@m4trix/evals 0.29.0 → 0.30.0

package/dist/index.d.ts

@@ -274,6 +274,11 @@ interface EvalMiddleware<TCtx> {
 interface EvaluateMeta {
     /** Identifier of the trigger that started the run (for example, a CLI invocation). */
     triggerId: string;
+    /**
+     * Milliseconds since Unix epoch when the run was triggered (e.g. `Date.now()` at CLI start, or when
+     * `runDatasetWith` / `runDatasetJobsWithSharedConcurrency` was invoked). Shared across all jobs in a batch.
+     */
+    triggerTimestamp: number;
     /**
      * Identifier of the current test-case execution shared across all evaluators
      * for this specific test-case run.
@@ -593,6 +598,10 @@ interface RunDatasetRequest {
      * When omitted, the runner generates one in the format `trg-[uuid]`.
      */
     triggerId?: string;
+    /**
+     * When the run was triggered (`Date.now()` ms); defaults to now. Forwarded as `meta.triggerTimestamp`.
+     */
+    triggerTimestamp?: number;
     datasetId: string;
     evaluatorIds: ReadonlyArray<string>;
     /** RunConfig name surfaced on evaluator `meta` (from the job or `PROGRAMMATIC_RUN_CONFIG`). */
@@ -697,6 +706,10 @@ interface RunDatasetJobsWithSharedConcurrencyRequest {
     jobs: ReadonlyArray<RunDatasetJob>;
     globalConcurrency: number;
     triggerId?: string;
+    /**
+     * When the batch was triggered (`Date.now()` ms); defaults to now. CLI sets this once at command start.
+     */
+    triggerTimestamp?: number;
     /** Applied to every job in this batch (e.g. CLI `--experiment`). */
     experimentName?: string;
 }

package/dist/index.js

@@ -1566,6 +1566,7 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
               output,
               meta: {
                 triggerId: task.triggerId,
+                triggerTimestamp: task.triggerTimestamp,
                 runId: evaluatorRunId,
                 datasetName: task.dataset.getDisplayLabel(),
                 testCaseId: testCaseItem.id,
@@ -2046,6 +2047,7 @@ var EffectRunner = class {
     const globalConcurrency = Math.max(1, request.globalConcurrency);
     const sem = Effect.unsafeMakeSemaphore(globalConcurrency);
     const triggerId = request.triggerId ?? `trg-${randomUUID()}`;
+    const triggerTimestamp = request.triggerTimestamp ?? Date.now();
     const snapshots = [];
     for (const job of request.jobs) {
       snapshots.push(
@@ -2053,6 +2055,7 @@ var EffectRunner = class {
           datasetId: job.datasetId,
           evaluatorIds: job.evaluatorIds,
           triggerId,
+          triggerTimestamp,
           maxConcurrency: this.config.maxConcurrency ?? 1,
           globalEvaluationSemaphore: sem,
           runConfigName: job.runConfigName,
@@ -2090,6 +2093,7 @@ var EffectRunner = class {
       datasetId: request.datasetId,
       evaluatorIds: request.evaluatorIds,
       triggerId: request.triggerId,
+      triggerTimestamp: request.triggerTimestamp ?? Date.now(),
       maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
       repetitions: request.repetitions,
       runConfigName,
@@ -2117,6 +2121,7 @@ var EffectRunner = class {
     const totalEvaluations = selectedTestCases.length * repetitions;
     const runConfigTags = [...params.runConfigTags ?? []];
     const triggerId = params.triggerId ?? `trg-${randomUUID()}`;
+    const triggerTimestamp = params.triggerTimestamp ?? Date.now();
     const runId = `run-${randomUUID()}`;
     const artifactPath = createArtifactPath(this.config.artifactDirectory, params.datasetId, runId);
     const snapshot = {
@@ -2160,6 +2165,7 @@ var EffectRunner = class {
       Queue.offer(this.runQueue, {
         runId,
         triggerId,
+        triggerTimestamp,
         datasetId: params.datasetId,
         dataset: dataset.dataset,
         evaluators: selectedEvaluators,