@tangle-network/agent-eval 0.117.0 → 0.118.0

package/dist/index.js

@@ -9,7 +9,7 @@ import {
   checkBehavioralCanary,
   checkCanaries,
   runBehavioralCanaries
-} from "./chunk-FQNLDL4D.js";
+} from "./chunk-SVH2ANFD.js";
 import {
   BENCHMARK_SPLIT_SEED,
   benchmarks_exports,
@@ -23,7 +23,7 @@ import {
   computeToolUseMetrics,
   iqr,
   welchsTTest
-} from "./chunk-ODVOOEWQ.js";
+} from "./chunk-QBRSJK47.js";
 import {
   analyzeSeries
 } from "./chunk-BOD4O7OF.js";
@@ -36,7 +36,7 @@ import {
   pytestTestParser,
   runTestGradedScenario,
   vitestTestParser
-} from "./chunk-HZHNRYHK.js";
+} from "./chunk-K6N6XJJX.js";
 import {
   CODING_HARNESSES,
   HARNESS_NATIVE_MODEL,
@@ -53,7 +53,7 @@ import {
   pairArms,
   parseCorrectnessResponse,
   verifyCompletion
-} from "./chunk-ZUXV7UWZ.js";
+} from "./chunk-Q442S5AS.js";
 import {
   DEFAULT_MUTATION_PRIMITIVES,
   DEFAULT_RED_TEAM_CORPUS,
@@ -91,7 +91,7 @@ import {
   scoreRedTeamOutput,
   surfaceContentHash,
   toolNamesForRun
-} from "./chunk-HQPHZGL6.js";
+} from "./chunk-YLMUS4MM.js";
 import {
   BackendIntegrityError,
   assertRealBackend,
@@ -115,12 +115,12 @@ import {
   defaultIsMaterial,
   diffFindings,
   runSemanticConceptJudge
-} from "./chunk-CCZIVI3F.js";
+} from "./chunk-BFW56GTT.js";
 import {
   buildDefaultAnalystRegistry,
   computeTraceMetrics,
   createChatClient
-} from "./chunk-VF3XSYTI.js";
+} from "./chunk-33JA4TFA.js";
 import "./chunk-HHWE3POT.js";
 import {
   Mutex
@@ -154,7 +154,7 @@ import {
   scorePolicyEditReadiness,
   validatePolicyEdit,
   validatePolicyEditCandidateRecord
-} from "./chunk-MGEHEHSN.js";
+} from "./chunk-OIIMMLRB.js";
 import {
   allCriticalPassed,
   controlFailureClassFromVerification,
@@ -175,7 +175,7 @@ import {
   stopOnNoProgress,
   stopOnRepeatedAction,
   subjectiveEval
-} from "./chunk-DXZRATT5.js";
+} from "./chunk-OYZAPX5G.js";
 import {
   assertReleaseConfidence,
   bootstrapCi,
@@ -185,8 +185,24 @@ import {
 } from "./chunk-MOXWMGPC.js";
 import {
   runEvalCampaign
-} from "./chunk-GQCZRZ7L.js";
+} from "./chunk-U5CHZ5M3.js";
 import "./chunk-ARU2PZFM.js";
+import {
+  LlmCallError,
+  LlmClient,
+  LlmResponseError,
+  LlmRouteAssertionError,
+  assertLlmRoute,
+  backoffMs,
+  callLlm,
+  callLlmJson,
+  costReceiptFromLlm,
+  costReceiptFromLlmError,
+  isTransientLlmError,
+  maximumChargeForLlmRequest,
+  probeLlm,
+  stripFencedJson
+} from "./chunk-NJC7U437.js";
 import {
   evaluateInterimReleaseConfidence,
   pairedEvalueSequence
@@ -238,6 +254,26 @@ import {
   wilcoxonSignedRank,
   wilson
 } from "./chunk-PJQFMIOX.js";
+import {
+  CostAccountingIncompleteError,
+  CostCallConflictError,
+  CostCeilingReachedError,
+  CostLedger,
+  CostLedgerPersistenceError,
+  CostReceiptCaptureError,
+  CostReservationExceededError,
+  costForUsage,
+  modelPriceKey
+} from "./chunk-VCTY3W6J.js";
+import {
+  MODEL_PRICING,
+  MetricsCollector,
+  TokenCounter,
+  estimateCost,
+  estimateTokens,
+  isModelPriced,
+  resolveModelPricing
+} from "./chunk-VI2UW6B6.js";
 import {
   FileSystemTraceStore,
   InMemoryTraceStore,
@@ -255,9 +291,6 @@ import {
   describeTraceInsightScope,
   domainEvidencePattern,
   exportRunAsOtlp,
-  extractUsage,
-  extractUsageFromResponse,
-  extractUsageFromSse,
   flattenOtlpExportToNdjson,
   inferDomainKeywords,
   iterateRawCalls,
@@ -268,31 +301,30 @@ import {
   scoreTraceInsightReadiness,
   tokenizeDomainWords,
   traceAnalystOnRunComplete
-} from "./chunk-YZPO4UHR.js";
+} from "./chunk-DP3BBMYJ.js";
+import "./chunk-KSDQVPLR.js";
+import {
+  extractUsage,
+  extractUsageFromResponse,
+  extractUsageFromSse
+} from "./chunk-PXE2VKMX.js";
 import {
-  FAILURE_CLASSES,
-  TRACE_SCHEMA_VERSION,
   aggregateLlm,
   argHash,
   groupBy,
   hasCapturedToolArgs,
-  isJudgeSpan,
-  isLlmSpan,
-  isRetrievalSpan,
-  isSandboxSpan,
-  isToolSpan,
   judgeSpans,
   llmSpans,
   runFailureClass,
   runsForScenario,
   toolSpans
-} from "./chunk-LQUTGLOZ.js";
+} from "./chunk-ZET2UAYW.js";
 import {
   TRACE_ANALYST_ACTOR_DESCRIPTION,
   TRACE_ANALYST_ACTOR_DESCRIPTION_VERSION,
   TRACE_ANALYST_SUBAGENT_DESCRIPTION,
   analyzeTraces
-} from "./chunk-4JLWXDYA.js";
+} from "./chunk-3EHHMC6E.js";
 import {
   DEFAULT_REDACTION_RULES,
   REDACTION_VERSION,
@@ -301,25 +333,17 @@ import {
 } from "./chunk-GGE4NNQT.js";
 import {
   DEFAULT_TRACE_ANALYST_BUDGETS,
-  INPUT_VALUE,
-  LLM_CACHED_TOKENS,
   LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_CACHE_WRITE_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,
-  OUTPUT_VALUE,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   OtlpFileTraceStore,
+  RUN_COST_ATTR_KEYS,
   SPAN_KIND_ATTR_KEYS,
   SpanNotFoundError,
-  TOOL_ARGS_CAPTURED,
-  TOOL_LATENCY_MS,
-  TOOL_NAME,
   TOOL_NAME_ATTR_KEYS,
   TRACE_ANALYST_TRUNCATION_MARKER_PREFIX,
   TraceFileMissingError,
@@ -337,16 +361,23 @@ import {
   stringField,
   traceAnalystFunctionGroup,
   traceSpanKindToOpenInferenceKind
-} from "./chunk-S2F4J57L.js";
+} from "./chunk-QKEGNI5B.js";
 import {
   RunIntegrityError,
   assertRunCaptured,
   throwIfRunIncomplete
 } from "./chunk-TT4KNT67.js";
+import {
+  FileSystemRawProviderSink,
+  InMemoryRawProviderSink,
+  NoopRawProviderSink,
+  defaultProviderRedactor,
+  providerFromBaseUrl
+} from "./chunk-PC4UYEBM.js";
 import {
   TraceEmitter,
   llmSpanFromProvider
-} from "./chunk-TVVP3ZZQ.js";
+} from "./chunk-VQMK5FMP.js";
 import {
   RunRecordValidationError,
   isRunRecord,
@@ -355,7 +386,16 @@ import {
   resolveRunCostProvenance,
   roundTripRunRecord,
   validateRunRecord
-} from "./chunk-5UF54T55.js";
+} from "./chunk-S3UZOQ5Y.js";
+import {
+  FAILURE_CLASSES,
+  TRACE_SCHEMA_VERSION,
+  isJudgeSpan,
+  isLlmSpan,
+  isRetrievalSpan,
+  isSandboxSpan,
+  isToolSpan
+} from "./chunk-MA6HLL3S.js";
 import {
   AGENT_PROFILE_KINDS,
   AgentProfileCellValidationError,
@@ -377,49 +417,6 @@ import {
   signManifest,
   verifyManifest
 } from "./chunk-VSMTAMNK.js";
-import {
-  LlmCallError,
-  LlmClient,
-  LlmResponseError,
-  LlmRouteAssertionError,
-  assertLlmRoute,
-  backoffMs,
-  callLlm,
-  callLlmJson,
-  costReceiptFromLlm,
-  costReceiptFromLlmError,
-  isTransientLlmError,
-  maximumChargeForLlmRequest,
-  probeLlm,
-  stripFencedJson
-} from "./chunk-NJC7U437.js";
-import {
-  CostAccountingIncompleteError,
-  CostCallConflictError,
-  CostCeilingReachedError,
-  CostLedger,
-  CostLedgerPersistenceError,
-  CostReceiptCaptureError,
-  CostReservationExceededError,
-  costForUsage,
-  modelPriceKey
-} from "./chunk-VCTY3W6J.js";
-import {
-  MODEL_PRICING,
-  MetricsCollector,
-  TokenCounter,
-  estimateCost,
-  estimateTokens,
-  isModelPriced,
-  resolveModelPricing
-} from "./chunk-VI2UW6B6.js";
-import {
-  FileSystemRawProviderSink,
-  InMemoryRawProviderSink,
-  NoopRawProviderSink,
-  defaultProviderRedactor,
-  providerFromBaseUrl
-} from "./chunk-PC4UYEBM.js";
 import {
   AgentEvalError,
   CaptureIntegrityError,
@@ -430,6 +427,21 @@ import {
   ValidationError,
   VerificationError
 } from "./chunk-ONWEPEDO.js";
+import {
+  INPUT_VALUE,
+  LLM_CACHED_TOKENS,
+  LLM_CACHE_WRITE_TOKENS,
+  LLM_COST_USD,
+  LLM_INPUT_TOKENS,
+  LLM_MODEL_NAME,
+  LLM_OUTPUT_TOKENS,
+  LLM_REASONING_TOKENS,
+  OPENINFERENCE_SPAN_KIND,
+  OUTPUT_VALUE,
+  TOOL_ARGS_CAPTURED,
+  TOOL_LATENCY_MS,
+  TOOL_NAME
+} from "./chunk-WDHBCA3M.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";
@@ -11582,6 +11594,8 @@ export {
   KNOWLEDGE_POISONING_KIND_SPEC,
   LLM_CACHED_TOKENS,
   LLM_CACHED_TOKEN_ATTR_KEYS,
+  LLM_CACHE_WRITE_TOKENS,
+  LLM_CACHE_WRITE_TOKEN_ATTR_KEYS,
   LLM_COST_ATTR_KEYS,
   LLM_COST_USD,
   LLM_INPUT_TOKENS,
@@ -11590,6 +11604,8 @@ export {
   LLM_MODEL_NAME,
   LLM_OUTPUT_TOKENS,
   LLM_OUTPUT_TOKEN_ATTR_KEYS,
+  LLM_REASONING_TOKENS,
+  LLM_REASONING_TOKEN_ATTR_KEYS,
   LlmCallError,
   LlmClient,
   LlmResponseError,
@@ -11618,6 +11634,7 @@ export {
   REFERENCE_EQUIVALENCE_INPUT_LIMITS,
   REFERENCE_EQUIVALENCE_JUDGE_VERSION,
   RESEARCH_REPORT_HARD_PAIR_FLOOR,
+  RUN_COST_ATTR_KEYS,
   ReplayCache,
   ReplayCacheMissError,
   ReplayError,