@mastra/evals 1.1.0 → 1.1.1-alpha.0

package/dist/scorers/prebuilt/index.js

@@ -704,7 +704,9 @@ function createFaithfulnessScorer({
     type: "agent"
   }).preprocess({
     description: "Extract relevant statements from the LLM output",
-    outputSchema: z.array(z.string()),
+    outputSchema: z.object({
+      claims: z.array(z.string())
+    }),
     createPrompt: ({ run }) => {
       const prompt = createFaithfulnessExtractPrompt({ output: getAssistantMessageFromRunOutput(run.output) ?? "" });
       return prompt;
@@ -718,7 +720,7 @@ function createFaithfulnessScorer({
         (toolCall) => toolCall.state === "result" ? JSON.stringify(toolCall.result) : ""
       ) ?? [];
       const prompt = createFaithfulnessAnalyzePrompt({
-        claims: results.preprocessStepResult || [],
+        claims: results.preprocessStepResult?.claims || [],
         context
       });
       return prompt;
@@ -2117,6 +2119,7 @@ Example responses:
 }
 
 // src/scorers/llm/noise-sensitivity/index.ts
+var scoreSchema = z.number().refine((n) => n >= 0 && n <= 1, { message: "Score must be between 0 and 1" });
 var analyzeOutputSchema4 = z.object({
   dimensions: z.array(
     z.object({
@@ -2128,7 +2131,7 @@ var analyzeOutputSchema4 = z.object({
   ),
   overallAssessment: z.string(),
   majorIssues: z.array(z.string()).optional().default([]),
-  robustnessScore: z.number().min(0).max(1)
+  robustnessScore: scoreSchema
 });
 var DEFAULT_IMPACT_WEIGHTS = {
   none: 1,
@@ -2462,9 +2465,10 @@ Example responses:
 }
 
 // src/scorers/llm/prompt-alignment/index.ts
+var scoreSchema2 = z.number().refine((n) => n >= 0 && n <= 1, { message: "Score must be between 0 and 1" });
 var analyzeOutputSchema5 = z.object({
   intentAlignment: z.object({
-    score: z.number().min(0).max(1),
+    score: scoreSchema2,
     primaryIntent: z.string(),
     isAddressed: z.boolean(),
     reasoning: z.string()
@@ -2477,15 +2481,15 @@ var analyzeOutputSchema5 = z.object({
         reasoning: z.string()
       })
     ),
-    overallScore: z.number().min(0).max(1)
+    overallScore: scoreSchema2
   }),
   completeness: z.object({
-    score: z.number().min(0).max(1),
+    score: scoreSchema2,
     missingElements: z.array(z.string()),
     reasoning: z.string()
   }),
   responseAppropriateness: z.object({
-    score: z.number().min(0).max(1),
+    score: scoreSchema2,
     formatAlignment: z.boolean(),
     toneAlignment: z.boolean(),
     reasoning: z.string()