@tangle-network/agent-eval 0.94.0 → 0.95.1

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-2KNZHH3P.js";
+} from "./chunk-Z6L6YSU6.js";
 import {
   classifyEuAiRisk,
   euAiActReport,
@@ -35,13 +35,15 @@ import {
 } from "./chunk-STGVSCDH.js";
 import {
   agentProfileHash,
+  agentProfileId,
+  agentProfileModelId,
   completionVerdict,
   createLlmCorrectnessChecker,
   createTokenRecallChecker,
   extractProducedState,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-S6OZEZQK.js";
+} from "./chunk-HMA63UEO.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -51,7 +53,7 @@ import {
   campaignMeanComposite,
   crowdingDistance,
   dominates,
-  gepaDriver,
+  gepaProposer,
   hashScenarios,
   heldOutGate,
   paretoFrontier,
@@ -65,12 +67,12 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-EGPMSBEZ.js";
+} from "./chunk-7QTQKIDD.js";
 import {
   BackendIntegrityError,
   assertRealBackend,
   summarizeBackendIntegrity
-} from "./chunk-2K6UUZ7P.js";
+} from "./chunk-2T4EZACH.js";
 import {
   MODEL_PRICING,
   MetricsCollector,
@@ -114,8 +116,6 @@ import {
   DEFAULT_COMPLEXITY_WEIGHTS,
   DEFAULT_RUN_SCORE_WEIGHTS,
   FindingsStore,
-  LockedJsonlAppender,
-  Mutex,
   RunCritic,
   SEMANTIC_CONCEPT_JUDGE_VERSION,
   SKILL_USAGE_ANALYST,
@@ -127,13 +127,16 @@ import {
   createSemanticConceptJudge,
   defaultIsMaterial,
   diffFindings,
-  resetLockedAppendersForTesting,
   runSemanticConceptJudge
-} from "./chunk-MPQWFX6Y.js";
+} from "./chunk-PSWWQXHF.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics
-} from "./chunk-MIFZUPEK.js";
+} from "./chunk-AQ5WQAIV.js";
+import {
+  LockedJsonlAppender,
+  Mutex
+} from "./chunk-DJWX3GVS.js";
 import {
   AnalystRegistry,
   DEFAULT_TRACE_ANALYST_KINDS,
@@ -143,7 +146,7 @@ import {
   KNOWLEDGE_POISONING_KIND_SPEC,
   createTraceAnalystKind,
   renderPriorFindings
-} from "./chunk-Q5LIB7BC.js";
+} from "./chunk-S4SYLDFX.js";
 import {
   controlFailureClassFromVerification,
   controlRunToRunRecord,
@@ -154,7 +157,7 @@ import {
   runProposeReview,
   runProposeReviewAsControlLoop,
   scoreFromEvals
-} from "./chunk-E4GH6USR.js";
+} from "./chunk-NZEQVRH5.js";
 import {
   allCriticalPassed,
   errorStreakDetector,
@@ -176,7 +179,7 @@ import {
 } from "./chunk-QTMYW64F.js";
 import {
   runEvalCampaign
-} from "./chunk-KW53MSA5.js";
+} from "./chunk-X74V6ESX.js";
 import {
   LlmCallError,
   LlmClient,
@@ -276,7 +279,13 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-CTBHKLEU.js";
+} from "./chunk-77T4STFI.js";
+import {
+  TRACE_ANALYST_ACTOR_DESCRIPTION,
+  TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
+  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
+  analyzeTraces
+} from "./chunk-IZCEK2HR.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -302,16 +311,24 @@ import {
   isSandboxSpan,
   isToolSpan
 } from "./chunk-5BKGXME7.js";
-import {
-  TRACE_ANALYST_ACTOR_DESCRIPTION,
-  TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
-  TRACE_ANALYST_SUBAGENT_DESCRIPTION,
-  analyzeTraces
-} from "./chunk-TBDR6PAI.js";
 import {
   DEFAULT_TRACE_ANALYST_BUDGETS,
+  LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_COST_ATTR_KEYS,
+  LLM_COST_USD,
+  LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
+  LLM_MODEL_NAME,
+  LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
+  OPENINFERENCE_SPAN_KIND,
   OtlpFileTraceStore,
+  SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
+  TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TraceFileMissingError,
   TraceNotFoundError,
@@ -325,8 +342,9 @@ import {
   projectOtlpFlatLine,
   readOtlpStatus,
   stringField,
-  traceAnalystFunctionGroup
-} from "./chunk-QMUEXQJS.js";
+  traceAnalystFunctionGroup,
+  traceSpanKindToOpenInferenceKind
+} from "./chunk-YBIGNSCZ.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
@@ -346,8 +364,8 @@ import {
   agentProfileCellHashMaterial,
   agentProfileCellKey,
   assertRunAgentProfileCell,
+  buildAgentInterfaceProfileCell,
   buildAgentProfileCell,
-  buildSandboxAgentProfileCell,
   groupRunsByAgentProfileCell,
   isRunRecord,
   parseRunRecordSafe,
@@ -357,7 +375,7 @@ import {
   validateAgentProfileCell,
   validateRunRecord,
   verifyAgentProfileCell
-} from "./chunk-KWRRMR3J.js";
+} from "./chunk-LO6IOIJ2.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
@@ -5136,6 +5154,7 @@ function aggregatePerDimension(runs) {
 }
 function recordRuns(runs, opts) {
   const profileHash = agentProfileHash(opts.profile);
+  const model = agentProfileModelId(opts.profile);
   const timestamp = opts.timestamp ?? (/* @__PURE__ */ new Date()).toISOString();
   const byScenario = /* @__PURE__ */ new Map();
   for (const run of runs) {
@@ -5162,7 +5181,7 @@ function recordRuns(runs, opts) {
     lines.push({
       scenarioId,
       profileHash,
-      model: opts.profile.model,
+      model,
       profile: opts.profile,
       entry
     });
@@ -9578,7 +9597,7 @@ async function runDistillation(opts) {
   const runDir = opts.runDir ?? `.evolve/distillation/${Date.now()}`;
   const studentTemperature = opts.studentTemperature ?? 0;
   const studentMaxTokens = opts.studentMaxTokens ?? 1024;
-  const driver = gepaDriver({
+  const proposer = gepaProposer({
     llm: opts.reflectionLlm,
     model: opts.optimizerModel,
     target: "a cheap single-shot analyst system prompt that reproduces an expensive workflow gold verdict",
@@ -9594,7 +9613,7 @@ async function runDistillation(opts) {
     scenarios: opts.train,
     holdoutScenarios: opts.holdout,
     judges: [opts.judge],
-    driver,
+    proposer,
     gate,
     autoOnPromote: "none",
     // the loop NEVER opens a PR — the caller decides
@@ -9602,8 +9621,8 @@ async function runDistillation(opts) {
     maxGenerations: opts.maxGenerations ?? 3,
     reps: opts.reps ?? 1,
     runDir,
-    // The student spends tokens; tracing must stay on (the driver is wired and
-    // runImprovementLoop refuses tracing='off' with a driver).
+    // The student spends tokens; tracing must stay on (the proposer is wired and
+    // runImprovementLoop refuses tracing='off' with a proposer).
     tracing: "on",
     dispatchWithSurface: async (surface, scenario, ctx) => {
       const prompt = render({
@@ -10142,6 +10161,16 @@ export {
   JudgeRunner,
   KNOWLEDGE_GAP_KIND_SPEC,
   KNOWLEDGE_POISONING_KIND_SPEC,
+  LLM_CACHED_TOKENS,
+  LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_COST_ATTR_KEYS,
+  LLM_COST_USD,
+  LLM_INPUT_TOKENS,
+  LLM_INPUT_TOKEN_ATTR_KEYS,
+  LLM_MODEL_ATTR_KEYS,
+  LLM_MODEL_NAME,
+  LLM_OUTPUT_TOKENS,
+  LLM_OUTPUT_TOKEN_ATTR_KEYS,
   LlmCallError,
   LlmClient,
   LlmRouteAssertionError,
@@ -10154,6 +10183,7 @@ export {
   NoopRawProviderSink,
   NoopResearcher,
   NotFoundError,
+  OPENINFERENCE_SPAN_KIND,
   OTEL_AGENT_EVAL_SCOPE,
   OtlpFileTraceStore,
   PairwiseSteeringOptimizer,
@@ -10169,6 +10199,7 @@ export {
   RunRecordValidationError,
   SEMANTIC_CONCEPT_JUDGE_VERSION,
   SKILL_USAGE_ANALYST,
+  SPAN_KIND_ATTR_KEYS,
   SandboxHarness,
   ScenarioRegistry,
   SeatUnsetError,
@@ -10176,6 +10207,8 @@ export {
   SkillUsageAnalyst,
   SpanNotFoundError,
   SubprocessSandboxDriver,
+  TOOL_NAME,
+  TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
@@ -10196,6 +10229,8 @@ export {
   agentProfileCellHashMaterial,
   agentProfileCellKey,
   agentProfileHash,
+  agentProfileId,
+  agentProfileModelId,
   aggregateJudgeVerdicts,
   aggregateLlm,
   aggregatePrReviewScore,
@@ -10230,13 +10265,13 @@ export {
   blockingKnowledgeEval,
   bonferroni,
   bootstrapCi,
+  buildAgentInterfaceProfileCell,
   buildAgentProfileCell,
   buildAgreementJudge,
   buildDefaultAnalystRegistry,
   buildDriverSystemPrompt,
   buildReflectionPrompt,
   buildReviewerPrompt,
-  buildSandboxAgentProfileCell,
   buildTraceAnalystTools,
   buildTraceInsightContext,
   buildTraceInsightPrompt,
@@ -10515,7 +10550,6 @@ export {
   requireAgentProfileCell,
   requiredSampleSize,
   researchReport,
-  resetLockedAppendersForTesting,
   resolveModelPricing,
   resolveSeat,
   roundTripRunRecord,
@@ -10593,6 +10627,7 @@ export {
   traceContract,
   traceJudge,
   traceJudgeEnsemble,
+  traceSpanKindToOpenInferenceKind,
   tracedAnalyzeTraces,
   typoMutator,
   urlContains,