@fallom/trace 0.2.25 → 0.2.26

package/dist/{core-Q3IHBEHB.mjs → core-46Z4Q54J.mjs}

@@ -7,7 +7,7 @@ import {
   evaluate,
   init,
   uploadResultsPublic
-} from "./chunk-3VWF2OJX.mjs";
+} from "./chunk-2NGJF2JZ.mjs";
 import "./chunk-7P6ASYW6.mjs";
 export {
   DEFAULT_JUDGE_MODEL,

package/dist/core-4L56QWI7.mjs

@@ -0,0 +1,21 @@
+import {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+} from "./chunk-3HBKT4HK.mjs";
+import "./chunk-7P6ASYW6.mjs";
+export {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+};

package/dist/core-DUG2SP2V.mjs

@@ -0,0 +1,21 @@
+import {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+} from "./chunk-GZ6TE7G4.mjs";
+import "./chunk-7P6ASYW6.mjs";
+export {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+};

package/dist/core-JLHYFVYS.mjs

@@ -0,0 +1,21 @@
+import {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+} from "./chunk-XBZ3ESNV.mjs";
+import "./chunk-7P6ASYW6.mjs";
+export {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+};

package/dist/core-NTEI2B5Z.mjs

@@ -0,0 +1,21 @@
+import {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+} from "./chunk-FTZVXPQN.mjs";
+import "./chunk-7P6ASYW6.mjs";
+export {
+  DEFAULT_JUDGE_MODEL,
+  _apiKey,
+  _baseUrl,
+  _initialized,
+  compareModels,
+  evaluate,
+  init,
+  uploadResultsPublic
+};

package/dist/index.d.mts

@@ -445,6 +445,12 @@ interface EvalResult {
     input: string;
     output: string;
     systemMessage?: string;
+    /** Expected/golden output for comparison (if provided) */
+    expectedOutput?: string;
+    /** Retrieved documents/context for RAG evaluation */
+    context?: string[];
+    /** Additional metadata */
+    metadata?: Record<string, unknown>;
     model: string;
     isProduction: boolean;
     answerRelevancy?: number;

package/dist/index.d.ts

@@ -445,6 +445,12 @@ interface EvalResult {
     input: string;
     output: string;
     systemMessage?: string;
+    /** Expected/golden output for comparison (if provided) */
+    expectedOutput?: string;
+    /** Retrieved documents/context for RAG evaluation */
+    context?: string[];
+    /** Additional metadata */
+    metadata?: Record<string, unknown>;
     model: string;
     isProduction: boolean;
     answerRelevancy?: number;

package/dist/index.js

@@ -1047,13 +1047,22 @@ 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 {
@@ -1072,10 +1081,14 @@ 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: {}
@@ -1175,6 +1188,7 @@ async function compareModels(options) {
           input: item.input,
           output,
           systemMessage: item.systemMessage,
+          metadata: item.metadata,
           model: model.name,
           isProduction: false,
           reasoning: {},
@@ -1286,6 +1300,9 @@ 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,
@@ -1395,7 +1412,7 @@ var import_exporter_trace_otlp_http = require("@opentelemetry/exporter-trace-otl
 // node_modules/@opentelemetry/resources/build/esm/Resource.js
 var import_api = require("@opentelemetry/api");
 
-// node_modules/@opentelemetry/resources/node_modules/@opentelemetry/semantic-conventions/build/esm/resource/SemanticResourceAttributes.js
+// node_modules/@opentelemetry/semantic-conventions/build/esm/resource/SemanticResourceAttributes.js
 var SemanticResourceAttributes = {
   /**
    * Name of the cloud provider.

package/dist/index.mjs

@@ -23,7 +23,7 @@ import {
   isCustomMetric,
   runGEval,
   uploadResultsPublic
-} from "./chunk-3VWF2OJX.mjs";
+} from "./chunk-FTZVXPQN.mjs";
 import {
   __export
 } from "./chunk-7P6ASYW6.mjs";
@@ -45,7 +45,7 @@ import { OTLPTraceExporter } from "@opentelemetry/exporter-trace-otlp-http";
 // node_modules/@opentelemetry/resources/build/esm/Resource.js
 import { diag } from "@opentelemetry/api";
 
-// node_modules/@opentelemetry/resources/node_modules/@opentelemetry/semantic-conventions/build/esm/resource/SemanticResourceAttributes.js
+// node_modules/@opentelemetry/semantic-conventions/build/esm/resource/SemanticResourceAttributes.js
 var SemanticResourceAttributes = {
   /**
    * Name of the cloud provider.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@fallom/trace",
-  "version": "0.2.25",
+  "version": "0.2.26",
   "description": "Model A/B testing and tracing for LLM applications. Zero latency, production-ready.",
   "main": "./dist/index.js",
   "module": "./dist/index.mjs",