@m4trix/evals 0.27.0 → 0.29.0

package/dist/index.cjs

@@ -841,7 +841,7 @@ var RunConfig = class _RunConfig {
   getDisplayLabel() {
     return this._displayName ?? this._name;
   }
-  /** Tags from `RunConfig.define({ tags })`; surfaced as `runConfigTags` on evaluator callbacks. */
+  /** Tags from `RunConfig.define({ tags })`; surfaced as `meta.runConfigTags` on evaluator callbacks. */
   getTags() {
     return [...this._tags];
   }
@@ -1014,10 +1014,11 @@ var TestCase = class _TestCase {
   static describe(config) {
     const name = validateTestCaseName(config.name, "TestCase.describe");
     const displayName = normalizeOptionalDisplayName(config.displayName);
+    const tags = config.tags !== void 0 ? [...config.tags] : [];
     return new _TestCase({
       name,
       displayName,
-      tags: config.tags,
+      tags,
       inputSchema: config.inputSchema,
       input: config.input,
       outputSchema: config.outputSchema,
@@ -1034,7 +1035,7 @@ var TestCase = class _TestCase {
     return this._config.displayName ?? this._config.name;
   }
   getTags() {
-    return this._config.tags;
+    return [...this._config.tags];
   }
   getInputSchema() {
     return this._config.inputSchema;
@@ -1592,14 +1593,17 @@ function processOneEvaluation(task, unit, totalEvaluations, publishEvent, persis
                 triggerId: task.triggerId,
                 runId: evaluatorRunId,
                 datasetName: task.dataset.getDisplayLabel(),
+                testCaseId: testCaseItem.id,
+                testCaseName: getTestCaseDisplayLabel(testCaseItem.testCase),
                 repetitionId,
                 repetitionIndex,
                 repetitionCount,
-                runConfigName: task.runConfigName
+                runConfigName: task.runConfigName,
+                ...task.experimentName !== void 0 && task.experimentName !== "" ? { experimentName: task.experimentName } : {},
+                testCaseTags: getTestCaseTagList(testCaseItem.testCase),
+                runConfigTags: task.runConfigTags,
+                evaluatorTags: getEvaluatorTagList(evaluator)
               },
-              testCaseTags: getTestCaseTagList(testCaseItem.testCase),
-              runConfigTags: task.runConfigTags,
-              evaluatorTags: getEvaluatorTagList(evaluator),
               logDiff,
               log,
               createError
@@ -2078,7 +2082,8 @@ var EffectRunner = class {
           globalEvaluationSemaphore: sem,
           runConfigName: job.runConfigName,
           runConfigTags: job.runConfigTags,
-          repetitions: job.repetitions
+          repetitions: job.repetitions,
+          experimentName: request.experimentName
         })
       );
     }
@@ -2113,7 +2118,8 @@ var EffectRunner = class {
       maxConcurrency: request.concurrency ?? this.config.maxConcurrency ?? 1,
       repetitions: request.repetitions,
       runConfigName,
-      runConfigTags: request.runConfigTags
+      runConfigTags: request.runConfigTags,
+      experimentName: request.experimentName
     });
   }
   async startDatasetRun(params) {
@@ -2188,7 +2194,8 @@ var EffectRunner = class {
         globalEvaluationSemaphore: params.globalEvaluationSemaphore,
         runConfigName: params.runConfigName,
         runConfigTags,
-        repetitions
+        repetitions,
+        experimentName: params.experimentName
       })
     );
     return snapshot;