queuedAt
+    );
   }
   async loadRunSnapshotsFromArtifacts() {
     return loadRunSnapshotsFromArtifacts(this.config);