@agentv/core 3.4.0 → 3.6.0

package/dist/index.cjs

@@ -55,7 +55,7 @@ function createLanguageModel(modelString) {
     case "anthropic":
       return (0, import_anthropic.createAnthropic)()(modelName);
     case "azure":
-      return (0, import_azure.createAzure)()(modelName);
+      return (0, import_azure.createAzure)().chat(modelName);
     case "google":
       return (0, import_google.createGoogleGenerativeAI)()(modelName);
     default:
@@ -1580,7 +1580,6 @@ __export(index_exports, {
   freeformEvaluationSchema: () => freeformEvaluationSchema,
   generateRubrics: () => generateRubrics,
   getAgentvHome: () => getAgentvHome,
-  getHitCount: () => getHitCount,
   getOutputFilenames: () => getOutputFilenames,
   getSubagentsRoot: () => getSubagentsRoot,
   getTraceStateRoot: () => getTraceStateRoot,
@@ -1730,9 +1729,6 @@ var EVALUATOR_KIND_SET = new Set(EVALUATOR_KIND_VALUES);
 function isEvaluatorKind(value) {
   return typeof value === "string" && EVALUATOR_KIND_SET.has(value);
 }
-function getHitCount(result) {
-  return result.hits.length;
-}
 
 // src/evaluation/trace.ts
 function computeTraceSummary(messages) {
@@ -2449,14 +2445,8 @@ var import_promises5 = require("fs/promises");
 
 // src/evaluation/template-variables.ts
 var TEMPLATE_VARIABLES = {
-  /** @deprecated Use OUTPUT_TEXT instead */
-  ANSWER: "answer",
   EXPECTED_OUTPUT: "expected_output",
-  /** @deprecated Use INPUT_TEXT instead */
-  QUESTION: "question",
   CRITERIA: "criteria",
-  /** @deprecated Use EXPECTED_OUTPUT_TEXT instead */
-  REFERENCE_ANSWER: "reference_answer",
   INPUT: "input",
   OUTPUT: "output",
   FILE_CHANGES: "file_changes",
@@ -2466,9 +2456,8 @@ var TEMPLATE_VARIABLES = {
 };
 var VALID_TEMPLATE_VARIABLES = new Set(Object.values(TEMPLATE_VARIABLES));
 var REQUIRED_TEMPLATE_VARIABLES = /* @__PURE__ */ new Set([
-  TEMPLATE_VARIABLES.ANSWER,
-  TEMPLATE_VARIABLES.EXPECTED_OUTPUT,
-  TEMPLATE_VARIABLES.OUTPUT_TEXT
+  TEMPLATE_VARIABLES.OUTPUT_TEXT,
+  TEMPLATE_VARIABLES.EXPECTED_OUTPUT
 ]);
 
 // src/evaluation/validation/prompt-validator.ts
@@ -2491,13 +2480,13 @@ function validateTemplateVariables(content, source) {
     }
     match = variablePattern.exec(content);
   }
-  const hasCandidateAnswer = foundVariables.has(TEMPLATE_VARIABLES.ANSWER) || foundVariables.has(TEMPLATE_VARIABLES.OUTPUT_TEXT);
+  const hasCandidateAnswer = foundVariables.has(TEMPLATE_VARIABLES.OUTPUT_TEXT);
   const hasExpectedOutput = foundVariables.has(TEMPLATE_VARIABLES.EXPECTED_OUTPUT);
   const hasRequiredFields = hasCandidateAnswer || hasExpectedOutput;
   if (!hasRequiredFields) {
     throw new Error(
       `Missing required fields. Must include at least one of:
-  - {{ ${TEMPLATE_VARIABLES.ANSWER} }} or {{ ${TEMPLATE_VARIABLES.OUTPUT_TEXT} }}
+  - {{ ${TEMPLATE_VARIABLES.OUTPUT_TEXT} }}
   - {{ ${TEMPLATE_VARIABLES.EXPECTED_OUTPUT} }}`
     );
   }
@@ -5576,7 +5565,7 @@ var AzureProvider = class {
     };
     this.retryConfig = config.retry;
     const azure = (0, import_azure2.createAzure)(buildAzureOptions(config));
-    this.model = azure(config.deploymentName);
+    this.model = azure.chat(config.deploymentName);
   }
   id;
   kind = "azure";
@@ -5799,6 +5788,8 @@ async function invokeModel(options) {
   const { model, request, defaults, retryConfig, providerOptions } = options;
   const chatPrompt = buildChatPrompt(request);
   const { temperature, maxOutputTokens } = resolveModelSettings(request, defaults);
+  const startTime = (/* @__PURE__ */ new Date()).toISOString();
+  const startMs = Date.now();
   const result = await withRetry(
     () => (0, import_ai.generateText)({
       model,
@@ -5812,9 +5803,11 @@ async function invokeModel(options) {
     retryConfig,
     request.signal
   );
-  return mapResponse(result);
+  const endTime = (/* @__PURE__ */ new Date()).toISOString();
+  const durationMs = Date.now() - startMs;
+  return mapResponse(result, { durationMs, startTime, endTime });
 }
-function mapResponse(result) {
+function mapResponse(result, timing) {
   const content = result.text ?? "";
   const rawUsage = result.totalUsage ?? result.usage;
   const reasoning = rawUsage?.outputTokenDetails?.reasoningTokens ?? void 0;
@@ -5829,7 +5822,10 @@ function mapResponse(result) {
     raw: result,
     usage: toJsonObject(rawUsage),
     output: [{ role: "assistant", content }],
-    tokenUsage
+    tokenUsage,
+    durationMs: timing?.durationMs,
+    startTime: timing?.startTime,
+    endTime: timing?.endTime
   };
 }
 function toJsonObject(value) {
@@ -6707,10 +6703,12 @@ var ClaudeSdkProvider = class {
             if (usage) {
               const inputTokens = (usage.input_tokens ?? 0) + (usage.cache_read_input_tokens ?? 0) + (usage.cache_creation_input_tokens ?? 0);
               const outputTokens = usage.output_tokens ?? 0;
+              const reasoningTokens = usage.reasoning_tokens ?? void 0;
               tokenUsage = {
                 input: inputTokens,
                 output: outputTokens,
-                cached: usage.cache_read_input_tokens ?? void 0
+                cached: usage.cache_read_input_tokens ?? void 0,
+                reasoning: reasoningTokens
               };
               request.streamCallbacks?.onLlmCallEnd?.(this.config.model ?? "claude", tokenUsage);
             }
@@ -7724,7 +7722,8 @@ ${basePrompt}` : basePrompt;
           onUsage({
             input: usage.input_tokens ?? 0,
             output: usage.output_tokens ?? 0,
-            cached: usage.cached_input_tokens ?? void 0
+            cached: usage.cached_input_tokens ?? void 0,
+            reasoning: usage.reasoning_tokens ?? void 0
           });
         }
       }
@@ -9739,10 +9738,12 @@ function extractTokenUsage(events) {
           output: output ?? 0
         };
         const cached = toFiniteNumber(u.cache_read_input_tokens ?? u.cached ?? u.cachedTokens);
-        if (cached !== void 0) {
-          return { ...result, cached };
-        }
-        return result;
+        const reasoning = toFiniteNumber(u.reasoning_tokens ?? u.reasoningTokens ?? u.reasoning);
+        return {
+          ...result,
+          ...cached !== void 0 ? { cached } : {},
+          ...reasoning !== void 0 ? { reasoning } : {}
+        };
       }
     }
     const messages = record.messages;
@@ -12807,9 +12808,11 @@ function negateScore(score) {
     ...score,
     score: negatedScore,
     verdict: negatedVerdict,
-    reasoning: score.reasoning ? `[Negated] ${score.reasoning} (original score: ${score.score.toFixed(2)})` : `[Negated] Original score: ${score.score.toFixed(2)}`,
-    hits: score.misses,
-    misses: score.hits
+    assertions: score.assertions.map((a) => ({
+      ...a,
+      passed: !a.passed,
+      evidence: a.evidence ? `[Negated] ${a.evidence}` : void 0
+    }))
   };
 }
 
@@ -13267,11 +13270,9 @@ var CodeEvaluator = class {
       }
     }
     const payload = {
-      question: context2.evalCase.question,
       criteria: context2.evalCase.criteria,
       expectedOutput: context2.evalCase.expected_output,
-      referenceAnswer: context2.evalCase.reference_answer,
-      answer: context2.candidate,
+      outputText: context2.candidate,
       output: outputForPayload,
       outputPath,
       guidelineFiles: context2.evalCase.guideline_paths,
@@ -13288,9 +13289,7 @@ var CodeEvaluator = class {
       fileChanges: context2.fileChanges ?? null,
       workspacePath: context2.workspacePath ?? null,
       config: this.config ?? null,
-      // Text convenience accessors (new names, always strings)
       inputText: context2.evalCase.question,
-      outputText: context2.candidate,
       expectedOutputText: context2.evalCase.reference_answer ?? ""
     };
     const inputPayload = JSON.stringify(toSnakeCaseDeep(payload), null, 2);
@@ -13324,9 +13323,13 @@ var CodeEvaluator = class {
       );
       const parsed = parseJsonSafe(stdout);
       const score = clampScore(typeof parsed?.score === "number" ? parsed.score : 0);
-      const hits = Array.isArray(parsed?.hits) ? parsed.hits.filter(isNonEmptyString) : [];
-      const misses = Array.isArray(parsed?.misses) ? parsed.misses.filter(isNonEmptyString) : [];
-      const reasoning = typeof parsed?.reasoning === "string" ? parsed.reasoning : void 0;
+      const assertions = Array.isArray(parsed?.assertions) ? parsed.assertions.filter(
+        (a) => typeof a === "object" && a !== null && typeof a.text === "string"
+      ).map((a) => ({
+        text: String(a.text),
+        passed: Boolean(a.passed),
+        ...typeof a.evidence === "string" ? { evidence: a.evidence } : {}
+      })) : [];
       const details = parsed?.details && typeof parsed.details === "object" && !Array.isArray(parsed.details) ? parsed.details : void 0;
       const proxyUsage = getProxyUsage?.();
       const evaluatorRawRequest = {
@@ -13342,10 +13345,8 @@ var CodeEvaluator = class {
       return {
         score,
         verdict: scoreToVerdict(score),
-        hits,
-        misses,
-        expectedAspectCount: hits.length + misses.length || 1,
-        reasoning,
+        assertions,
+        expectedAspectCount: assertions.length || 1,
         evaluatorRawRequest,
         ...details ? { details } : {},
         tokenUsage: proxyUsage?.tokenUsage
@@ -13356,10 +13357,8 @@ var CodeEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`Code evaluator failed: ${message}`],
+        assertions: [{ text: `Code evaluator failed: ${message}`, passed: false }],
         expectedAspectCount: 1,
-        reasoning: message,
         evaluatorRawRequest: {
           command: this.command,
           ...this.cwd ? { cwd: this.cwd } : {},
@@ -13490,18 +13489,22 @@ Be concise and focused in your evaluation. Provide succinct, specific feedback r
 {{${TEMPLATE_VARIABLES.CRITERIA}}}
 
 [[ ## question ## ]]
-{{${TEMPLATE_VARIABLES.QUESTION}}}
+{{${TEMPLATE_VARIABLES.INPUT_TEXT}}}
 
 [[ ## reference_answer ## ]]
-{{${TEMPLATE_VARIABLES.REFERENCE_ANSWER}}}
+{{${TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT}}}
 
 [[ ## answer ## ]]
-{{${TEMPLATE_VARIABLES.ANSWER}}}`;
+{{${TEMPLATE_VARIABLES.OUTPUT_TEXT}}}`;
 var freeformEvaluationSchema = import_zod4.z.object({
   score: import_zod4.z.number().min(0).max(1).describe("Score between 0.0 and 1.0"),
-  hits: import_zod4.z.array(import_zod4.z.string()).describe("Brief specific achievements").optional(),
-  misses: import_zod4.z.array(import_zod4.z.string()).describe("Brief failures or omissions").optional(),
-  reasoning: import_zod4.z.string().describe("Concise explanation (1-2 sentences)").optional()
+  assertions: import_zod4.z.array(
+    import_zod4.z.object({
+      text: import_zod4.z.string().describe("Brief description of what was checked"),
+      passed: import_zod4.z.boolean().describe("Whether this aspect was satisfied"),
+      evidence: import_zod4.z.string().describe("Concise evidence (1-2 sentences)").optional()
+    })
+  ).describe("Per-aspect evaluation results \u2014 one entry per aspect checked").optional()
 });
 var rubricCheckResultSchema = import_zod4.z.object({
   id: import_zod4.z.string().describe("The ID of the rubric item being checked"),
@@ -13570,12 +13573,8 @@ var LlmGraderEvaluator = class {
         2
       ),
       [TEMPLATE_VARIABLES.OUTPUT]: JSON.stringify(context2.output ?? [], null, 2),
-      [TEMPLATE_VARIABLES.ANSWER]: context2.candidate.trim(),
-      [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (context2.evalCase.reference_answer ?? "").trim(),
       [TEMPLATE_VARIABLES.CRITERIA]: context2.evalCase.criteria.trim(),
-      [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
       [TEMPLATE_VARIABLES.FILE_CHANGES]: context2.fileChanges ?? "",
-      // Text convenience accessors (new names, always strings)
       [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
       [TEMPLATE_VARIABLES.OUTPUT_TEXT]: context2.candidate.trim(),
       [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (context2.evalCase.reference_answer ?? "").trim()
@@ -13603,17 +13602,12 @@ ${context2.fileChanges}`;
         schema: freeformEvaluationSchema
       });
       const score = clampScore(data.score);
-      const hits = Array.isArray(data.hits) ? data.hits.filter(isNonEmptyString).slice(0, 4) : [];
-      const misses = Array.isArray(data.misses) ? data.misses.filter(isNonEmptyString).slice(0, 4) : [];
-      const reasoning = data.reasoning;
-      const expectedAspectCount = Math.max(hits.length + misses.length, 1);
+      const assertions = Array.isArray(data.assertions) ? data.assertions.slice(0, 8) : [];
       return {
         score,
         verdict: scoreToVerdict(score),
-        hits,
-        misses,
-        expectedAspectCount,
-        reasoning,
+        assertions,
+        expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
         tokenUsage
       };
@@ -13624,10 +13618,8 @@ ${context2.fileChanges}`;
       return {
         score: 0,
         verdict: "skip",
-        hits: [],
-        misses: [`Grader parse failure after 3 attempts: ${message}`],
+        assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: 1,
-        reasoning: `Grader parse failure after 3 attempts: ${message}`,
         evaluatorRawRequest
       };
     }
@@ -13657,14 +13649,12 @@ ${context2.fileChanges}`;
         userPrompt: prompt,
         schema: rubricEvaluationSchema
       });
-      const { score, verdict, hits, misses } = calculateRubricScore(data, rubrics);
+      const { score, verdict, assertions } = calculateRubricScore(data, rubrics);
       return {
         score,
         verdict,
-        hits,
-        misses,
+        assertions,
         expectedAspectCount: rubrics.length,
-        reasoning: data.overall_reasoning,
         evaluatorRawRequest,
         tokenUsage
       };
@@ -13675,10 +13665,8 @@ ${context2.fileChanges}`;
       return {
         score: 0,
         verdict: "skip",
-        hits: [],
-        misses: [`Grader parse failure after 3 attempts: ${message}`],
+        assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: rubrics.length,
-        reasoning: `Grader parse failure after 3 attempts: ${message}`,
         evaluatorRawRequest
       };
     }
@@ -13703,14 +13691,12 @@ ${context2.fileChanges}`;
         userPrompt: prompt,
         schema: scoreRangeEvaluationSchema
       });
-      const { score, verdict, hits, misses, details } = calculateScoreRangeResult(data, rubrics);
+      const { score, verdict, assertions, details } = calculateScoreRangeResult(data, rubrics);
       return {
         score,
         verdict,
-        hits,
-        misses,
+        assertions,
         expectedAspectCount: rubrics.length,
-        reasoning: data.overall_reasoning,
         evaluatorRawRequest,
         details,
         tokenUsage
@@ -13722,10 +13708,8 @@ ${context2.fileChanges}`;
       return {
         score: 0,
         verdict: "skip",
-        hits: [],
-        misses: [`Grader parse failure after 3 attempts: ${message}`],
+        assertions: [{ text: `Grader parse failure after 3 attempts: ${message}`, passed: false }],
         expectedAspectCount: rubrics.length,
-        reasoning: `Grader parse failure after 3 attempts: ${message}`,
         evaluatorRawRequest
       };
     }
@@ -13782,8 +13766,7 @@ ${context2.fileChanges}`;
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`llm-grader built-in evaluation failed: ${message}`],
+        assertions: [{ text: `llm-grader built-in evaluation failed: ${message}`, passed: false }],
         expectedAspectCount: 1,
         evaluatorRawRequest,
         details: { mode: "built-in", error: message }
@@ -13833,8 +13816,9 @@ ${context2.fileChanges}`;
         return {
           score: 0,
           verdict: "fail",
-          hits: [],
-          misses: [`llm-grader ${modeLabel} returned no assistant response`],
+          assertions: [
+            { text: `llm-grader ${modeLabel} returned no assistant response`, passed: false }
+          ],
           expectedAspectCount: 1,
           evaluatorRawRequest,
           details: { mode: modeLabel, grader_target: provider.targetName }
@@ -13852,8 +13836,9 @@ ${context2.fileChanges}`;
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`llm-grader ${modeLabel} evaluation failed: ${message}`],
+        assertions: [
+          { text: `llm-grader ${modeLabel} evaluation failed: ${message}`, passed: false }
+        ],
         expectedAspectCount: 1,
         evaluatorRawRequest,
         details: {
@@ -13894,10 +13879,10 @@ ${context2.fileChanges}`;
   buildAgentUserPrompt(context2) {
     const formattedQuestion = context2.promptInputs.question && context2.promptInputs.question.trim().length > 0 ? context2.promptInputs.question : context2.evalCase.question;
     const variables = {
-      [TEMPLATE_VARIABLES.ANSWER]: context2.candidate.trim(),
-      [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (context2.evalCase.reference_answer ?? "").trim(),
       [TEMPLATE_VARIABLES.CRITERIA]: context2.evalCase.criteria.trim(),
-      [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
+      [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
+      [TEMPLATE_VARIABLES.OUTPUT_TEXT]: context2.candidate.trim(),
+      [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (context2.evalCase.reference_answer ?? "").trim(),
       [TEMPLATE_VARIABLES.FILE_CHANGES]: context2.fileChanges ?? ""
     };
     if (this.evaluatorTemplate) {
@@ -13950,10 +13935,10 @@ ${context2.fileChanges}`;
     const rubrics = config?.type === "llm-grader" || config?.type === "llm-judge" ? config.rubrics : void 0;
     if (this.evaluatorTemplate) {
       const variables = {
-        [TEMPLATE_VARIABLES.ANSWER]: context2.candidate.trim(),
-        [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (context2.evalCase.reference_answer ?? "").trim(),
         [TEMPLATE_VARIABLES.CRITERIA]: context2.evalCase.criteria.trim(),
-        [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
+        [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
+        [TEMPLATE_VARIABLES.OUTPUT_TEXT]: context2.candidate.trim(),
+        [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (context2.evalCase.reference_answer ?? "").trim(),
         [TEMPLATE_VARIABLES.FILE_CHANGES]: context2.fileChanges ?? ""
       };
       const customPrompt = substituteVariables(this.evaluatorTemplate, variables);
@@ -14005,29 +13990,24 @@ ${outputSchema}`;
       const parsed = parseJsonFromText(text);
       if (rubrics && rubrics.length > 0) {
         const data2 = rubricEvaluationSchema.parse(parsed);
-        const { score: score2, verdict, hits: hits2, misses: misses2 } = calculateRubricScore(data2, rubrics);
+        const { score: score2, verdict, assertions: assertions2 } = calculateRubricScore(data2, rubrics);
         return {
           score: score2,
           verdict,
-          hits: hits2,
-          misses: misses2,
+          assertions: assertions2,
           expectedAspectCount: rubrics.length,
-          reasoning: data2.overall_reasoning,
           evaluatorRawRequest,
           details
         };
       }
       const data = freeformEvaluationSchema.parse(parsed);
       const score = clampScore(data.score);
-      const hits = Array.isArray(data.hits) ? data.hits.filter(isNonEmptyString).slice(0, 4) : [];
-      const misses = Array.isArray(data.misses) ? data.misses.filter(isNonEmptyString).slice(0, 4) : [];
+      const assertions = Array.isArray(data.assertions) ? data.assertions.slice(0, 8) : [];
       return {
         score,
         verdict: scoreToVerdict(score),
-        hits,
-        misses,
-        expectedAspectCount: Math.max(hits.length + misses.length, 1),
-        reasoning: data.reasoning,
+        assertions,
+        expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
         details
       };
@@ -14035,8 +14015,12 @@ ${outputSchema}`;
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["Failed to parse llm-grader agent response as valid evaluation JSON"],
+        assertions: [
+          {
+            text: "Failed to parse llm-grader agent response as valid evaluation JSON",
+            passed: false
+          }
+        ],
         expectedAspectCount: 1,
         evaluatorRawRequest,
         details
@@ -14165,9 +14149,13 @@ function buildOutputSchema() {
     "",
     "{",
     '  "score": <number between 0.0 and 1.0>,',
-    '  "hits": [<array of strings, max 4 items, brief specific achievements>],',
-    '  "misses": [<array of strings, max 4 items, brief specific failures or omissions, empty if none>],',
-    '  "reasoning": "<string, concise explanation for the score, 1-2 sentences max>"',
+    '  "assertions": [',
+    "    {",
+    '      "text": "<brief description of what was checked>",',
+    '      "passed": <boolean>,',
+    '      "evidence": "<concise evidence, 1-2 sentences, optional>"',
+    "    }",
+    "  ]",
     "}"
   ].join("\n");
 }
@@ -14192,8 +14180,7 @@ function substituteVariables(template, variables) {
 }
 function calculateRubricScore(result, rubrics) {
   const rubricMap = new Map(rubrics.map((rubric) => [rubric.id, rubric]));
-  const hits = [];
-  const misses = [];
+  const assertions = [];
   let totalWeight = 0;
   let earnedWeight = 0;
   let failedRequired = false;
@@ -14203,19 +14190,20 @@ function calculateRubricScore(result, rubrics) {
       continue;
     }
     totalWeight += rubric.weight;
+    assertions.push({
+      text: `[${rubric.id}] ${rubric.outcome}`,
+      passed: check.satisfied,
+      evidence: check.reasoning
+    });
     if (check.satisfied) {
       earnedWeight += rubric.weight;
-      hits.push(`[${rubric.id}] ${rubric.outcome}: ${check.reasoning}`);
-    } else {
-      misses.push(`[${rubric.id}] ${rubric.outcome}: ${check.reasoning}`);
-      if (rubric.required) {
-        failedRequired = true;
-      }
+    } else if (rubric.required) {
+      failedRequired = true;
     }
   }
   const score = totalWeight > 0 ? Math.min(1, Math.max(0, earnedWeight / totalWeight)) : 0;
   const verdict = failedRequired ? "fail" : scoreToVerdict(score);
-  return { score, verdict, hits, misses };
+  return { score, verdict, assertions };
 }
 function buildScoreRangeOutputSchema() {
   return `You are an expert evaluator. Score the candidate answer on each criterion.
@@ -14235,8 +14223,7 @@ Important: The "score" must be an integer from 0 to 10 that falls within one of
 }
 function calculateScoreRangeResult(result, rubrics) {
   const rubricMap = new Map(rubrics.map((rubric) => [rubric.id, rubric]));
-  const hits = [];
-  const misses = [];
+  const assertions = [];
   const rawScores = {};
   let totalWeight = 0;
   let weightedScoreSum = 0;
@@ -14262,24 +14249,22 @@ function calculateScoreRangeResult(result, rubrics) {
     );
     const rangeDescription = matchingRange?.outcome ?? "";
     const criterionLabel = rubric.outcome ?? rubric.id;
-    const reasoningText = check.reasoning ? `: ${check.reasoning}` : "";
-    const scoreInfo = `[${rubric.id}] ${criterionLabel} - Score: ${rawScore}/10 (${rangeDescription})${reasoningText}`;
+    const passed = !(requiredMinScore !== void 0 && rawScore < requiredMinScore) && rawScore >= 7;
     if (requiredMinScore !== void 0 && rawScore < requiredMinScore) {
       failedRequired = true;
-      misses.push(scoreInfo);
-    } else if (rawScore >= 7) {
-      hits.push(scoreInfo);
-    } else {
-      misses.push(scoreInfo);
     }
+    assertions.push({
+      text: `[${rubric.id}] ${criterionLabel} - Score: ${rawScore}/10 (${rangeDescription})`,
+      passed,
+      evidence: check.reasoning
+    });
   }
   const score = totalWeight > 0 ? Math.min(1, Math.max(0, weightedScoreSum / totalWeight)) : 0;
   const verdict = failedRequired ? "fail" : scoreToVerdict(score);
   return {
     score,
     verdict,
-    hits,
-    misses,
+    assertions,
     details: {
       raw_scores: rawScores,
       normalization: "score / 10",
@@ -14455,9 +14440,7 @@ var CompositeEvaluator = class {
     let totalWeight = 0;
     let weightedSum = 0;
     let evaluatedCount = 0;
-    const allHits = [];
-    const allMisses = [];
-    const reasoningParts = [];
+    const allAssertions = [];
     const scores = [];
     for (const member of results) {
       const weight = weights?.[member.id] ?? 1;
@@ -14467,9 +14450,7 @@ var CompositeEvaluator = class {
         score: member.result.score,
         weight,
         verdict: member.result.verdict,
-        hits: [...member.result.hits],
-        misses: [...member.result.misses],
-        reasoning: member.result.reasoning,
+        assertions: [...member.result.assertions],
         evaluatorRawRequest: member.result.evaluatorRawRequest,
         scores: member.result.scores,
         details: member.result.details,
@@ -14481,20 +14462,16 @@ var CompositeEvaluator = class {
       evaluatedCount++;
       totalWeight += weight;
       weightedSum += member.result.score * weight;
-      allHits.push(...member.result.hits.map((h) => `[${member.id}] ${h}`));
-      allMisses.push(...member.result.misses.map((m) => `[${member.id}] ${m}`));
-      if (member.result.reasoning) {
-        reasoningParts.push(`${member.id}: ${member.result.reasoning}`);
-      }
+      allAssertions.push(
+        ...member.result.assertions.map((a) => ({ ...a, text: `[${member.id}] ${a.text}` }))
+      );
     }
     if (evaluatedCount === 0 && results.length > 0) {
       return {
         score: 0,
         verdict: "skip",
-        hits: [],
-        misses: [],
+        assertions: [{ text: "All evaluators skipped (infrastructure failure)", passed: false }],
         expectedAspectCount: 1,
-        reasoning: "All evaluators skipped (infrastructure failure)",
         evaluatorRawRequest: {
           aggregator: "weighted_average",
           ...weights ? { weights } : {}
@@ -14506,10 +14483,8 @@ var CompositeEvaluator = class {
     return {
       score: clampScore(finalScore),
       verdict: scoreToVerdict(finalScore),
-      hits: allHits,
-      misses: allMisses,
-      expectedAspectCount: Math.max(allHits.length + allMisses.length, 1),
-      reasoning: reasoningParts.length > 0 ? reasoningParts.join("; ") : void 0,
+      assertions: allAssertions,
+      expectedAspectCount: allAssertions.length || 1,
       evaluatorRawRequest: {
         aggregator: "weighted_average",
         ...weights ? { weights } : {}
@@ -14519,11 +14494,8 @@ var CompositeEvaluator = class {
   }
   runThreshold(results, threshold) {
     const scores = [];
-    const allHits = [];
-    const allMisses = [];
-    const reasoningParts = [];
+    const allAssertions = [];
     let passingCount = 0;
-    let borderlineCount = 0;
     let evaluatedCount = 0;
     for (const member of results) {
       scores.push({
@@ -14531,9 +14503,7 @@ var CompositeEvaluator = class {
         type: member.type,
         score: member.result.score,
         verdict: member.result.verdict,
-        hits: [...member.result.hits],
-        misses: [...member.result.misses],
-        reasoning: member.result.reasoning,
+        assertions: [...member.result.assertions],
         evaluatorRawRequest: member.result.evaluatorRawRequest,
         scores: member.result.scores,
         details: member.result.details,
@@ -14546,24 +14516,17 @@ var CompositeEvaluator = class {
       const isPassing = member.result.verdict === "pass" || member.result.verdict === "borderline";
       if (isPassing) {
         passingCount++;
-        if (member.result.verdict === "borderline") {
-          borderlineCount++;
-        }
-      }
-      allHits.push(...member.result.hits.map((h) => `[${member.id}] ${h}`));
-      allMisses.push(...member.result.misses.map((m) => `[${member.id}] ${m}`));
-      if (member.result.reasoning) {
-        reasoningParts.push(`${member.id}: ${member.result.reasoning}`);
       }
+      allAssertions.push(
+        ...member.result.assertions.map((a) => ({ ...a, text: `[${member.id}] ${a.text}` }))
+      );
     }
     if (evaluatedCount === 0 && results.length > 0) {
       return {
         score: 0,
         verdict: "skip",
-        hits: [],
-        misses: [],
+        assertions: [{ text: "All evaluators skipped (infrastructure failure)", passed: false }],
         expectedAspectCount: 1,
-        reasoning: "All evaluators skipped (infrastructure failure)",
         evaluatorRawRequest: {
           aggregator: "threshold",
           threshold
@@ -14574,19 +14537,15 @@ var CompositeEvaluator = class {
     const totalCount = evaluatedCount;
     const score = totalCount > 0 ? passingCount / totalCount : 0;
     const pass = score >= threshold;
-    if (pass && borderlineCount > 0) {
-      reasoningParts.push(`Warning: ${borderlineCount} borderline evaluator(s) counted as passing`);
-    }
-    reasoningParts.unshift(
-      `${passingCount}/${totalCount} evaluators passed (threshold: ${threshold})`
-    );
+    allAssertions.unshift({
+      text: `${passingCount}/${totalCount} evaluators passed (threshold: ${threshold})`,
+      passed: pass
+    });
     return {
       score: clampScore(score),
       verdict: pass ? "pass" : "fail",
-      hits: allHits,
-      misses: allMisses,
-      expectedAspectCount: Math.max(allHits.length + allMisses.length, 1),
-      reasoning: reasoningParts.join("; "),
+      assertions: allAssertions,
+      expectedAspectCount: allAssertions.length || 1,
       evaluatorRawRequest: {
         aggregator: "threshold",
         threshold
@@ -14603,9 +14562,7 @@ var CompositeEvaluator = class {
       score: member.result.score,
       weight: weights?.[member.id] ?? 1,
       verdict: member.result.verdict,
-      hits: [...member.result.hits],
-      misses: [...member.result.misses],
-      reasoning: member.result.reasoning,
+      assertions: [...member.result.assertions],
       evaluatorRawRequest: member.result.evaluatorRawRequest,
       scores: member.result.scores,
       details: member.result.details
@@ -14614,17 +14571,19 @@ var CompositeEvaluator = class {
       const stdout = await executeScript(scriptPath, inputPayload, void 0, cwd);
       const parsed = parseJsonSafe(stdout);
       const score = clampScore(typeof parsed?.score === "number" ? parsed.score : 0);
-      const hits = Array.isArray(parsed?.hits) ? parsed.hits.filter(isNonEmptyString) : [];
-      const misses = Array.isArray(parsed?.misses) ? parsed.misses.filter(isNonEmptyString) : [];
-      const reasoning = typeof parsed?.reasoning === "string" ? parsed.reasoning : void 0;
+      const assertions = Array.isArray(parsed?.assertions) ? parsed.assertions.filter(
+        (a) => typeof a === "object" && a !== null && typeof a.text === "string"
+      ).map((a) => ({
+        text: String(a.text),
+        passed: Boolean(a.passed),
+        ...typeof a.evidence === "string" ? { evidence: a.evidence } : {}
+      })) : [];
       const verdict = typeof parsed?.verdict === "string" && (parsed.verdict === "pass" || parsed.verdict === "fail" || parsed.verdict === "borderline") ? parsed.verdict : scoreToVerdict(score);
       return {
         score,
         verdict,
-        hits,
-        misses,
-        expectedAspectCount: hits.length + misses.length || 1,
-        reasoning,
+        assertions,
+        expectedAspectCount: assertions.length || 1,
         evaluatorRawRequest: {
           aggregator: "code-grader",
           script: scriptPath
@@ -14636,10 +14595,8 @@ var CompositeEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`Code aggregator failed: ${message}`],
+        assertions: [{ text: `Code aggregator failed: ${message}`, passed: false }],
         expectedAspectCount: 1,
-        reasoning: message,
         evaluatorRawRequest: {
           aggregator: "code-grader",
           script: scriptPath,
@@ -14661,9 +14618,7 @@ var CompositeEvaluator = class {
       type: member.type,
       score: member.result.score,
       verdict: member.result.verdict,
-      hits: [...member.result.hits],
-      misses: [...member.result.misses],
-      reasoning: member.result.reasoning,
+      assertions: [...member.result.assertions],
       evaluatorRawRequest: member.result.evaluatorRawRequest,
       scores: member.result.scores,
       details: member.result.details
@@ -14687,16 +14642,12 @@ var CompositeEvaluator = class {
         });
         const data2 = freeformEvaluationSchema.parse(parseJsonFromText(text));
         const score2 = clampScore(data2.score);
-        const hits2 = Array.isArray(data2.hits) ? data2.hits.filter(isNonEmptyString).slice(0, 4) : [];
-        const misses2 = Array.isArray(data2.misses) ? data2.misses.filter(isNonEmptyString).slice(0, 4) : [];
-        const reasoning2 = data2.reasoning;
+        const assertions2 = Array.isArray(data2.assertions) ? data2.assertions.slice(0, 8) : [];
         return {
           score: score2,
           verdict: scoreToVerdict(score2),
-          hits: hits2,
-          misses: misses2,
-          expectedAspectCount: Math.max(hits2.length + misses2.length, 1),
-          reasoning: reasoning2,
+          assertions: assertions2,
+          expectedAspectCount: Math.max(assertions2.length, 1),
           evaluatorRawRequest,
           scores
         };
@@ -14711,16 +14662,12 @@ var CompositeEvaluator = class {
         parseJsonFromText(extractLastAssistantContent2(response.output))
       );
       const score = clampScore(data.score);
-      const hits = Array.isArray(data.hits) ? data.hits.filter(isNonEmptyString).slice(0, 4) : [];
-      const misses = Array.isArray(data.misses) ? data.misses.filter(isNonEmptyString).slice(0, 4) : [];
-      const reasoning = data.reasoning;
+      const assertions = Array.isArray(data.assertions) ? data.assertions.slice(0, 8) : [];
       return {
         score,
         verdict: scoreToVerdict(score),
-        hits,
-        misses,
-        expectedAspectCount: Math.max(hits.length + misses.length, 1),
-        reasoning,
+        assertions,
+        expectedAspectCount: Math.max(assertions.length, 1),
         evaluatorRawRequest,
         scores
       };
@@ -14728,8 +14675,7 @@ var CompositeEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [],
+        assertions: [{ text: "LLM aggregator failed", passed: false }],
         expectedAspectCount: 1,
         evaluatorRawRequest,
         scores
@@ -14752,10 +14698,8 @@ var CostEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No cost data available in trace"],
+        assertions: [{ text: "No cost data available in trace", passed: false }],
         expectedAspectCount: 1,
-        reasoning: "Execution cost not reported by provider",
         evaluatorRawRequest: {
           type: "cost",
           budget,
@@ -14769,10 +14713,10 @@ var CostEvaluator = class {
     return {
       score,
       verdict: passed ? "pass" : "fail",
-      hits: passed ? [`Cost ${formatCost(costUsd)} <= ${formatCost(budget)} budget`] : [],
-      misses: passed ? [] : [`Cost ${formatCost(costUsd)} > ${formatCost(budget)} budget`],
+      assertions: [
+        passed ? { text: `Cost ${formatCost(costUsd)} <= ${formatCost(budget)} budget`, passed: true } : { text: `Cost ${formatCost(costUsd)} > ${formatCost(budget)} budget`, passed: false }
+      ],
       expectedAspectCount: 1,
-      reasoning: `Execution cost ${formatCost(costUsd)} (budget: ${formatCost(budget)})`,
       evaluatorRawRequest: {
         type: "cost",
         budget,
@@ -14805,10 +14749,8 @@ var ExecutionMetricsEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No trace summary available"],
+        assertions: [{ text: "No trace summary available", passed: false }],
         expectedAspectCount: 1,
-        reasoning: "Execution metrics not available - no trace summary provided",
         evaluatorRawRequest: {
           type: "execution-metrics",
           config: this.extractConfiguredThresholds(),
@@ -14817,116 +14759,114 @@ var ExecutionMetricsEvaluator = class {
       };
     }
     const narrowedTrace = trace2;
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     const actualMetrics = {};
     if (max_tool_calls !== void 0 && narrowedTrace) {
       const toolCalls = narrowedTrace.eventCount;
       actualMetrics.tool_calls = toolCalls;
       if (toolCalls <= max_tool_calls) {
-        hits.push(`Tool calls ${toolCalls} <= ${max_tool_calls} max`);
+        assertions.push({ text: `Tool calls ${toolCalls} <= ${max_tool_calls} max`, passed: true });
       } else {
-        misses.push(`Tool calls ${toolCalls} > ${max_tool_calls} max`);
+        assertions.push({ text: `Tool calls ${toolCalls} > ${max_tool_calls} max`, passed: false });
       }
     }
     if (max_llm_calls !== void 0 && narrowedTrace) {
       const llmCalls = narrowedTrace.llmCallCount;
       if (llmCalls === void 0) {
-        misses.push("LLM call count data not available");
+        assertions.push({ text: "LLM call count data not available", passed: false });
       } else {
         actualMetrics.llm_calls = llmCalls;
         if (llmCalls <= max_llm_calls) {
-          hits.push(`LLM calls ${llmCalls} <= ${max_llm_calls} max`);
+          assertions.push({ text: `LLM calls ${llmCalls} <= ${max_llm_calls} max`, passed: true });
         } else {
-          misses.push(`LLM calls ${llmCalls} > ${max_llm_calls} max`);
+          assertions.push({ text: `LLM calls ${llmCalls} > ${max_llm_calls} max`, passed: false });
         }
       }
     }
     if (max_tokens !== void 0) {
       if (!tokenUsage) {
-        misses.push("Token usage data not available");
+        assertions.push({ text: "Token usage data not available", passed: false });
       } else {
         const totalTokens = tokenUsage.input + tokenUsage.output;
         actualMetrics.tokens = totalTokens;
         if (totalTokens <= max_tokens) {
-          hits.push(`Total tokens ${totalTokens} <= ${max_tokens} max`);
+          assertions.push({
+            text: `Total tokens ${totalTokens} <= ${max_tokens} max`,
+            passed: true
+          });
         } else {
-          misses.push(`Total tokens ${totalTokens} > ${max_tokens} max`);
+          assertions.push({
+            text: `Total tokens ${totalTokens} > ${max_tokens} max`,
+            passed: false
+          });
         }
       }
     }
     if (max_cost_usd !== void 0) {
       if (costUsd === void 0) {
-        misses.push("Cost data not available");
+        assertions.push({ text: "Cost data not available", passed: false });
       } else {
         actualMetrics.cost_usd = costUsd;
         const formatCost = (n) => `$${n.toFixed(4)}`;
         if (costUsd <= max_cost_usd) {
-          hits.push(`Cost ${formatCost(costUsd)} <= ${formatCost(max_cost_usd)} max`);
+          assertions.push({
+            text: `Cost ${formatCost(costUsd)} <= ${formatCost(max_cost_usd)} max`,
+            passed: true
+          });
         } else {
-          misses.push(`Cost ${formatCost(costUsd)} > ${formatCost(max_cost_usd)} max`);
+          assertions.push({
+            text: `Cost ${formatCost(costUsd)} > ${formatCost(max_cost_usd)} max`,
+            passed: false
+          });
         }
       }
     }
     if (max_duration_ms !== void 0) {
       if (durationMs === void 0) {
-        misses.push("Duration data not available");
+        assertions.push({ text: "Duration data not available", passed: false });
       } else {
         actualMetrics.duration_ms = durationMs;
         if (durationMs <= max_duration_ms) {
-          hits.push(`Duration ${durationMs}ms <= ${max_duration_ms}ms max`);
+          assertions.push({
+            text: `Duration ${durationMs}ms <= ${max_duration_ms}ms max`,
+            passed: true
+          });
         } else {
-          misses.push(`Duration ${durationMs}ms > ${max_duration_ms}ms max`);
+          assertions.push({
+            text: `Duration ${durationMs}ms > ${max_duration_ms}ms max`,
+            passed: false
+          });
         }
       }
     }
     if (target_exploration_ratio !== void 0 && narrowedTrace) {
       const ratio = explorationRatio(narrowedTrace);
       if (ratio === void 0) {
-        misses.push("Exploration ratio not available (no tool calls)");
+        assertions.push({ text: "Exploration ratio not available (no tool calls)", passed: false });
       } else {
         actualMetrics.exploration_ratio = ratio;
         const diff = Math.abs(ratio - target_exploration_ratio);
         if (diff <= exploration_tolerance) {
-          hits.push(
-            `Exploration ratio ${ratio.toFixed(2)} within tolerance of target ${target_exploration_ratio}`
-          );
+          assertions.push({
+            text: `Exploration ratio ${ratio.toFixed(2)} within tolerance of target ${target_exploration_ratio}`,
+            passed: true
+          });
         } else {
-          misses.push(
-            `Exploration ratio ${ratio.toFixed(2)} outside tolerance of target ${target_exploration_ratio} (diff: ${diff.toFixed(2)}, tolerance: ${exploration_tolerance})`
-          );
+          assertions.push({
+            text: `Exploration ratio ${ratio.toFixed(2)} outside tolerance of target ${target_exploration_ratio} (diff: ${diff.toFixed(2)}, tolerance: ${exploration_tolerance})`,
+            passed: false
+          });
         }
       }
     }
-    const totalChecks = hits.length + misses.length;
-    const score = totalChecks > 0 ? hits.length / totalChecks : 0;
-    const reasoningParts = [];
-    if (actualMetrics.tool_calls !== void 0) {
-      reasoningParts.push(`tool_calls=${actualMetrics.tool_calls}`);
-    }
-    if (actualMetrics.llm_calls !== void 0) {
-      reasoningParts.push(`llm_calls=${actualMetrics.llm_calls}`);
-    }
-    if (actualMetrics.tokens !== void 0) {
-      reasoningParts.push(`tokens=${actualMetrics.tokens}`);
-    }
-    if (actualMetrics.cost_usd !== void 0) {
-      reasoningParts.push(`cost=$${actualMetrics.cost_usd.toFixed(4)}`);
-    }
-    if (actualMetrics.duration_ms !== void 0) {
-      reasoningParts.push(`duration=${actualMetrics.duration_ms}ms`);
-    }
-    if (actualMetrics.exploration_ratio !== void 0) {
-      reasoningParts.push(`exploration_ratio=${actualMetrics.exploration_ratio.toFixed(2)}`);
-    }
-    const reasoning = reasoningParts.length > 0 ? `execution-metrics ${reasoningParts.join(", ")}` : "No metrics evaluated";
+    const totalChecks = assertions.length;
+    const passedCount = assertions.filter((a) => a.passed).length;
+    const score = totalChecks > 0 ? passedCount / totalChecks : 0;
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: totalChecks || 1,
-      reasoning,
       evaluatorRawRequest: {
         type: "execution-metrics",
         config: this.extractConfiguredThresholds(),
@@ -15030,10 +14970,8 @@ var FieldAccuracyEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["Failed to parse candidate answer as JSON"],
-        expectedAspectCount: this.config.fields.length,
-        reasoning: "Candidate answer is not valid JSON"
+        assertions: [{ text: "Failed to parse candidate answer as JSON", passed: false }],
+        expectedAspectCount: this.config.fields.length
       };
     }
     const expectedData = this.extractExpectedData(evalCase.expected_output);
@@ -15041,10 +14979,8 @@ var FieldAccuracyEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No expected data found in expected_output"],
-        expectedAspectCount: this.config.fields.length,
-        reasoning: "Could not extract expected data from expected_output"
+        assertions: [{ text: "No expected data found in expected_output", passed: false }],
+        expectedAspectCount: this.config.fields.length
       };
     }
     const fieldResults = [];
@@ -15262,18 +15198,14 @@ var FieldAccuracyEvaluator = class {
    */
   aggregateResults(results) {
     const aggregation = this.config.aggregation ?? "weighted_average";
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     for (const result of results) {
-      if (result.hit) {
-        hits.push(result.message);
-      } else {
-        misses.push(result.message);
-      }
+      assertions.push({ text: result.message, passed: result.hit });
     }
     let score;
     if (aggregation === "all_or_nothing") {
-      score = misses.length === 0 ? 1 : 0;
+      const hasFailed = assertions.some((a) => !a.passed);
+      score = hasFailed ? 0 : 1;
     } else {
       const totalWeight = results.reduce((sum, r) => sum + r.weight, 0);
       if (totalWeight === 0) {
@@ -15283,15 +15215,11 @@ var FieldAccuracyEvaluator = class {
         score = weightedSum / totalWeight;
       }
     }
-    const reasoning = `${hits.length}/${results.length} fields matched`;
     return {
       score: clampScore(score),
       verdict: scoreToVerdict(score),
-      hits: hits.slice(0, 4),
-      // Cap at 4 to keep output concise
-      misses: misses.slice(0, 4),
-      expectedAspectCount: results.length,
-      reasoning
+      assertions,
+      expectedAspectCount: results.length
     };
   }
 };
@@ -15400,10 +15328,8 @@ var LatencyEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No duration data available in trace"],
+        assertions: [{ text: "No duration data available in trace", passed: false }],
         expectedAspectCount: 1,
-        reasoning: "Execution duration not reported by provider",
         evaluatorRawRequest: {
           type: "latency",
           threshold,
@@ -15416,10 +15342,10 @@ var LatencyEvaluator = class {
     return {
       score,
       verdict: passed ? "pass" : "fail",
-      hits: passed ? [`Duration ${durationMs}ms <= ${threshold}ms threshold`] : [],
-      misses: passed ? [] : [`Duration ${durationMs}ms > ${threshold}ms threshold`],
+      assertions: [
+        passed ? { text: `Duration ${durationMs}ms <= ${threshold}ms threshold`, passed: true } : { text: `Duration ${durationMs}ms > ${threshold}ms threshold`, passed: false }
+      ],
       expectedAspectCount: 1,
-      reasoning: `Execution took ${durationMs}ms (threshold: ${threshold}ms)`,
       evaluatorRawRequest: {
         type: "latency",
         threshold,
@@ -15440,7 +15366,10 @@ var COPILOT_MATCHER = {
   skillTools: ["Skill", "skill"],
   skillInputField: "skill",
   readTools: ["Read File", "readFile", "Read", "readTextFile"],
-  readInputField: "file_path"
+  readInputField: "file_path",
+  skillToolPrefixes: ["Using skill: "],
+  readToolPrefixes: ["Viewing "],
+  readInputFields: ["file_path", "path"]
 };
 var PROVIDER_TOOL_SEMANTICS = {
   claude: CLAUDE_MATCHER,
@@ -15482,12 +15411,22 @@ var SkillTriggerEvaluator = class {
           triggered = true;
           evidence = `Skill tool invoked with ${matcher.skillInputField}="${skillArg}"`;
         }
+      } else if (matcher.skillToolPrefixes?.some(
+        (prefix) => firstTool.tool.startsWith(prefix) && firstTool.tool.includes(skillName)
+      )) {
+        triggered = true;
+        evidence = `Skill tool invoked via tool name "${firstTool.tool}"`;
       } else if (matcher.readTools.includes(firstTool.tool)) {
-        const filePath = String(input[matcher.readInputField] ?? "");
+        const filePath = this.readPathFromInput(input, matcher);
         if (filePath.includes(skillName)) {
           triggered = true;
           evidence = `Read tool loaded skill file: ${filePath}`;
         }
+      } else if (matcher.readToolPrefixes?.some(
+        (prefix) => firstTool.tool.startsWith(prefix) && firstTool.tool.includes(skillName)
+      )) {
+        triggered = true;
+        evidence = `Read tool loaded skill file via tool name "${firstTool.tool}"`;
       }
     }
     const pass = triggered === shouldTrigger;
@@ -15495,25 +15434,37 @@ var SkillTriggerEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: [
-          shouldTrigger ? evidence || `Skill "${skillName}" triggered as expected` : `Skill "${skillName}" correctly did not trigger`
+        assertions: [
+          {
+            text: shouldTrigger ? evidence || `Skill "${skillName}" triggered as expected` : `Skill "${skillName}" correctly did not trigger`,
+            passed: true
+          }
         ],
-        misses: [],
-        expectedAspectCount: 1,
-        reasoning: shouldTrigger ? "Skill triggered correctly" : "No false trigger"
+        expectedAspectCount: 1
       };
     }
     return {
       score: 0,
       verdict: "fail",
-      hits: [],
-      misses: [
-        shouldTrigger ? firstTool ? `First tool was "${firstTool.tool}" \u2014 not a skill/read tool for "${skillName}"` : "No tool calls recorded" : evidence || `Skill "${skillName}" triggered unexpectedly`
+      assertions: [
+        {
+          text: shouldTrigger ? firstTool ? `First tool was "${firstTool.tool}" \u2014 not a skill/read tool for "${skillName}"` : "No tool calls recorded" : evidence || `Skill "${skillName}" triggered unexpectedly`,
+          passed: false
+        }
       ],
-      expectedAspectCount: 1,
-      reasoning: shouldTrigger ? `Skill "${skillName}" was not triggered` : "False trigger: skill fired when it should not have"
+      expectedAspectCount: 1
     };
   }
+  readPathFromInput(input, matcher) {
+    const fields = matcher.readInputFields ?? [matcher.readInputField];
+    for (const field of fields) {
+      const value = input[field];
+      if (value !== void 0 && value !== null) {
+        return String(value);
+      }
+    }
+    return "";
+  }
 };
 
 // src/evaluation/evaluators/llm-grader-prompt.ts
@@ -15548,12 +15499,8 @@ function assembleFreeform(evalCase, candidate, promptInputs, fileChanges, evalua
     [TEMPLATE_VARIABLES.INPUT]: JSON.stringify(evalCase.input_segments, null, 2),
     [TEMPLATE_VARIABLES.EXPECTED_OUTPUT]: JSON.stringify(evalCase.expected_output, null, 2),
     [TEMPLATE_VARIABLES.OUTPUT]: JSON.stringify([], null, 2),
-    [TEMPLATE_VARIABLES.ANSWER]: candidate.trim(),
-    [TEMPLATE_VARIABLES.REFERENCE_ANSWER]: (evalCase.reference_answer ?? "").trim(),
     [TEMPLATE_VARIABLES.CRITERIA]: evalCase.criteria.trim(),
-    [TEMPLATE_VARIABLES.QUESTION]: formattedQuestion.trim(),
     [TEMPLATE_VARIABLES.FILE_CHANGES]: fileChanges ?? "",
-    // Text convenience accessors (new names, always strings)
     [TEMPLATE_VARIABLES.INPUT_TEXT]: formattedQuestion.trim(),
     [TEMPLATE_VARIABLES.OUTPUT_TEXT]: candidate.trim(),
     [TEMPLATE_VARIABLES.EXPECTED_OUTPUT_TEXT]: (evalCase.reference_answer ?? "").trim()
@@ -15680,10 +15627,8 @@ var TokenUsageEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No token usage data available in trace"],
+        assertions: [{ text: "No token usage data available in trace", passed: false }],
         expectedAspectCount,
-        reasoning: "Token usage not reported by provider",
         evaluatorRawRequest: {
           type: "token-usage",
           max_total: maxTotal ?? null,
@@ -15697,37 +15642,34 @@ var TokenUsageEvaluator = class {
     const output = usage.output;
     const cached = usage.cached ?? 0;
     const total = input + output + cached;
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     if (typeof maxInput === "number") {
       if (input <= maxInput) {
-        hits.push(`Input tokens ${input} <= ${maxInput}`);
+        assertions.push({ text: `Input tokens ${input} <= ${maxInput}`, passed: true });
       } else {
-        misses.push(`Input tokens ${input} > ${maxInput}`);
+        assertions.push({ text: `Input tokens ${input} > ${maxInput}`, passed: false });
       }
     }
     if (typeof maxOutput === "number") {
       if (output <= maxOutput) {
-        hits.push(`Output tokens ${output} <= ${maxOutput}`);
+        assertions.push({ text: `Output tokens ${output} <= ${maxOutput}`, passed: true });
       } else {
-        misses.push(`Output tokens ${output} > ${maxOutput}`);
+        assertions.push({ text: `Output tokens ${output} > ${maxOutput}`, passed: false });
       }
     }
     if (typeof maxTotal === "number") {
       if (total <= maxTotal) {
-        hits.push(`Total tokens ${total} <= ${maxTotal}`);
+        assertions.push({ text: `Total tokens ${total} <= ${maxTotal}`, passed: true });
       } else {
-        misses.push(`Total tokens ${total} > ${maxTotal}`);
+        assertions.push({ text: `Total tokens ${total} > ${maxTotal}`, passed: false });
       }
     }
-    const passed = misses.length === 0;
+    const passed = assertions.every((a) => a.passed);
     return {
       score: passed ? 1 : 0,
       verdict: passed ? "pass" : "fail",
-      hits,
-      misses,
+      assertions,
       expectedAspectCount,
-      reasoning: `token-usage input=${input}, output=${output}, cached=${cached}, total=${total}`,
       evaluatorRawRequest: {
         type: "token-usage",
         max_total: maxTotal ?? null,
@@ -15827,8 +15769,7 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: ["No trace available for evaluation"],
+        assertions: [{ text: "No trace available for evaluation", passed: false }],
         expectedAspectCount: 1
       };
     }
@@ -15839,8 +15780,7 @@ var ToolTrajectoryEvaluator = class {
           return {
             score: 0,
             verdict: "fail",
-            hits: [],
-            misses: ["No trace available for evaluation"],
+            assertions: [{ text: "No trace available for evaluation", passed: false }],
             expectedAspectCount: 1
           };
         }
@@ -15858,8 +15798,7 @@ var ToolTrajectoryEvaluator = class {
         return {
           score: 0,
           verdict: "fail",
-          hits: [],
-          misses: [`Unknown mode: ${this.config.mode}`],
+          assertions: [{ text: `Unknown mode: ${this.config.mode}`, passed: false }],
           expectedAspectCount: 1
         };
     }
@@ -15908,28 +15847,32 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: ["No tool requirements specified"],
-        misses: [],
+        assertions: [{ text: "No tool requirements specified", passed: true }],
         expectedAspectCount: 0
       };
     }
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     for (const toolName of toolNames) {
       const required = minimums[toolName];
       const actual = summary.toolCallsByName[toolName] ?? 0;
       if (actual >= required) {
-        hits.push(`${toolName}: called ${actual} times (required >=${required})`);
+        assertions.push({
+          text: `${toolName}: called ${actual} times (required >=${required})`,
+          passed: true
+        });
       } else {
-        misses.push(`${toolName}: called ${actual} times (required >=${required})`);
+        assertions.push({
+          text: `${toolName}: called ${actual} times (required >=${required})`,
+          passed: false
+        });
       }
     }
-    const score = hits.length / toolNames.length;
+    const passedCount = assertions.filter((a) => a.passed).length;
+    const score = passedCount / toolNames.length;
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: toolNames.length
     };
   }
@@ -15939,13 +15882,11 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: ["No tool sequence specified"],
-        misses: [],
+        assertions: [{ text: "No tool sequence specified", passed: true }],
         expectedAspectCount: 0
       };
     }
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     const warnings = [];
     let actualIndex = 0;
     let sequenceHits = 0;
@@ -15965,16 +15906,20 @@ var ToolTrajectoryEvaluator = class {
         const actualCall = toolCalls[actualIndex];
         if (actualCall.name === expectedTool) {
           if (argsMatch(expectedItem.args, actualCall.args, mode)) {
-            hits.push(`Found ${expectedTool} at position ${actualIndex}`);
+            assertions.push({
+              text: `Found ${expectedTool} at position ${actualIndex}`,
+              passed: true
+            });
             sequenceHits++;
             matchedCall = actualCall;
             actualIndex++;
             found = true;
             break;
           }
-          misses.push(
-            `Expected ${expectedTool} at position ${i}: tool found at ${actualIndex} but args mismatch`
-          );
+          assertions.push({
+            text: `Expected ${expectedTool} at position ${i}: tool found at ${actualIndex} but args mismatch`,
+            passed: false
+          });
           actualIndex++;
           argsMismatch = true;
           break;
@@ -15982,7 +15927,10 @@ var ToolTrajectoryEvaluator = class {
         actualIndex++;
       }
       if (!found && !argsMismatch) {
-        misses.push(`Expected ${expectedTool} at position ${i}, not found in remaining trace`);
+        assertions.push({
+          text: `Expected ${expectedTool} at position ${i}, not found in remaining trace`,
+          passed: false
+        });
       }
       if (found && matchedCall) {
         const latencyResult = checkLatency(
@@ -15991,10 +15939,10 @@ var ToolTrajectoryEvaluator = class {
           matchedCall.durationMs
         );
         if (latencyResult.status === "pass") {
-          hits.push(latencyResult.message);
+          assertions.push({ text: latencyResult.message, passed: true });
           latencyHits++;
         } else if (latencyResult.status === "fail") {
-          misses.push(latencyResult.message);
+          assertions.push({ text: latencyResult.message, passed: false });
         } else if (latencyResult.message) {
           warnings.push(latencyResult.message);
           latencySkips++;
@@ -16010,8 +15958,7 @@ var ToolTrajectoryEvaluator = class {
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: totalAssertions
     };
   }
@@ -16021,13 +15968,11 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: ["No tool sequence specified"],
-        misses: [],
+        assertions: [{ text: "No tool sequence specified", passed: true }],
         expectedAspectCount: 0
       };
     }
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     const warnings = [];
     let sequenceHits = 0;
     let latencyHits = 0;
@@ -16036,7 +15981,10 @@ var ToolTrajectoryEvaluator = class {
       (item) => item.maxDurationMs !== void 0
     ).length;
     if (toolCalls.length !== expected.length) {
-      misses.push(`Expected ${expected.length} tool calls, got ${toolCalls.length}`);
+      assertions.push({
+        text: `Expected ${expected.length} tool calls, got ${toolCalls.length}`,
+        passed: false
+      });
     }
     const checkLength = Math.min(expected.length, toolCalls.length);
     for (let i = 0; i < checkLength; i++) {
@@ -16048,14 +15996,17 @@ var ToolTrajectoryEvaluator = class {
       let sequenceMatched = false;
       if (actualTool === expectedTool) {
         if (argsMatch(expectedItem.args, actualCall.args, mode)) {
-          hits.push(`Position ${i}: ${expectedTool}`);
+          assertions.push({ text: `Position ${i}: ${expectedTool}`, passed: true });
           sequenceHits++;
           sequenceMatched = true;
         } else {
-          misses.push(`Position ${i}: ${expectedTool} args mismatch`);
+          assertions.push({ text: `Position ${i}: ${expectedTool} args mismatch`, passed: false });
         }
       } else {
-        misses.push(`Position ${i}: expected ${expectedTool}, got ${actualTool}`);
+        assertions.push({
+          text: `Position ${i}: expected ${expectedTool}, got ${actualTool}`,
+          passed: false
+        });
       }
       if (sequenceMatched) {
         const latencyResult = checkLatency(
@@ -16064,10 +16015,10 @@ var ToolTrajectoryEvaluator = class {
           actualCall.durationMs
         );
         if (latencyResult.status === "pass") {
-          hits.push(latencyResult.message);
+          assertions.push({ text: latencyResult.message, passed: true });
           latencyHits++;
         } else if (latencyResult.status === "fail") {
-          misses.push(latencyResult.message);
+          assertions.push({ text: latencyResult.message, passed: false });
         } else if (latencyResult.message) {
           warnings.push(latencyResult.message);
           latencySkips++;
@@ -16075,7 +16026,10 @@ var ToolTrajectoryEvaluator = class {
       }
     }
     for (let i = checkLength; i < expected.length; i++) {
-      misses.push(`Position ${i}: expected ${expected[i].tool}, got nothing`);
+      assertions.push({
+        text: `Position ${i}: expected ${expected[i].tool}, got nothing`,
+        passed: false
+      });
     }
     for (const warning of warnings) {
       console.warn(`[tool-trajectory] ${warning}`);
@@ -16086,8 +16040,7 @@ var ToolTrajectoryEvaluator = class {
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: totalAssertions
     };
   }
@@ -16102,13 +16055,11 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: ["No expected tools specified"],
-        misses: [],
+        assertions: [{ text: "No expected tools specified", passed: true }],
         expectedAspectCount: 0
       };
     }
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     const consumed = /* @__PURE__ */ new Set();
     for (let i = 0; i < expected.length; i++) {
       const expectedItem = expected[i];
@@ -16119,22 +16070,25 @@ var ToolTrajectoryEvaluator = class {
         if (consumed.has(j)) continue;
         const actualCall = toolCalls[j];
         if (actualCall.name === expectedTool && argsMatch(expectedItem.args, actualCall.args, mode)) {
-          hits.push(`Found ${expectedTool} at position ${j}`);
+          assertions.push({ text: `Found ${expectedTool} at position ${j}`, passed: true });
           consumed.add(j);
           found = true;
           break;
         }
       }
       if (!found) {
-        misses.push(`Expected ${expectedTool} not found in actual trajectory`);
+        assertions.push({
+          text: `Expected ${expectedTool} not found in actual trajectory`,
+          passed: false
+        });
       }
     }
-    const score = expected.length > 0 ? hits.length / expected.length : 1;
+    const passedCount = assertions.filter((a) => a.passed).length;
+    const score = expected.length > 0 ? passedCount / expected.length : 1;
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: expected.length
     };
   }
@@ -16150,16 +16104,19 @@ var ToolTrajectoryEvaluator = class {
         return {
           score: 1,
           verdict: "pass",
-          hits: ["No tool calls and no expected tools"],
-          misses: [],
+          assertions: [{ text: "No tool calls and no expected tools", passed: true }],
           expectedAspectCount: 0
         };
       }
       return {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`${toolCalls.length} unexpected tool call(s) with empty allowed list`],
+        assertions: [
+          {
+            text: `${toolCalls.length} unexpected tool call(s) with empty allowed list`,
+            passed: false
+          }
+        ],
         expectedAspectCount: toolCalls.length
       };
     }
@@ -16167,13 +16124,11 @@ var ToolTrajectoryEvaluator = class {
       return {
         score: 1,
         verdict: "pass",
-        hits: ["No actual tool calls (trivially a subset)"],
-        misses: [],
+        assertions: [{ text: "No actual tool calls (trivially a subset)", passed: true }],
         expectedAspectCount: 0
       };
     }
-    const hits = [];
-    const misses = [];
+    const assertions = [];
     for (let i = 0; i < toolCalls.length; i++) {
       const actualCall = toolCalls[i];
       let allowed = false;
@@ -16185,17 +16140,23 @@ var ToolTrajectoryEvaluator = class {
         }
       }
       if (allowed) {
-        hits.push(`Position ${i}: ${actualCall.name} is in allowed set`);
+        assertions.push({
+          text: `Position ${i}: ${actualCall.name} is in allowed set`,
+          passed: true
+        });
       } else {
-        misses.push(`Position ${i}: ${actualCall.name} is not in allowed set`);
+        assertions.push({
+          text: `Position ${i}: ${actualCall.name} is not in allowed set`,
+          passed: false
+        });
       }
     }
-    const score = toolCalls.length > 0 ? hits.length / toolCalls.length : 1;
+    const passedCount = assertions.filter((a) => a.passed).length;
+    const score = toolCalls.length > 0 ? passedCount / toolCalls.length : 1;
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits,
-      misses,
+      assertions,
       expectedAspectCount: toolCalls.length
     };
   }
@@ -16206,8 +16167,12 @@ function runContainsAssertion(output, value) {
   const passed = output.includes(value);
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains "${value}"`] : [],
-    misses: passed ? [] : [`Output does not contain "${value}"`]
+    assertions: [
+      {
+        text: passed ? `Output contains "${value}"` : `Output does not contain "${value}"`,
+        passed
+      }
+    ]
   };
 }
 function runContainsAnyAssertion(output, values) {
@@ -16215,8 +16180,12 @@ function runContainsAnyAssertion(output, values) {
   const passed = matched.length > 0;
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains "${matched[0]}"`] : [],
-    misses: passed ? [] : [`Output does not contain any of: ${values.map((v) => `"${v}"`).join(", ")}`]
+    assertions: [
+      {
+        text: passed ? `Output contains "${matched[0]}"` : `Output does not contain any of: ${values.map((v) => `"${v}"`).join(", ")}`,
+        passed
+      }
+    ]
   };
 }
 function runContainsAllAssertion(output, values) {
@@ -16224,16 +16193,24 @@ function runContainsAllAssertion(output, values) {
   const passed = missing.length === 0;
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains all ${values.length} expected strings`] : [],
-    misses: passed ? [] : [`Output missing: ${missing.map((v) => `"${v}"`).join(", ")}`]
+    assertions: [
+      {
+        text: passed ? `Output contains all ${values.length} expected strings` : `Output missing: ${missing.map((v) => `"${v}"`).join(", ")}`,
+        passed
+      }
+    ]
   };
 }
 function runIcontainsAssertion(output, value) {
   const passed = output.toLowerCase().includes(value.toLowerCase());
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains "${value}" (case-insensitive)`] : [],
-    misses: passed ? [] : [`Output does not contain "${value}" (case-insensitive)`]
+    assertions: [
+      {
+        text: passed ? `Output contains "${value}" (case-insensitive)` : `Output does not contain "${value}" (case-insensitive)`,
+        passed
+      }
+    ]
   };
 }
 function runIcontainsAnyAssertion(output, values) {
@@ -16242,9 +16219,11 @@ function runIcontainsAnyAssertion(output, values) {
   const passed = matched.length > 0;
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains "${matched[0]}" (case-insensitive)`] : [],
-    misses: passed ? [] : [
-      `Output does not contain any of: ${values.map((v) => `"${v}"`).join(", ")} (case-insensitive)`
+    assertions: [
+      {
+        text: passed ? `Output contains "${matched[0]}" (case-insensitive)` : `Output does not contain any of: ${values.map((v) => `"${v}"`).join(", ")} (case-insensitive)`,
+        passed
+      }
     ]
   };
 }
@@ -16254,24 +16233,36 @@ function runIcontainsAllAssertion(output, values) {
   const passed = missing.length === 0;
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output contains all ${values.length} expected strings (case-insensitive)`] : [],
-    misses: passed ? [] : [`Output missing (case-insensitive): ${missing.map((v) => `"${v}"`).join(", ")}`]
+    assertions: [
+      {
+        text: passed ? `Output contains all ${values.length} expected strings (case-insensitive)` : `Output missing (case-insensitive): ${missing.map((v) => `"${v}"`).join(", ")}`,
+        passed
+      }
+    ]
   };
 }
 function runStartsWithAssertion(output, value) {
   const passed = output.trim().startsWith(value.trim());
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output starts with "${value}"`] : [],
-    misses: passed ? [] : [`Output does not start with "${value}"`]
+    assertions: [
+      {
+        text: passed ? `Output starts with "${value}"` : `Output does not start with "${value}"`,
+        passed
+      }
+    ]
   };
 }
 function runEndsWithAssertion(output, value) {
   const passed = output.trim().endsWith(value.trim());
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output ends with "${value}"`] : [],
-    misses: passed ? [] : [`Output does not end with "${value}"`]
+    assertions: [
+      {
+        text: passed ? `Output ends with "${value}"` : `Output does not end with "${value}"`,
+        passed
+      }
+    ]
   };
 }
 function runRegexAssertion(output, pattern, flags) {
@@ -16280,8 +16271,12 @@ function runRegexAssertion(output, pattern, flags) {
   const flagsLabel = flags ? ` (flags: ${flags})` : "";
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output matches pattern /${pattern}/${flags ?? ""}${flagsLabel}`] : [],
-    misses: passed ? [] : [`Output does not match pattern /${pattern}/${flags ?? ""}${flagsLabel}`]
+    assertions: [
+      {
+        text: passed ? `Output matches pattern /${pattern}/${flags ?? ""}${flagsLabel}` : `Output does not match pattern /${pattern}/${flags ?? ""}${flagsLabel}`,
+        passed
+      }
+    ]
   };
 }
 function runIsJsonAssertion(output) {
@@ -16293,16 +16288,24 @@ function runIsJsonAssertion(output) {
   }
   return {
     score: passed ? 1 : 0,
-    hits: passed ? ["Output is valid JSON"] : [],
-    misses: passed ? [] : ["Output is not valid JSON"]
+    assertions: [
+      {
+        text: passed ? "Output is valid JSON" : "Output is not valid JSON",
+        passed
+      }
+    ]
   };
 }
 function runEqualsAssertion(output, value) {
   const passed = output.trim() === value.trim();
   return {
     score: passed ? 1 : 0,
-    hits: passed ? [`Output equals "${value}"`] : [],
-    misses: passed ? [] : [`Output does not equal "${value}"`]
+    assertions: [
+      {
+        text: passed ? `Output equals "${value}"` : `Output does not equal "${value}"`,
+        passed
+      }
+    ]
   };
 }
 
@@ -16515,10 +16518,8 @@ var InlineAssertEvaluator = class {
     return {
       score,
       verdict: scoreToVerdict(score),
-      hits: score >= 0.8 ? [result.name] : [],
-      misses: score < 0.5 ? [result.name] : [],
+      assertions: [{ text: result.name, passed: score >= 0.5 }],
       expectedAspectCount: 1,
-      reasoning: void 0,
       details: result.metadata ? result.metadata : void 0
     };
   }
@@ -16556,11 +16557,9 @@ async function resolveCustomPrompt(promptConfig, context2, timeoutMs) {
 }
 async function executePromptTemplate(script, context2, config, timeoutMs) {
   const payload = {
-    question: context2.evalCase.question,
     criteria: context2.evalCase.criteria,
     expectedOutput: context2.evalCase.expected_output,
-    referenceAnswer: context2.evalCase.reference_answer,
-    answer: context2.candidate,
+    outputText: context2.candidate,
     output: context2.output ?? null,
     guidelineFiles: context2.evalCase.guideline_paths,
     inputFiles: context2.evalCase.file_paths.filter(
@@ -16571,9 +16570,7 @@ async function executePromptTemplate(script, context2, config, timeoutMs) {
     fileChanges: context2.fileChanges ?? null,
     workspacePath: context2.workspacePath ?? null,
     config: config ?? context2.config ?? null,
-    // Text convenience accessors (new names, always strings)
     inputText: context2.evalCase.question,
-    outputText: context2.candidate,
     expectedOutputText: context2.evalCase.reference_answer ?? ""
   };
   const inputJson = JSON.stringify(toSnakeCaseDeep(payload), null, 2);
@@ -16711,9 +16708,7 @@ var containsFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? `Output contains "${c.value}"` : `Output does not contain "${c.value}"`,
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16725,9 +16720,7 @@ var regexFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? `Output matches pattern /${c.value}/${c.flags ?? ""}` : `Output does not match pattern /${c.value}/${c.flags ?? ""}`,
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16738,9 +16731,7 @@ var isJsonFactory = () => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? "Output is valid JSON" : "Output is not valid JSON",
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16752,9 +16743,7 @@ var equalsFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? `Output equals "${c.value}"` : `Output does not equal "${c.value}"`,
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16766,9 +16755,7 @@ var containsAnyFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16780,9 +16767,7 @@ var containsAllFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16794,9 +16779,7 @@ var icontainsFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16808,9 +16791,7 @@ var icontainsAnyFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16822,9 +16803,7 @@ var icontainsAllFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16836,9 +16815,7 @@ var startsWithFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -16850,9 +16827,7 @@ var endsWithFactory = (config) => {
     return {
       score: result.score,
       verdict: result.score === 1 ? "pass" : "fail",
-      hits: result.hits,
-      misses: result.misses,
-      reasoning: result.score === 1 ? result.hits[0] : result.misses[0],
+      assertions: result.assertions,
       expectedAspectCount: 1
     };
   });
@@ -18258,9 +18233,8 @@ async function runEvaluation(options) {
             testId: evalCase.id,
             dataset: evalCase.dataset,
             score: 0,
-            hits: [],
-            misses: [],
-            answer: "",
+            assertions: [],
+            outputText: "",
             target: target.name,
             error: `Suite budget exceeded ($${cumulativeBudgetCost.toFixed(4)} / $${totalBudgetUsd.toFixed(4)})`,
             budgetExceeded: true,
@@ -18295,9 +18269,8 @@ async function runEvaluation(options) {
             testId: evalCase.id,
             dataset: evalCase.dataset,
             score: 0,
-            hits: [],
-            misses: [],
-            answer: "",
+            assertions: [],
+            outputText: "",
             target: target.name,
             error: errorMsg,
             executionStatus: "execution_error",
@@ -19263,11 +19236,9 @@ async function evaluateCandidate(options) {
     dataset: evalCase.dataset,
     conversationId: evalCase.conversation_id,
     score: score.score,
-    hits: score.hits,
-    misses: score.misses,
-    answer: candidate,
+    assertions: score.assertions,
+    outputText: candidate,
     target: target.name,
-    reasoning: score.reasoning,
     tokenUsage,
     costUsd,
     durationMs,
@@ -19441,9 +19412,7 @@ async function runEvaluatorList(options) {
         score: score2.score,
         weight,
         verdict: score2.verdict,
-        hits: score2.hits,
-        misses: score2.misses,
-        reasoning: score2.reasoning,
+        assertions: score2.assertions,
         evaluatorProviderRequest: score2.evaluatorRawRequest,
         details: score2.details,
         scores: mapChildResults(score2.scores),
@@ -19458,10 +19427,10 @@ async function runEvaluatorList(options) {
       const fallbackScore = {
         score: 0,
         verdict: "fail",
-        hits: [],
-        misses: [`Evaluator '${evaluatorConfig.name}' failed: ${message}`],
-        expectedAspectCount: 1,
-        reasoning: message
+        assertions: [
+          { text: `Evaluator '${evaluatorConfig.name}' failed: ${message}`, passed: false }
+        ],
+        expectedAspectCount: 1
       };
       const weight = evaluatorConfig.weight ?? 1;
       scored.push({
@@ -19477,9 +19446,12 @@ async function runEvaluatorList(options) {
         score: 0,
         weight,
         verdict: "fail",
-        hits: [],
-        misses: [`Evaluator '${evaluatorConfig.name ?? "unknown"}' failed: ${message}`],
-        reasoning: message,
+        assertions: [
+          {
+            text: `Evaluator '${evaluatorConfig.name ?? "unknown"}' failed: ${message}`,
+            passed: false
+          }
+        ],
         durationMs: endedAt.getTime() - startedAt.getTime(),
         startedAt: startedAt.toISOString(),
         endedAt: endedAt.toISOString()
@@ -19495,9 +19467,7 @@ async function runEvaluatorList(options) {
           ...scores[lastScoresIdx],
           score: negated.score,
           verdict: negated.verdict,
-          hits: [...negated.hits],
-          misses: [...negated.misses],
-          reasoning: negated.reasoning
+          assertions: [...negated.assertions]
         };
       }
     }
@@ -19512,21 +19482,13 @@ async function runEvaluatorList(options) {
   const aggregateScore = hasRequiredFailure ? 0 : scorable.length > 0 ? computeWeightedMean(
     scorable.map((entry) => ({ score: entry.score.score, weight: entry.weight }))
   ) : 0;
-  const hits = scored.flatMap((entry) => entry.score.hits);
-  const misses = scored.flatMap((entry) => entry.score.misses);
-  const expectedAspectCount = scored.reduce(
-    (total, entry) => total + (entry.score.expectedAspectCount ?? 0),
-    0
-  );
-  const reasoningParts = scored.map((entry) => entry.score.reasoning ? `${entry.name}: ${entry.score.reasoning}` : void 0).filter(isNonEmptyString);
-  const reasoning = reasoningParts.length > 0 ? reasoningParts.join(" | ") : void 0;
+  const assertions = scored.flatMap((entry) => entry.score.assertions);
+  const expectedAspectCount = assertions.length || 1;
   const score = {
     score: aggregateScore,
     verdict: scoreToVerdict(aggregateScore),
-    hits,
-    misses,
-    expectedAspectCount,
-    reasoning
+    assertions,
+    expectedAspectCount
   };
   return { score, scores };
 }
@@ -19630,9 +19592,8 @@ function buildErrorResult(evalCase, targetName, timestamp, error, promptInputs,
     dataset: evalCase.dataset,
     conversationId: evalCase.conversation_id,
     score: 0,
-    hits: [],
-    misses: [`Error: ${message}`],
-    answer: `Error occurred: ${message}`,
+    assertions: [{ text: `Error: ${message}`, passed: false }],
+    outputText: `Error occurred: ${message}`,
     target: targetName,
     requests,
     input,
@@ -19741,9 +19702,7 @@ function mapChildResults(children) {
     score: child.score,
     weight: child.weight,
     verdict: child.verdict,
-    hits: child.hits,
-    misses: child.misses,
-    reasoning: child.reasoning,
+    assertions: child.assertions,
     evaluatorProviderRequest: child.evaluatorRawRequest,
     scores: mapChildResults(child.scores),
     details: child.details,
@@ -20172,7 +20131,7 @@ function shouldSkipCacheForTemperature(targetConfig) {
 
 // src/evaluation/baseline.ts
 var STRIPPED_TOP_LEVEL_FIELDS = /* @__PURE__ */ new Set([
-  "answer",
+  "outputText",
   "requests",
   "trace",
   "workspacePath",
@@ -20346,7 +20305,7 @@ var OtelTraceExporter = class {
         rootSpan.setAttribute("agentv.target", result.target);
         if (result.dataset) rootSpan.setAttribute("agentv.dataset", result.dataset);
         rootSpan.setAttribute("agentv.score", result.score);
-        if (captureContent) rootSpan.setAttribute("agentv.answer", result.answer);
+        if (captureContent) rootSpan.setAttribute("agentv.output_text", result.outputText);
         if (result.durationMs != null)
           rootSpan.setAttribute("agentv.trace.duration_ms", result.durationMs);
         if (result.costUsd != null) rootSpan.setAttribute("agentv.trace.cost_usd", result.costUsd);
@@ -20713,7 +20672,6 @@ function createAgentKernel() {
   freeformEvaluationSchema,
   generateRubrics,
   getAgentvHome,
-  getHitCount,
   getOutputFilenames,
   getSubagentsRoot,
   getTraceStateRoot,