@mastra/evals 0.10.8-alpha.0 → 0.11.0-alpha.2

package/dist/_tsup-dts-rollup.d.cts

@@ -405,7 +405,10 @@ export { createTextualDifferenceScorer }
 export { createTextualDifferenceScorer as createTextualDifferenceScorer_alias_1 }
 export { createTextualDifferenceScorer as createTextualDifferenceScorer_alias_2 }
 
-export declare function createToneScorer(): MastraScorer;
+declare function createToneScorer(): MastraScorer;
+export { createToneScorer }
+export { createToneScorer as createToneScorer_alias_1 }
+export { createToneScorer as createToneScorer_alias_2 }
 
 export declare function createToxicityAnalyzePrompt({ input, output }: {
     input: string;

package/dist/_tsup-dts-rollup.d.ts

@@ -405,7 +405,10 @@ export { createTextualDifferenceScorer }
 export { createTextualDifferenceScorer as createTextualDifferenceScorer_alias_1 }
 export { createTextualDifferenceScorer as createTextualDifferenceScorer_alias_2 }
 
-export declare function createToneScorer(): MastraScorer;
+declare function createToneScorer(): MastraScorer;
+export { createToneScorer }
+export { createToneScorer as createToneScorer_alias_1 }
+export { createToneScorer as createToneScorer_alias_2 }
 
 export declare function createToxicityAnalyzePrompt({ input, output }: {
     input: string;

package/dist/scorers/code/index.cjs

@@ -5,12 +5,14 @@ var nlp = require('compromise');
 var difflib = require('difflib');
 var keyword_extractor = require('keyword-extractor');
 var stringSimilarity = require('string-similarity');
+var Sentiment = require('sentiment');
 
 function _interopDefault (e) { return e && e.__esModule ? e : { default: e }; }
 
 var nlp__default = /*#__PURE__*/_interopDefault(nlp);
 var keyword_extractor__default = /*#__PURE__*/_interopDefault(keyword_extractor);
 var stringSimilarity__default = /*#__PURE__*/_interopDefault(stringSimilarity);
+var Sentiment__default = /*#__PURE__*/_interopDefault(Sentiment);
 
 function normalizeString(str) {
   return str.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase();
@@ -63,7 +65,7 @@ function createCompletenessScorer() {
       if (isInputInvalid || isOutputInvalid) {
         throw new Error("Inputs cannot be null or undefined");
       }
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       const inputToProcess = input;
       const outputToProcess = output;
@@ -100,7 +102,7 @@ function createTextualDifferenceScorer() {
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     analyze: async (run) => {
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       const matcher = new difflib.SequenceMatcher(null, input, output);
       const ratio = matcher.ratio();
@@ -125,7 +127,7 @@ function createKeywordCoverageScorer() {
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     extract: async (run) => {
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       if (!input && !output) {
         return {
@@ -182,7 +184,7 @@ function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace } = { igno
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     extract: async (run) => {
-      let processedInput = run.input.map((i) => i.content).join(", ");
+      let processedInput = run.input?.map((i) => i.content).join(", ") || "";
       let processedOutput = run.output.text;
       if (ignoreCase) {
         processedInput = processedInput.toLowerCase();
@@ -213,8 +215,46 @@ function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace } = { igno
     }
   });
 }
+function createToneScorer() {
+  return scores.createScorer({
+    name: "Completeness",
+    description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
+    analyze: async (run) => {
+      const sentiment = new Sentiment__default.default();
+      const input = run.input?.map((i) => i.content).join(", ") || "";
+      const output = run.output.text;
+      const responseSentiment = sentiment.analyze(input);
+      if (output) {
+        const referenceSentiment = sentiment.analyze(output);
+        const sentimentDiff = Math.abs(responseSentiment.comparative - referenceSentiment.comparative);
+        const normalizedScore = Math.max(0, 1 - sentimentDiff);
+        return {
+          score: normalizedScore,
+          result: {
+            responseSentiment: responseSentiment.comparative,
+            referenceSentiment: referenceSentiment.comparative,
+            difference: sentimentDiff
+          }
+        };
+      }
+      const sentences = input.match(/[^.!?]+[.!?]+/g) || [input];
+      const sentiments = sentences.map((s) => sentiment.analyze(s).comparative);
+      const avgSentiment = sentiments.reduce((a, b) => a + b, 0) / sentiments.length;
+      const variance = sentiments.reduce((sum, s) => sum + Math.pow(s - avgSentiment, 2), 0) / sentiments.length;
+      const stability = Math.max(0, 1 - variance);
+      return {
+        score: stability,
+        result: {
+          avgSentiment,
+          sentimentVariance: variance
+        }
+      };
+    }
+  });
+}
 
 exports.createCompletenessScorer = createCompletenessScorer;
 exports.createContentSimilarityScorer = createContentSimilarityScorer;
 exports.createKeywordCoverageScorer = createKeywordCoverageScorer;
 exports.createTextualDifferenceScorer = createTextualDifferenceScorer;
+exports.createToneScorer = createToneScorer;

package/dist/scorers/code/index.d.cts

@@ -2,3 +2,4 @@ export { createCompletenessScorer_alias_1 as createCompletenessScorer } from '..
 export { createTextualDifferenceScorer_alias_1 as createTextualDifferenceScorer } from '../../_tsup-dts-rollup.cjs';
 export { createKeywordCoverageScorer_alias_1 as createKeywordCoverageScorer } from '../../_tsup-dts-rollup.cjs';
 export { createContentSimilarityScorer_alias_1 as createContentSimilarityScorer } from '../../_tsup-dts-rollup.cjs';
+export { createToneScorer_alias_1 as createToneScorer } from '../../_tsup-dts-rollup.cjs';

package/dist/scorers/code/index.d.ts

@@ -2,3 +2,4 @@ export { createCompletenessScorer_alias_1 as createCompletenessScorer } from '..
 export { createTextualDifferenceScorer_alias_1 as createTextualDifferenceScorer } from '../../_tsup-dts-rollup.js';
 export { createKeywordCoverageScorer_alias_1 as createKeywordCoverageScorer } from '../../_tsup-dts-rollup.js';
 export { createContentSimilarityScorer_alias_1 as createContentSimilarityScorer } from '../../_tsup-dts-rollup.js';
+export { createToneScorer_alias_1 as createToneScorer } from '../../_tsup-dts-rollup.js';

package/dist/scorers/code/index.js

@@ -3,6 +3,7 @@ import nlp from 'compromise';
 import { SequenceMatcher } from 'difflib';
 import keyword_extractor from 'keyword-extractor';
 import stringSimilarity from 'string-similarity';
+import Sentiment from 'sentiment';
 
 function normalizeString(str) {
   return str.normalize("NFD").replace(/[\u0300-\u036f]/g, "").toLowerCase();
@@ -55,7 +56,7 @@ function createCompletenessScorer() {
       if (isInputInvalid || isOutputInvalid) {
         throw new Error("Inputs cannot be null or undefined");
       }
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       const inputToProcess = input;
       const outputToProcess = output;
@@ -92,7 +93,7 @@ function createTextualDifferenceScorer() {
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     analyze: async (run) => {
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       const matcher = new SequenceMatcher(null, input, output);
       const ratio = matcher.ratio();
@@ -117,7 +118,7 @@ function createKeywordCoverageScorer() {
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     extract: async (run) => {
-      const input = run.input.map((i) => i.content).join(", ");
+      const input = run.input?.map((i) => i.content).join(", ") || "";
       const output = run.output.text;
       if (!input && !output) {
         return {
@@ -174,7 +175,7 @@ function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace } = { igno
     name: "Completeness",
     description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
     extract: async (run) => {
-      let processedInput = run.input.map((i) => i.content).join(", ");
+      let processedInput = run.input?.map((i) => i.content).join(", ") || "";
       let processedOutput = run.output.text;
       if (ignoreCase) {
         processedInput = processedInput.toLowerCase();
@@ -205,5 +206,42 @@ function createContentSimilarityScorer({ ignoreCase, ignoreWhitespace } = { igno
     }
   });
 }
+function createToneScorer() {
+  return createScorer({
+    name: "Completeness",
+    description: 'Leverage the nlp method from "compromise" to extract elements from the input and output and calculate the coverage.',
+    analyze: async (run) => {
+      const sentiment = new Sentiment();
+      const input = run.input?.map((i) => i.content).join(", ") || "";
+      const output = run.output.text;
+      const responseSentiment = sentiment.analyze(input);
+      if (output) {
+        const referenceSentiment = sentiment.analyze(output);
+        const sentimentDiff = Math.abs(responseSentiment.comparative - referenceSentiment.comparative);
+        const normalizedScore = Math.max(0, 1 - sentimentDiff);
+        return {
+          score: normalizedScore,
+          result: {
+            responseSentiment: responseSentiment.comparative,
+            referenceSentiment: referenceSentiment.comparative,
+            difference: sentimentDiff
+          }
+        };
+      }
+      const sentences = input.match(/[^.!?]+[.!?]+/g) || [input];
+      const sentiments = sentences.map((s) => sentiment.analyze(s).comparative);
+      const avgSentiment = sentiments.reduce((a, b) => a + b, 0) / sentiments.length;
+      const variance = sentiments.reduce((sum, s) => sum + Math.pow(s - avgSentiment, 2), 0) / sentiments.length;
+      const stability = Math.max(0, 1 - variance);
+      return {
+        score: stability,
+        result: {
+          avgSentiment,
+          sentimentVariance: variance
+        }
+      };
+    }
+  });
+}
 
-export { createCompletenessScorer, createContentSimilarityScorer, createKeywordCoverageScorer, createTextualDifferenceScorer };
+export { createCompletenessScorer, createContentSimilarityScorer, createKeywordCoverageScorer, createTextualDifferenceScorer, createToneScorer };

package/dist/scorers/llm/index.cjs

@@ -243,7 +243,7 @@ function createAnswerRelevancyScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         return createReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           score: run.score,
           results: run.analyzeStepResult.results,
@@ -477,7 +477,7 @@ function createFaithfulnessScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         const prompt = createFaithfulnessReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           context: options?.context || [],
           score: run.score,
@@ -885,7 +885,7 @@ function createHallucinationScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         const prompt = createHallucinationReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           context: run?.additionalContext?.context || [],
           score: run.score,
@@ -999,7 +999,7 @@ function createToxicityScorer({ model, options }) {
       outputSchema: zod.z.object({ verdicts: zod.z.array(zod.z.object({ verdict: zod.z.string(), reason: zod.z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createToxicityAnalyzePrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text
         });
         return prompt;

package/dist/scorers/llm/index.js

@@ -241,7 +241,7 @@ function createAnswerRelevancyScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         return createReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           score: run.score,
           results: run.analyzeStepResult.results,
@@ -475,7 +475,7 @@ function createFaithfulnessScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         const prompt = createFaithfulnessReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           context: options?.context || [],
           score: run.score,
@@ -883,7 +883,7 @@ function createHallucinationScorer({
       description: "Reason about the results",
       createPrompt: ({ run }) => {
         const prompt = createHallucinationReasonPrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text,
           context: run?.additionalContext?.context || [],
           score: run.score,
@@ -997,7 +997,7 @@ function createToxicityScorer({ model, options }) {
       outputSchema: z.object({ verdicts: z.array(z.object({ verdict: z.string(), reason: z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createToxicityAnalyzePrompt({
-          input: run.input.map((input) => input.content).join(", "),
+          input: run.input?.map((input) => input.content).join(", ") || "",
           output: run.output.text
         });
         return prompt;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mastra/evals",
-  "version": "0.10.8-alpha.0",
+  "version": "0.11.0-alpha.2",
   "description": "",
   "type": "module",
   "files": [
@@ -84,7 +84,7 @@
     "zod": "^3.25.67"
   },
   "peerDependencies": {
-    "@mastra/core": ">=0.11.0-0 <0.12.0-0",
+    "@mastra/core": ">=0.12.0-0 <0.13.0-0",
     "ai": "^4.0.0"
   },
   "devDependencies": {
@@ -101,7 +101,7 @@
     "typescript": "^5.8.3",
     "vitest": "^3.2.4",
     "@internal/lint": "0.0.23",
-    "@mastra/core": "0.12.0-alpha.0"
+    "@mastra/core": "0.12.0-alpha.5"
   },
   "scripts": {
     "check": "tsc --noEmit",