@mastra/evals 0.10.6 → 0.10.8-alpha.0

package/dist/{dist-IVAARSAW.cjs → dist-JD6MNRVB.cjs}

@@ -11988,7 +11988,7 @@ function createTestHook(name, handler) {
   };
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/utils.XdZDrNZV.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/utils.XdZDrNZV.js
 var NAME_WORKER_STATE = "__vitest_worker__";
 function getWorkerState() {
   const workerState = globalThis[NAME_WORKER_STATE];
@@ -12036,7 +12036,7 @@ async function waitForImportsToResolve() {
   await waitForImportsToResolve();
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/_commonjsHelpers.BFTU3MAI.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/_commonjsHelpers.BFTU3MAI.js
 var commonjsGlobal = typeof globalThis !== "undefined" ? globalThis : typeof window !== "undefined" ? window : typeof global !== "undefined" ? global : typeof self !== "undefined" ? self : {};
 function getDefaultExportFromCjs3(x) {
   return x && x.__esModule && Object.prototype.hasOwnProperty.call(x, "default") ? x["default"] : x;
@@ -12889,7 +12889,7 @@ function offsetToLineNumber(source, offset) {
   return line + 1;
 }
 async function saveInlineSnapshots(environment, snapshots) {
-  const MagicString = (await import('./magic-string.es-66FD77JZ.cjs')).default;
+  const MagicString = (await import('./magic-string.es-T2QO2IBJ.cjs')).default;
   const files = new Set(snapshots.map((i) => i.file));
   await Promise.all(Array.from(files).map(async (file) => {
     const snaps = snapshots.filter((i) => i.file === file);
@@ -13666,7 +13666,7 @@ var SnapshotClient = class {
   }
 };
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/date.Bq6ZW5rf.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/date.Bq6ZW5rf.js
 var RealDate = Date;
 var now2 = null;
 var MockDate = class _MockDate extends RealDate {
@@ -13714,7 +13714,7 @@ function resetDate() {
   globalThis.Date = RealDate;
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/vi.bdSIJ99Y.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/vi.bdSIJ99Y.js
 var unsupported = [
   "matchSnapshot",
   "toMatchSnapshot",
@@ -16400,7 +16400,7 @@ function getImporter(name) {
   return stack?.file || "";
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/benchmark.CYdenmiT.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/benchmark.CYdenmiT.js
 var benchFns = /* @__PURE__ */ new WeakMap();
 var benchOptsMap = /* @__PURE__ */ new WeakMap();
 var bench = createBenchmark(function(name, fn2 = noop, options = {}) {
@@ -16426,12 +16426,12 @@ function formatName2(name) {
   return typeof name === "string" ? name : typeof name === "function" ? name.name || "<anonymous>" : String(name);
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/index.CdQS2e2Q.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/index.CdQS2e2Q.js
 chunkIS3BZTWE_cjs.__toESM(require_dist(), 1);
 var assertType = function assertType2() {
 };
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/index.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/index.js
 var import_expect_type2 = chunkIS3BZTWE_cjs.__toESM(require_dist(), 1);
 var export_expectTypeOf = import_expect_type2.expectTypeOf;
 /*! Bundled license information:

package/dist/{dist-5JXLPLM2.js → dist-ZXFGMR47.js}

@@ -11986,7 +11986,7 @@ function createTestHook(name, handler) {
   };
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/utils.XdZDrNZV.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/utils.XdZDrNZV.js
 var NAME_WORKER_STATE = "__vitest_worker__";
 function getWorkerState() {
   const workerState = globalThis[NAME_WORKER_STATE];
@@ -12034,7 +12034,7 @@ async function waitForImportsToResolve() {
   await waitForImportsToResolve();
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/_commonjsHelpers.BFTU3MAI.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/_commonjsHelpers.BFTU3MAI.js
 var commonjsGlobal = typeof globalThis !== "undefined" ? globalThis : typeof window !== "undefined" ? window : typeof global !== "undefined" ? global : typeof self !== "undefined" ? self : {};
 function getDefaultExportFromCjs3(x) {
   return x && x.__esModule && Object.prototype.hasOwnProperty.call(x, "default") ? x["default"] : x;
@@ -12887,7 +12887,7 @@ function offsetToLineNumber(source, offset) {
   return line + 1;
 }
 async function saveInlineSnapshots(environment, snapshots) {
-  const MagicString = (await import('./magic-string.es-LD4FLE5J.js')).default;
+  const MagicString = (await import('./magic-string.es-MNZ6ZGOL.js')).default;
   const files = new Set(snapshots.map((i) => i.file));
   await Promise.all(Array.from(files).map(async (file) => {
     const snaps = snapshots.filter((i) => i.file === file);
@@ -13664,7 +13664,7 @@ var SnapshotClient = class {
   }
 };
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/date.Bq6ZW5rf.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/date.Bq6ZW5rf.js
 var RealDate = Date;
 var now2 = null;
 var MockDate = class _MockDate extends RealDate {
@@ -13712,7 +13712,7 @@ function resetDate() {
   globalThis.Date = RealDate;
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/vi.bdSIJ99Y.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/vi.bdSIJ99Y.js
 var unsupported = [
   "matchSnapshot",
   "toMatchSnapshot",
@@ -16398,7 +16398,7 @@ function getImporter(name) {
   return stack?.file || "";
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/benchmark.CYdenmiT.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/benchmark.CYdenmiT.js
 var benchFns = /* @__PURE__ */ new WeakMap();
 var benchOptsMap = /* @__PURE__ */ new WeakMap();
 var bench = createBenchmark(function(name, fn2 = noop, options = {}) {
@@ -16424,12 +16424,12 @@ function formatName2(name) {
   return typeof name === "string" ? name : typeof name === "function" ? name.name || "<anonymous>" : String(name);
 }
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/chunks/index.CdQS2e2Q.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/chunks/index.CdQS2e2Q.js
 __toESM(require_dist(), 1);
 var assertType = function assertType2() {
 };
 
-// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.2_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_bcb3b08cf6e48f0a4d90e52d1b155bc5/node_modules/vitest/dist/index.js
+// ../../node_modules/.pnpm/vitest@3.2.4_@types+debug@4.1.12_@types+node@20.19.9_@vitest+ui@3.2.3_jiti@2.4.2_jsdom@_0090e69ea15e68f4eaa34b37eb448faf/node_modules/vitest/dist/index.js
 var import_expect_type2 = __toESM(require_dist(), 1);
 var export_expectTypeOf = import_expect_type2.expectTypeOf;
 /*! Bundled license information:

package/dist/index.cjs

@@ -41,7 +41,7 @@ var getCurrentTestInfo = async () => {
     };
   }
   try {
-    const vitest = await import('./dist-IVAARSAW.cjs');
+    const vitest = await import('./dist-JD6MNRVB.cjs');
     if (typeof vitest !== "undefined" && vitest.expect?.getState) {
       const state = vitest.expect.getState();
       return {

package/dist/index.js

@@ -39,7 +39,7 @@ var getCurrentTestInfo = async () => {
     };
   }
   try {
-    const vitest = await import('./dist-5JXLPLM2.js');
+    const vitest = await import('./dist-ZXFGMR47.js');
     if (typeof vitest !== "undefined" && vitest.expect?.getState) {
       const state = vitest.expect.getState();
       return {

package/dist/{magic-string.es-LD4FLE5J.js → magic-string.es-MNZ6ZGOL.js}

@@ -1,4 +1,4 @@
-// ../../node_modules/.pnpm/@jridgewell+sourcemap-codec@1.5.3/node_modules/@jridgewell/sourcemap-codec/dist/sourcemap-codec.mjs
+// ../../node_modules/.pnpm/@jridgewell+sourcemap-codec@1.5.4/node_modules/@jridgewell/sourcemap-codec/dist/sourcemap-codec.mjs
 var comma = ",".charCodeAt(0);
 var semicolon = ";".charCodeAt(0);
 var chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";

package/dist/{magic-string.es-66FD77JZ.cjs → magic-string.es-T2QO2IBJ.cjs}

@@ -2,7 +2,7 @@
 
 Object.defineProperty(exports, '__esModule', { value: true });
 
-// ../../node_modules/.pnpm/@jridgewell+sourcemap-codec@1.5.3/node_modules/@jridgewell/sourcemap-codec/dist/sourcemap-codec.mjs
+// ../../node_modules/.pnpm/@jridgewell+sourcemap-codec@1.5.4/node_modules/@jridgewell/sourcemap-codec/dist/sourcemap-codec.mjs
 var comma = ",".charCodeAt(0);
 var semicolon = ";".charCodeAt(0);
 var chars = "ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/";

package/dist/scorers/llm/index.cjs

@@ -236,7 +236,7 @@ function createAnswerRelevancyScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: zod.z.array(zod.z.object({ result: zod.z.string(), reason: zod.z.string() })),
+      outputSchema: zod.z.object({ results: zod.z.array(zod.z.object({ result: zod.z.string(), reason: zod.z.string() })) }),
       createPrompt: ({ run }) => createScorePrompt(JSON.stringify(run.input), run.extractStepResult?.statements || [])
     },
     reason: {
@@ -246,18 +246,18 @@ function createAnswerRelevancyScorer({
           input: run.input.map((input) => input.content).join(", "),
           output: run.output.text,
           score: run.score,
-          results: run.analyzeStepResult,
+          results: run.analyzeStepResult.results,
           scale: options.scale
         });
       }
     },
     calculateScore: ({ run }) => {
-      if (!run.analyzeStepResult || run.analyzeStepResult.length === 0) {
+      if (!run.analyzeStepResult || run.analyzeStepResult.results.length === 0) {
         return 0;
       }
-      const numberOfResults = run.analyzeStepResult.length;
+      const numberOfResults = run.analyzeStepResult.results.length;
       let relevancyCount = 0;
-      for (const { result } of run.analyzeStepResult) {
+      for (const { result } of run.analyzeStepResult.results) {
         if (result.trim().toLowerCase() === "yes") {
           relevancyCount++;
         } else if (result.trim().toLowerCase() === "unsure") {
@@ -455,7 +455,7 @@ function createFaithfulnessScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: zod.z.array(zod.z.object({ verdict: zod.z.string(), reason: zod.z.string() })),
+      outputSchema: zod.z.object({ verdicts: zod.z.array(zod.z.object({ verdict: zod.z.string(), reason: zod.z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createFaithfulnessAnalyzePrompt({
           claims: run.extractStepResult || [],
@@ -465,8 +465,8 @@ function createFaithfulnessScorer({
       }
     },
     calculateScore: ({ run }) => {
-      const totalClaims = run.analyzeStepResult.length;
-      const supportedClaims = run.analyzeStepResult.filter((v) => v.verdict === "yes").length;
+      const totalClaims = run.analyzeStepResult.verdicts.length;
+      const supportedClaims = run.analyzeStepResult.verdicts.filter((v) => v.verdict === "yes").length;
       if (totalClaims === 0) {
         return 0;
       }
@@ -482,7 +482,7 @@ function createFaithfulnessScorer({
           context: options?.context || [],
           score: run.score,
           scale: options?.scale || 1,
-          verdicts: run.analyzeStepResult || []
+          verdicts: run.analyzeStepResult?.verdicts || []
         });
         return prompt;
       }
@@ -617,7 +617,7 @@ function createBiasScorer({ model, options }) {
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: zod.z.array(zod.z.object({ result: zod.z.string(), reason: zod.z.string() })),
+      outputSchema: zod.z.object({ results: zod.z.array(zod.z.object({ result: zod.z.string(), reason: zod.z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createBiasAnalyzePrompt({
           output: run.output.text,
@@ -627,17 +627,20 @@ function createBiasScorer({ model, options }) {
       }
     },
     calculateScore: ({ run }) => {
-      if (!run.analyzeStepResult || run.analyzeStepResult.length === 0) {
+      if (!run.analyzeStepResult || run.analyzeStepResult.results.length === 0) {
         return 0;
       }
-      const biasedVerdicts = run.analyzeStepResult.filter((v) => v.result.toLowerCase() === "yes");
-      const score = biasedVerdicts.length / run.analyzeStepResult.length;
+      const biasedVerdicts = run.analyzeStepResult.results.filter((v) => v.result.toLowerCase() === "yes");
+      const score = biasedVerdicts.length / run.analyzeStepResult.results.length;
       return roundToTwoDecimals2(score * (options?.scale || 1));
     },
     reason: {
       description: "Reason about the results",
       createPrompt: ({ run }) => {
-        return createBiasReasonPrompt({ score: run.score, biases: run.analyzeStepResult?.map((v) => v.reason) || [] });
+        return createBiasReasonPrompt({
+          score: run.score,
+          biases: run.analyzeStepResult?.results.map((v) => v.reason) || []
+        });
       }
     }
   });
@@ -858,7 +861,9 @@ function createHallucinationScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: zod.z.array(zod.z.object({ statement: zod.z.string(), verdict: zod.z.string(), reason: zod.z.string() })),
+      outputSchema: zod.z.object({
+        verdicts: zod.z.array(zod.z.object({ statement: zod.z.string(), verdict: zod.z.string(), reason: zod.z.string() }))
+      }),
       createPrompt: ({ run }) => {
         const prompt = createHallucinationAnalyzePrompt({
           claims: run.extractStepResult.claims,
@@ -868,8 +873,8 @@ function createHallucinationScorer({
       }
     },
     calculateScore: ({ run }) => {
-      const totalStatements = run.analyzeStepResult.length;
-      const contradictedStatements = run.analyzeStepResult.filter((v) => v.verdict === "yes").length;
+      const totalStatements = run.analyzeStepResult.verdicts.length;
+      const contradictedStatements = run.analyzeStepResult.verdicts.filter((v) => v.verdict === "yes").length;
       if (totalStatements === 0) {
         return 0;
       }
@@ -885,7 +890,7 @@ function createHallucinationScorer({
           context: run?.additionalContext?.context || [],
           score: run.score,
           scale: options?.scale || 1,
-          verdicts: run.analyzeStepResult || []
+          verdicts: run.analyzeStepResult?.verdicts || []
         });
         return prompt;
       }
@@ -991,7 +996,7 @@ function createToxicityScorer({ model, options }) {
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: zod.z.array(zod.z.object({ verdict: zod.z.string(), reason: zod.z.string() })),
+      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(", "),
@@ -1001,12 +1006,12 @@ function createToxicityScorer({ model, options }) {
       }
     },
     calculateScore: ({ run }) => {
-      const numberOfVerdicts = run.analyzeStepResult?.length || 0;
+      const numberOfVerdicts = run.analyzeStepResult?.verdicts.length || 0;
       if (numberOfVerdicts === 0) {
         return 1;
       }
       let toxicityCount = 0;
-      for (const { verdict } of run.analyzeStepResult) {
+      for (const { verdict } of run.analyzeStepResult.verdicts) {
         if (verdict.trim().toLowerCase() === "yes") {
           toxicityCount++;
         }
@@ -1019,7 +1024,7 @@ function createToxicityScorer({ model, options }) {
       createPrompt: ({ run }) => {
         const prompt = createToxicityReasonPrompt({
           score: run.score,
-          toxics: run.analyzeStepResult?.map((v) => v.reason) || []
+          toxics: run.analyzeStepResult?.verdicts.map((v) => v.reason) || []
         });
         return prompt;
       }

package/dist/scorers/llm/index.js

@@ -234,7 +234,7 @@ function createAnswerRelevancyScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: z.array(z.object({ result: z.string(), reason: z.string() })),
+      outputSchema: z.object({ results: z.array(z.object({ result: z.string(), reason: z.string() })) }),
       createPrompt: ({ run }) => createScorePrompt(JSON.stringify(run.input), run.extractStepResult?.statements || [])
     },
     reason: {
@@ -244,18 +244,18 @@ function createAnswerRelevancyScorer({
           input: run.input.map((input) => input.content).join(", "),
           output: run.output.text,
           score: run.score,
-          results: run.analyzeStepResult,
+          results: run.analyzeStepResult.results,
           scale: options.scale
         });
       }
     },
     calculateScore: ({ run }) => {
-      if (!run.analyzeStepResult || run.analyzeStepResult.length === 0) {
+      if (!run.analyzeStepResult || run.analyzeStepResult.results.length === 0) {
         return 0;
       }
-      const numberOfResults = run.analyzeStepResult.length;
+      const numberOfResults = run.analyzeStepResult.results.length;
       let relevancyCount = 0;
-      for (const { result } of run.analyzeStepResult) {
+      for (const { result } of run.analyzeStepResult.results) {
         if (result.trim().toLowerCase() === "yes") {
           relevancyCount++;
         } else if (result.trim().toLowerCase() === "unsure") {
@@ -453,7 +453,7 @@ function createFaithfulnessScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: z.array(z.object({ verdict: z.string(), reason: z.string() })),
+      outputSchema: z.object({ verdicts: z.array(z.object({ verdict: z.string(), reason: z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createFaithfulnessAnalyzePrompt({
           claims: run.extractStepResult || [],
@@ -463,8 +463,8 @@ function createFaithfulnessScorer({
       }
     },
     calculateScore: ({ run }) => {
-      const totalClaims = run.analyzeStepResult.length;
-      const supportedClaims = run.analyzeStepResult.filter((v) => v.verdict === "yes").length;
+      const totalClaims = run.analyzeStepResult.verdicts.length;
+      const supportedClaims = run.analyzeStepResult.verdicts.filter((v) => v.verdict === "yes").length;
       if (totalClaims === 0) {
         return 0;
       }
@@ -480,7 +480,7 @@ function createFaithfulnessScorer({
           context: options?.context || [],
           score: run.score,
           scale: options?.scale || 1,
-          verdicts: run.analyzeStepResult || []
+          verdicts: run.analyzeStepResult?.verdicts || []
         });
         return prompt;
       }
@@ -615,7 +615,7 @@ function createBiasScorer({ model, options }) {
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: z.array(z.object({ result: z.string(), reason: z.string() })),
+      outputSchema: z.object({ results: z.array(z.object({ result: z.string(), reason: z.string() })) }),
       createPrompt: ({ run }) => {
         const prompt = createBiasAnalyzePrompt({
           output: run.output.text,
@@ -625,17 +625,20 @@ function createBiasScorer({ model, options }) {
       }
     },
     calculateScore: ({ run }) => {
-      if (!run.analyzeStepResult || run.analyzeStepResult.length === 0) {
+      if (!run.analyzeStepResult || run.analyzeStepResult.results.length === 0) {
         return 0;
       }
-      const biasedVerdicts = run.analyzeStepResult.filter((v) => v.result.toLowerCase() === "yes");
-      const score = biasedVerdicts.length / run.analyzeStepResult.length;
+      const biasedVerdicts = run.analyzeStepResult.results.filter((v) => v.result.toLowerCase() === "yes");
+      const score = biasedVerdicts.length / run.analyzeStepResult.results.length;
       return roundToTwoDecimals2(score * (options?.scale || 1));
     },
     reason: {
       description: "Reason about the results",
       createPrompt: ({ run }) => {
-        return createBiasReasonPrompt({ score: run.score, biases: run.analyzeStepResult?.map((v) => v.reason) || [] });
+        return createBiasReasonPrompt({
+          score: run.score,
+          biases: run.analyzeStepResult?.results.map((v) => v.reason) || []
+        });
       }
     }
   });
@@ -856,7 +859,9 @@ function createHallucinationScorer({
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: z.array(z.object({ statement: z.string(), verdict: z.string(), reason: z.string() })),
+      outputSchema: z.object({
+        verdicts: z.array(z.object({ statement: z.string(), verdict: z.string(), reason: z.string() }))
+      }),
       createPrompt: ({ run }) => {
         const prompt = createHallucinationAnalyzePrompt({
           claims: run.extractStepResult.claims,
@@ -866,8 +871,8 @@ function createHallucinationScorer({
       }
     },
     calculateScore: ({ run }) => {
-      const totalStatements = run.analyzeStepResult.length;
-      const contradictedStatements = run.analyzeStepResult.filter((v) => v.verdict === "yes").length;
+      const totalStatements = run.analyzeStepResult.verdicts.length;
+      const contradictedStatements = run.analyzeStepResult.verdicts.filter((v) => v.verdict === "yes").length;
       if (totalStatements === 0) {
         return 0;
       }
@@ -883,7 +888,7 @@ function createHallucinationScorer({
           context: run?.additionalContext?.context || [],
           score: run.score,
           scale: options?.scale || 1,
-          verdicts: run.analyzeStepResult || []
+          verdicts: run.analyzeStepResult?.verdicts || []
         });
         return prompt;
       }
@@ -989,7 +994,7 @@ function createToxicityScorer({ model, options }) {
     },
     analyze: {
       description: "Score the relevance of the statements to the input",
-      outputSchema: z.array(z.object({ verdict: z.string(), reason: z.string() })),
+      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(", "),
@@ -999,12 +1004,12 @@ function createToxicityScorer({ model, options }) {
       }
     },
     calculateScore: ({ run }) => {
-      const numberOfVerdicts = run.analyzeStepResult?.length || 0;
+      const numberOfVerdicts = run.analyzeStepResult?.verdicts.length || 0;
       if (numberOfVerdicts === 0) {
         return 1;
       }
       let toxicityCount = 0;
-      for (const { verdict } of run.analyzeStepResult) {
+      for (const { verdict } of run.analyzeStepResult.verdicts) {
         if (verdict.trim().toLowerCase() === "yes") {
           toxicityCount++;
         }
@@ -1017,7 +1022,7 @@ function createToxicityScorer({ model, options }) {
       createPrompt: ({ run }) => {
         const prompt = createToxicityReasonPrompt({
           score: run.score,
-          toxics: run.analyzeStepResult?.map((v) => v.reason) || []
+          toxics: run.analyzeStepResult?.verdicts.map((v) => v.reason) || []
         });
         return prompt;
       }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mastra/evals",
-  "version": "0.10.6",
+  "version": "0.10.8-alpha.0",
   "description": "",
   "type": "module",
   "files": [
@@ -84,7 +84,7 @@
     "zod": "^3.25.67"
   },
   "peerDependencies": {
-    "@mastra/core": "^0.10.1-alpha.0",
+    "@mastra/core": ">=0.11.0-0 <0.12.0-0",
     "ai": "^4.0.0"
   },
   "devDependencies": {
@@ -100,8 +100,8 @@
     "tsup": "^8.5.0",
     "typescript": "^5.8.3",
     "vitest": "^3.2.4",
-    "@internal/lint": "0.0.21",
-    "@mastra/core": "0.11.0"
+    "@internal/lint": "0.0.23",
+    "@mastra/core": "0.12.0-alpha.0"
   },
   "scripts": {
     "check": "tsc --noEmit",