@tangle-network/agent-eval 0.119.0 → 0.120.0

package/dist/index.js

@@ -9,12 +9,12 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-XJ7JVCHB.js";
+} from "./chunk-3A246TSA.js";
 import {
   BENCHMARK_SPLIT_SEED,
   benchmarks_exports,
   deterministicSplit
-} from "./chunk-D7AEXSM5.js";
+} from "./chunk-PICTDURQ.js";
 import {
   DEFAULT_RULES,
   buildTrajectory,
@@ -43,17 +43,15 @@ import {
   agentProfileHash,
   agentProfileId,
   agentProfileModelId,
-  comparePairedArms,
   completionVerdict,
   createLlmCorrectnessChecker,
   createTokenRecallChecker,
   expandProfileAxes,
   extractProducedState,
   harnessAxisOf,
-  pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-F6YUH3L4.js";
+} from "./chunk-JN2FCO5W.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -67,6 +65,7 @@ import {
   campaignMeanComposite,
   codeExecutionJudge,
   coherenceJudge,
+  comparePairedArms,
   costReceiptFromTCloud,
   createCustomJudge,
   createDomainExpertJudge,
@@ -79,6 +78,7 @@ import {
   heldOutGate,
   llmJudge,
   maximumChargeForTCloudRequest,
+  pairArms,
   paretoFrontier,
   paretoFrontierWithCrowding,
   parseReflectionResponse,
@@ -91,17 +91,19 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-RLCJQ6VZ.js";
+} from "./chunk-32BZXMSO.js";
 import {
   BackendIntegrityError,
+  assertRealAgentReceipts,
   assertRealBackend,
   cachedJudge,
   canonicalJson,
   contentHash,
   fileVerdictCache,
   inMemoryVerdictCache,
+  summarizeAgentReceiptIntegrity,
   summarizeBackendIntegrity
-} from "./chunk-PAHNGS65.js";
+} from "./chunk-XDIRG3TO.js";
 import {
   DEFAULT_COMPLEXITY_WEIGHTS,
   FindingsStore,
@@ -114,12 +116,12 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-LMJ2TGWJ.js";
+} from "./chunk-JM2SKQMS.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics,
   createChatClient
-} from "./chunk-GERDAIAL.js";
+} from "./chunk-QWMPPZ3X.js";
 import "./chunk-HHWE3POT.js";
 import {
   Mutex
@@ -155,7 +157,7 @@ import {
   scorePolicyEditReadiness,
   validatePolicyEdit,
   validatePolicyEditCandidateRecord
-} from "./chunk-7A4LIMMY.js";
+} from "./chunk-S5TT5R3L.js";
 import {
   allCriticalPassed,
   controlFailureClassFromVerification,
@@ -304,7 +306,7 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-ZYHJNKI3.js";
+} from "./chunk-ULOKLHIQ.js";
 import "./chunk-H5UD2323.js";
 import {
   extractUsage,
@@ -325,9 +327,8 @@ import {
 import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
-  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   analyzeTraces
-} from "./chunk-FIUFRXP3.js";
+} from "./chunk-WW2A73HW.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -352,7 +353,7 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-6QIM2EAP.js";
+} from "./chunk-PXD6ZFNY.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
@@ -4338,7 +4339,7 @@ function renderSteeringText(bundle) {
 }
 
 // src/steering-optimizer.ts
-import { AxGEPA, ai, ax } from "@ax-llm/ax";
+import { AxGEPA, AxSignature, ax } from "@ax-llm/ax";
 var PairwiseSteeringOptimizer = class {
   optimize(rows, config = {}) {
     const ranked = rankRows(rows, config.weights);
@@ -4370,7 +4371,8 @@ var AxGepaSteeringOptimizer = class {
       };
     }
     const signature = `task:string, split:string, seedPreview:string -> variantId:class "${variantIds.join(", ")}", rationale:string`;
-    const selector = ax(signature, {
+    const selectorSignature = AxSignature.from(signature);
+    const selector = ax(selectorSignature, {
       description: "Choose the best steering bundle variant for an autopilot task."
     });
     const shuffled = seededShuffle(byScenario, signature);
@@ -4399,15 +4401,11 @@ var AxGepaSteeringOptimizer = class {
       earlyStoppingTrials: 3,
       sampleCount: 1
     });
-    const compiled = await optimizer.compile(
-      selector,
-      train,
-      (input) => input.prediction?.variantId === input.example?.variantId ? 1 : 0,
-      {
-        validationExamples: validation,
-        maxMetricCalls: 64
-      }
-    );
+    const metric = ({ prediction, example }) => stringField2(prediction, "variantId") === stringField2(example, "variantId") ? 1 : 0;
+    const compiled = await optimizer.compile(selector, train, metric, {
+      validationExamples: validation,
+      maxMetricCalls: 64
+    });
     if (compiled.optimizedProgram !== void 0) {
       selector.applyOptimization(compiled.optimizedProgram);
     }
@@ -4463,12 +4461,17 @@ function collapseScenarioWinners(rows, weights) {
   });
 }
 function createAxService(provider, apiKey, model) {
-  return ai({
-    name: provider,
+  return createAnalystAi({
+    provider,
     apiKey,
-    config: { model }
+    model
   });
 }
+function stringField2(value, field) {
+  if (!value || typeof value !== "object") return void 0;
+  const candidate = value[field];
+  return typeof candidate === "string" ? candidate : void 0;
+}
 function seededShuffle(items, seed) {
   const rng = mulberry322(hashString(seed));
   const out = [...items];
@@ -5972,9 +5975,9 @@ function statusAdvanced(key, progression) {
     description: `"${key}" progressed along ${progression.join("\u2192")}`,
     score: ({ before, after }) => {
       const bi = progression.indexOf(String(before[key]));
-      const ai2 = progression.indexOf(String(after[key]));
-      if (bi === -1 || ai2 === -1) return 0;
-      return ai2 >= bi ? 1 : 0;
+      const ai = progression.indexOf(String(after[key]));
+      if (bi === -1 || ai === -1) return 0;
+      return ai >= bi ? 1 : 0;
     }
   };
 }
@@ -10036,12 +10039,11 @@ async function tracedAnalyzeTraces(input, options, traceOpts) {
     attributes: {
       "analyst.question_length": input.question.length,
       "analyst.max_turns": options.maxTurns ?? 12,
-      "analyst.max_depth": options.maxDepth ?? 1,
+      "analyst.max_subqueries": options.maxSubqueries ?? 4,
       "eval.phase": "analyst"
     }
   });
   const originalOnTurn = options.onTurn;
-  const turnSpanIds = [];
   const wrappedOptions = {
     ...options,
     onTurn: async (turn) => {
@@ -10050,6 +10052,7 @@ async function tracedAnalyzeTraces(input, options, traceOpts) {
         name: `analyst:turn-${turn.turn}`,
         parentSpanId: parentSpan.span.spanId,
         attributes: {
+          "analyst.stage": turn.stage,
           "analyst.turn": turn.turn,
           "analyst.is_error": turn.isError,
           "analyst.code_length": turn.code.length,
@@ -10057,7 +10060,6 @@ async function tracedAnalyzeTraces(input, options, traceOpts) {
           "eval.phase": "analyst"
         }
       });
-      turnSpanIds.push(turnSpan.span.spanId);
       if (turn.isError) {
         await turnSpan.fail("Turn produced an error");
       } else {
@@ -11664,7 +11666,6 @@ export {
   TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
-  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TRACE_SCHEMA_VERSION,
   TokenCounter,
@@ -11707,6 +11708,7 @@ export {
   assertModelsServed,
   assertNoHiddenLeak,
   assertProductBenchmarkRun,
+  assertRealAgentReceipts,
   assertRealBackend,
   assertReleaseConfidence,
   assertRunAgentProfileCell,
@@ -12109,6 +12111,7 @@ export {
   stringField,
   stripFencedJson,
   subjectiveEval,
+  summarizeAgentReceiptIntegrity,
   summarizeBackendIntegrity,
   summarizeHarnessResults,
   summarizePrReviewBenchmark,