@fallom/trace 0.2.25 → 0.2.26

package/dist/{chunk-3VWF2OJX.mjs → chunk-FTZVXPQN.mjs}

@@ -423,7 +423,7 @@ function datasetFromTraces(traces) {
   return items;
 }
 async function datasetFromFallom(datasetKey, version, config) {
-  const { _apiKey: _apiKey2, _baseUrl: _baseUrl2, _initialized: _initialized2 } = await import("./core-Q3IHBEHB.mjs").then(
+  const { _apiKey: _apiKey2, _baseUrl: _baseUrl2, _initialized: _initialized2 } = await import("./core-NTEI2B5Z.mjs").then(
     (m) => ({
       _apiKey: config?._apiKey ?? m._apiKey,
       _baseUrl: config?._baseUrl ?? m._baseUrl,
@@ -496,7 +496,7 @@ var EvaluationDataset = class {
    * @returns Self for chaining
    */
   async pull(alias, version) {
-    const { _apiKey: _apiKey2, _baseUrl: _baseUrl2, _initialized: _initialized2 } = await import("./core-Q3IHBEHB.mjs");
+    const { _apiKey: _apiKey2, _baseUrl: _baseUrl2, _initialized: _initialized2 } = await import("./core-NTEI2B5Z.mjs");
     if (!_initialized2) {
       throw new Error("Fallom evals not initialized. Call evals.init() first.");
     }
@@ -698,13 +698,22 @@ async function evaluate(options) {
     _skipUpload = false
   } = options;
   let dataset;
+  let testCaseExtras = /* @__PURE__ */ new Map();
   if (testCases !== void 0 && testCases.length > 0) {
-    dataset = testCases.map((tc) => ({
-      input: tc.input,
-      output: tc.actualOutput,
-      systemMessage: tc.systemMessage,
-      metadata: tc.metadata
-    }));
+    dataset = testCases.map((tc, idx) => {
+      if (tc.expectedOutput || tc.context) {
+        testCaseExtras.set(idx, {
+          expectedOutput: tc.expectedOutput,
+          context: tc.context
+        });
+      }
+      return {
+        input: tc.input,
+        output: tc.actualOutput,
+        systemMessage: tc.systemMessage,
+        metadata: tc.metadata
+      };
+    });
   } else if (datasetInput !== void 0) {
     dataset = await resolveDataset(datasetInput);
   } else {
@@ -723,10 +732,14 @@ async function evaluate(options) {
   for (let i = 0; i < dataset.length; i++) {
     const item = dataset[i];
     if (verbose) console.log(`Evaluating item ${i + 1}/${dataset.length}...`);
+    const extras = testCaseExtras.get(i);
     const result = {
       input: item.input,
       output: item.output,
       systemMessage: item.systemMessage,
+      expectedOutput: extras?.expectedOutput,
+      context: extras?.context,
+      metadata: item.metadata,
       model: "production",
       isProduction: true,
       reasoning: {}
@@ -826,6 +839,7 @@ async function compareModels(options) {
           input: item.input,
           output,
           systemMessage: item.systemMessage,
+          metadata: item.metadata,
           model: model.name,
           isProduction: false,
           reasoning: {},
@@ -937,6 +951,9 @@ async function uploadResults(results, name, description, judgeModel, verbose) {
     results: allResults.map((r) => ({
       input: r.input,
       system_message: r.systemMessage,
+      expected_output: r.expectedOutput,
+      context: r.context,
+      metadata: r.metadata,
       model: r.model,
       output: r.output,
       is_production: r.isProduction,