@anvia/core 1.0.0-rc.6 → 1.0.0-rc.9

package/dist/evals/index.js

@@ -1,23 +1,26 @@
 import {
   AgentRunBlockedError
-} from "../chunk-VGJRXUPT.js";
+} from "../chunk-X4LK4ODN.js";
 import {
   cosineSimilarity,
   embedText
-} from "../chunk-H7BI3BSP.js";
+} from "../chunk-ZA564323.js";
 import {
   mapWithConcurrency
 } from "../chunk-3RM57ZT2.js";
 import {
   extract
-} from "../chunk-BOCJCECN.js";
-import "../chunk-MGNTXGTX.js";
+} from "../chunk-KIXN5DGS.js";
+import "../chunk-D4PACGPA.js";
 import {
   Usage
-} from "../chunk-HT4P3SSG.js";
-import "../chunk-G5OACQEI.js";
-import "../chunk-SROQA7I7.js";
-import "../chunk-UBE4WH6G.js";
+} from "../chunk-WTRGYFCD.js";
+import "../chunk-QGX73TSQ.js";
+import "../chunk-7JLAIN6E.js";
+import {
+  isJsonValue
+} from "../chunk-3XQGVDU5.js";
+import "../chunk-3RWESPUG.js";
 
 // src/evals/advanced-metrics.ts
 import { z } from "zod";
@@ -238,13 +241,14 @@ ${actual}`,
         usage = addUsage(usage, verdictResult.usage);
       }
       const score = verdicts.length === 0 ? 1 : verdicts.filter((verdict) => verdict.verdict !== "no").length / verdicts.length;
+      const serializedVerdicts = serializeVerdicts(verdicts);
       const reasonResult = await maybeReason({
         model: options.model,
         includeReason: config.includeReason,
         retries: config.retries,
         metric: "answer relevancy",
         score,
-        evidence: { input, verdicts }
+        evidence: { input, verdicts: serializedVerdicts }
       });
       usage = addUsage(usage, reasonResult.usage);
       return higherOutcome({
@@ -252,7 +256,7 @@ ${actual}`,
         threshold: config.threshold,
         strictMode: config.strictMode,
         comment: reasonResult.reason,
-        details: { statements, verdicts },
+        details: { statements, verdicts: serializedVerdicts },
         usage
       });
     } catch (error) {
@@ -320,6 +324,9 @@ function jsonCorrectness(options) {
         let validationError;
         try {
           parsed = JSON.parse(actual);
+          if (!isJsonValue(parsed)) {
+            throw new TypeError("Generated output is not a JSON value.");
+          }
           const result = options.schema.safeParse(parsed);
           if (!result.success) {
             validationError = z.prettifyError(result.error);
@@ -452,13 +459,14 @@ ${actual}`,
         (verdict) => verdict.verdict === "yes" || verdict.verdict === "idk" && !penalizeAmbiguousClaims
       ).length;
       const score = verdicts.length === 0 ? 1 : supported / verdicts.length;
+      const serializedVerdicts = serializeVerdicts(verdicts);
       const reasonResult = await maybeReason({
         model: options.model,
         includeReason: config.includeReason,
         retries: config.retries,
         metric: "faithfulness",
         score,
-        evidence: { verdicts, penalizeAmbiguousClaims }
+        evidence: { verdicts: serializedVerdicts, penalizeAmbiguousClaims }
       });
       usage = addUsage(usage, reasonResult.usage);
       return higherOutcome({
@@ -466,7 +474,7 @@ ${actual}`,
         threshold: config.threshold,
         strictMode: config.strictMode,
         comment: reasonResult.reason,
-        details: { truths, claims, verdicts, penalizeAmbiguousClaims },
+        details: { truths, claims, verdicts: serializedVerdicts, penalizeAmbiguousClaims },
         usage
       });
     } catch (error) {
@@ -630,13 +638,19 @@ ${actual}`,
       });
       const coverageScore = coverageTotal === 0 ? 0 : coverageMatched / coverageTotal;
       const score = Math.min(alignmentScore, coverageScore);
+      const serializedAlignmentVerdicts = serializeVerdicts(alignmentVerdicts);
       const reasonResult = await maybeReason({
         model: options.model,
         includeReason: config.includeReason,
         retries: config.retries,
         metric: "summarization",
         score,
-        evidence: { alignmentVerdicts, coverageVerdicts, alignmentScore, coverageScore }
+        evidence: {
+          alignmentVerdicts: serializedAlignmentVerdicts,
+          coverageVerdicts,
+          alignmentScore,
+          coverageScore
+        }
       });
       const usage = addUsage(
         truthResult.usage,
@@ -656,7 +670,7 @@ ${actual}`,
           truths,
           claims,
           assessmentQuestions: questions,
-          alignmentVerdicts,
+          alignmentVerdicts: serializedAlignmentVerdicts,
           coverageVerdicts,
           scoreBreakdown: { alignment: alignmentScore, coverage: coverageScore }
         },
@@ -1042,15 +1056,33 @@ function assertSameLength(label, inputs, outputs) {
     throw new Error(`${label} count ${outputs.length} did not match input count ${inputs.length}.`);
   }
 }
+function serializeVerdicts(verdicts) {
+  return verdicts.map((verdict) => {
+    const serialized = { verdict: verdict.verdict };
+    if (verdict.reason !== void 0) serialized.reason = verdict.reason;
+    return serialized;
+  });
+}
 function jsonPrompt(value) {
   return JSON.stringify(value, null, 2);
 }
 function toJsonValue(value) {
-  try {
-    return JSON.parse(JSON.stringify(value));
-  } catch {
-    return String(value);
-  }
+  if (!isJsonValue(value)) {
+    throw new TypeError("G-Eval expectedOutput must be a JSON value.");
+  }
+  return cloneJsonValue(value);
+}
+function cloneJsonValue(value) {
+  if (value === null || typeof value !== "object") return value;
+  if (Array.isArray(value)) return value.map(cloneJsonValue);
+  return Object.fromEntries(
+    Object.entries(value).map(([key, item]) => {
+      if (item === void 0) {
+        throw new TypeError("G-Eval expectedOutput must be a JSON value.");
+      }
+      return [key, cloneJsonValue(item)];
+    })
+  );
 }
 function normalizeEvalTurns(value) {
   const source = conversationArray(value);