@tangle-network/agent-eval 0.145.11 → 0.145.13

package/dist/cli.js

@@ -1,6 +1,6 @@
 #!/usr/bin/env node
 import { o as runRolloutReleaseCli } from "./hf-dataset-XggBupCr.js";
-import { n as runAnalystBenchmarkCommand } from "./benchmark-command-D8K5YJNS.js";
+import { n as runAnalystBenchmarkCommand } from "./benchmark-command-RyFRoiYm.js";
 import { a as runRpcBatch, o as runRpcOnce, p as handleVersion, r as startServerAsync, s as buildOpenApi } from "./server-ulsOdrTI.js";
 import { writeFileSync } from "node:fs";
 //#region src/cli-config.ts

package/dist/contract/index.d.ts

@@ -595,7 +595,32 @@ interface CodeAgentSessionIntakeOptions {
    * JSON stream has no terminal event, as with `opencode run --format json`. */
   execution?: CodeAgentSessionExecutionReceipt;
 }
+/** One transcript line after the intake rule ran on it. A blank line produces
+ *  nothing, so every value here is either a parsed entry or a counted defect. */
+type CodeAgentJsonlLine = {
+  kind: 'entry';
+  lineNumber: number;
+  entry: unknown;
+} | {
+  kind: 'malformed';
+  lineNumber: number;
+};
 declare function parseCodeAgentJsonl(jsonl: string): ParsedCodeAgentJsonl;
+/** Reads a transcript one line at a time and never holds the file as a single
+ *  string. `parseCodeAgentJsonl` needs the whole file in one string, so a
+ *  session above V8's ~512MB string ceiling throws `ERR_STRING_TOO_LONG` and
+ *  cannot be ingested at all; the largest real Codex rollout on record is 695MB.
+ *
+ *  Lines break on `\n` only, which is what the string path's `split('\n')` does.
+ *  `node:readline` also breaks on a bare `\r`, so it is deliberately not used
+ *  here: a lone carriage return inside a line must stay inside that line for the
+ *  two paths to report the same malformed count. */
+declare function streamCodeAgentJsonlFile(path: string): AsyncGenerator<CodeAgentJsonlLine>;
+/** Streaming counterpart to `parseCodeAgentJsonl` for a transcript on disk.
+ *  It returns the same shape, so a caller that holds every entry keeps working
+ *  above the string ceiling. The entry array still grows with the transcript;
+ *  consume `streamCodeAgentJsonlFile` directly when memory must stay flat. */
+declare function parseCodeAgentJsonlFile(path: string): Promise<ParsedCodeAgentJsonl>;
 declare function fromCodexSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
 declare function fromClaudeCodeSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
 declare function fromOpenCodeSession(options: CodeAgentSessionIntakeOptions): CodeAgentSessionIntakeResult;
@@ -691,5 +716,5 @@ interface FromOtelSpansOptions {
 }
 declare function fromOtelSpans(opts: FromOtelSpansOptions): RunRecord[];
 //#endregion
-export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvalRunDiff, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
+export { type AgentEvalAgent, type AgentEvalEvaluateOptions, type AgentEvalImproveOptions, type AgentProfileImprovementExperimentExecutionInput, type AgentProfileImprovementExperimentRun, type AgentTraceContributor, type AgentTraceContributorType, type AgentTraceConversation, type AgentTraceFile, type AgentTraceIndex, type AgentTraceRange, type AgentTraceRecord, type AnalystFinding, type AnalyzeRunsOptions, type AuthoringProvenance, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type CampaignAggregates, type CampaignArtifactWriter, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignStorage, type CampaignTraceWriter, type CandidateExperimentExecutionInput, type CandidateExperimentRun, type ChatClient, type CodeAgentJsonlLine, type CodeAgentSessionAction, type CodeAgentSessionActionKind, type CodeAgentSessionActionStatus, type CodeAgentSessionActionSurface, type CodeAgentSessionDiagnostic, type CodeAgentSessionExecutionReceipt, type CodeAgentSessionIntakeOptions, type CodeAgentSessionIntakeResult, type CodeAgentSessionMetrics, type CodeAgentSessionObservation, type CodeAgentSessionSource, type CodeAgentSessionTerminalStatus, type CodeSurface, type CompareAgentProfileImprovementExperimentOptions, type CompareCandidateExperimentOptions, type CompareOptimizationMethodsOptions, type ComparisonCost, type CostLedgerHandle, type CostProvenanceSummary, type CreateChatClientOpts, type DefaultAnalystRegistryOptions, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DefineAgentEvalOptions, type DefinedAgentEval, type DeploymentOutcome, type DispatchFn as Dispatch, type DispatchContext, type EvalCellScoreDelta, type EvalDimensionDelta, type EvalGenerationDiff, type EvalReportingSuiteInput, type EvalReportingSuiteOptions, type EvalReportingSuiteResult, type EvalRunDiff, type EvaluatePairedMeasurementsOptions, type EvidenceVector, type ExecutionErrorOutcomeCell, type ExecutionInsight, type ExecutionReport, type ExternalOptimizationExample, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, type FailureClassTally, type FailureClusterInsight, type FeedbackTableMeta, type FeedbackTableRow, FileSystemOutcomeStore, type FileSystemOutcomeStoreOptions, type FromFeedbackTableOptions, type FromFeedbackTableResult, type FromOtelSpansOptions, type FromRunRecordDirOptions, type FromRunRecordDirResult, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaEngineOptions, type GepaEngineRun, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaRunnerCommand, type HeldOutGateOptions, type HostedTenant, InMemoryOutcomeStore, type InsightReport, type InterRaterInsight, type JudgeConfig, type JudgeDimension, type JudgeInsight, type JudgeScore, type LiftInsight, type LlmJudgeDimension, type LlmJudgeOptions, type MutableSurface, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodInput, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type OutcomeCorrelationInsight, type OutcomeStore, type PairedMeasurement, type PairedMeasurementAdapter, type PairedMeasurementEvaluation, type ParetoSignificanceGateOptions, type ParsedCodeAgentJsonl, type PartitionByAuthoringModelResult, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, type Recommendation, type ReferenceEquivalenceJudgeInput, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceJudgeResult, type ReferenceEquivalenceScenario, type ReleaseSummary, type RunAgentProfileImprovementExperimentOptions, type RunCampaignOptions, type RunCandidateExperimentOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunRecordRejection, type ScalarDistribution, type Scenario, type SealAgentProfileImprovementSuiteOptions, type SealCandidateBenchmarkSuiteOptions, type SelfImproveBudget, type SelfImproveOptions, type SelfImproveProgressEvent, type SelfImproveResult, SelfImproveRunError, type SessionScript, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SummarizeExecutionOptions, type SurfaceProposer, type TokenUsageInsight, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
 //# sourceMappingURL=index.d.ts.map

package/dist/contract/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;;;;;UAiCiB;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;iBAkHa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UC7zB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UCnPc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UCrCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;iBAGE,oBAAoB,gBAAgB;iBAepC,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;cAIU,2BAAkB;;;UCpJd;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}
+{"version":3,"file":"index.d.ts","names":[],"sources":["../../src/contract/measured-comparison.ts","../../src/contract/profile-measured-comparison.ts","../../src/contract/intake/run-record-dir.ts","../../src/contract/eval-reporting-suite.ts","../../src/contract/diff.ts","../../src/contract/intake/agent-trace.ts","../../src/contract/intake/code-agent-observation.ts","../../src/contract/intake/code-agent-session.ts","../../src/contract/intake/feedback-table.ts","../../src/contract/intake/otel-spans.ts"],"mappings":";;;;;;;;;;;;;;;;UAiCiB;EACf,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,QAAQ;EACR,MAAM;EACN,eAAe;EACf;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QAAQ,OAAO,oCAAoC,QAAQ;;EAE3D;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;UAII,kBAAkB;EACjC;EACA,UAAU;EACV,WAAW;;;UAII,yBAAyB;EACxC,MAAM,KAAK;EACX,WAAW,KAAK;IAAkB;IAAc;;EAChD,QAAQ,KAAK;EACb,eAAe,KAAK,OAAO;EAC3B,UAAU,KAAK;EACf,UAAU,KAAK;EACf,OAAO,KAAK;;UAGG,kCAAkC;EACjD,uBAAuB,kBAAkB;EACzC,QAAQ;EACR,SAAS,yBAAyB;;EAElC;;EAEA,kBAAkB;;EAElB,kBAAkB;;;KAIR,8BAA8B,KACxC;EAGA,iBAAiB;EACjB,WAAW;EACX;;;iBAIc,2BACd,UAAU,sCACT;;iBAQa,4BACd,SAAS,qCACR;;iBAiBa,wBACd,UAAU,mCACT;iBAQa,0BAA0B,iBAAiB;;iBAarC,uBACpB,SAAS,gCACR,QAAQ;;;;;;;;iBAkEK,2BAA2B,MACzC,SAAS,kCAAkC,QAC1C;;iBA8Ya,0CACd,SAAS,oCACR;;iBAuEa,oCACd,iBACC;iBAkHa,6BAA6B,iBAAiB;iBAM9C,oCACd,iBACC;iBAkBa,8BAA8B;;;;;;;;;;UC7zB7B;EACf,aAAa;EACb,QAAQ,gCAAgC;EACxC;EACA;;UAGe;EACf,YAAY;EACZ;EACA,aAAa;EACb,MAAM;EACN,SAAS;EACT;EACA,SAAS;;UAGM;EACf,YAAY;EACZ,QACE,OAAO,kDACN,QAAQ;;EAEX;;EAEA,aAAa;EACb,SAAS;;UAGM;EACf,cAAc;EACd;IACE;IACA,MAAM;;;UAIO;EACf,YAAY;EACZ,cAAc;EACd;IACE;IACA,MAAM;;EAER,aAAa;EACb;EACA,YAAY;EACZ;EACA,WAAW;;;iBAIG,gCACd,UAAU,sCACT;;iBAQa,iCACd,SAAS,0CACR;;iBAqBa,sCACd,UAAU,4CACT;iBAOa,kCAAkC,iBAAiB;iBAInD,yCACd,iBACC;iBAIa,wCACd,iBACC;;;;;;iBASmB,qCACpB,SAAS,8CACR,QAAQ;;iBA0CK,wDACd,SAAS,kDACR;;iBAuEa,kDACd,iBACC;;;;UCnPc;;EAEf;;EAEA;;EAEA;;UAGe;;;;;;EAMf;;;;;;;EAOA,WAAW;;;;;;EAMX;;UAGe;;EAEf,MAAM;;EAEN,UAAU;;EAEV;;;;;;;;;;iBAkBoB,iBACpB,cACA,UAAS,0BACR,QAAQ;;;;;KC7CC,0BAA0B;UAErB;;;;;EAKf,UAAU,KAAK;;EAEf,OAAO;;;;;;;;;;EAUP;;;;UAKe;;;EAGf,QAAQ;;EAER;;IAEE;;IAEA;;;IAGA;;IAEA;;;IAGA,UAAU;;;EAGZ;;;;;;;iBAUoB,mBACpB,OAAO,yBACP,UAAS,4BACR,QAAQ;;;;;;UC5DM;EACf;EACA;EACA;;;UAIe;EACf;EACA;EACA;EACA;EACA;;;EAGA,YAAY,eAAe,eAAe;;;;UAK3B;EACf;EACA;EACA;EACA;EACA;;EAEA,SAAS;;EAET,SAAS;;EAET,OAAO;;EAEP;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;;;;UAMe;EACf;EACA;EACA;EACA;EACA,oBAAoB;EACpB,mBAAmB;EACnB;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA,cAAc;;;;EAId,aAAa;;;;;;;;;iBAoDC,gBACd,QAAQ,2BACR,OAAO,4BACN;;;;;;iBAqEa,SAAS,QAAQ,cAAc,OAAO,eAAe;;;;;;;iBAsCrD,wBAAwB,KAAK,eAAe;;;KC1OhD;UAEK;EACf,MAAM;;EAEN;;UAGe;EACf;EACA;EACA;;;EAGA,cAAc;;UAGC;EACf;EACA,cAAc;EACd,QAAQ;;UAGO;EACf;EACA,eAAe;;UAGA;EACf;EACA;EACA;EACA;IAAQ;IAAc;;EACtB;IAAS;IAAe;;EACxB,OAAO;;;;UAOQ;EACf;;EAEA;;EAEA;EACA;EACA;;EAEA;;EAEA;;KAGU,kBAAkB,YAAY;;;;;;iBAW1B,gBAAgB,SAAS,qBAAqB;UAmE7C;;;;EAIf,SAAS,YAAY;;;EAGrB,cAAc;;;;;;;;iBASA,8BACd,MAAM,aACN,OAAO,kBACN;;;KCjLS;KAEA;KAEA;KAEA;KASA;UAEK;EACf;EACA;EACA;;UAGe;EACf;EACA;EACA,MAAM;EACN,SAAS;EACT;EACA,QAAQ;EACR;EACA;EACA,UAAU;;UAGK;EACf,QAAQ;EACR;EACA;EACA;IACE,QAAQ;IACR;;EAEF,SAAS;;UAGM;EACf,QAAQ;EACR;EACA;EACA,YAAY;;;;;;;iBAeE,wBACd,SAAS,iCACR;;;UClCc;EACf;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf,QAAQ;EACR;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;;UAGe;EACf,MAAM;EACN,aAAa;EACb,SAAS;EACT,cAAc;;UAGC;EACf;EACA;EACA;EACA;EACA;EACA;EACA,WAAW;EACX;EACA;EACA;EACA;EACA;EACA;;;EAGA,iBAAiB;;;EAGjB,YAAY;;;;KAKF;EACN;EAAe;EAAoB;;EACnC;EAAmB;;iBAcT,oBAAoB,gBAAgB;;;;;;;;;;iBAuB7B,yBAAyB,eAAe,eAAe;;;;;iBA6BxD,wBAAwB,eAAe,QAAQ;iBAUrD,iBACd,SAAS,gCACR;iBAIa,sBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,oBACd,SAAS,gCACR;iBAIa,cACd,SAAS,gCACR;cAIU,2BAAkB;;;UCxNd;;;EAGf;;EAEA;;;EAGA;;;EAGA,WAAW;;UAGI;EACf;;;EAGA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;EAEA;;;EAGA,WAAW;;;EAGX,SAAS;;UAGM;;EAEf,SAAS;;;EAGT,OAAO;;;;EAIP;IAAU;IAAa;;;;;EAIvB;;UAGe;EACf,MAAM;;;EAGN,aAAa;IAAQ;IAAe;IAAe;;;iBAGrC,kBAAkB,MAAM,2BAA2B;;;UCvBlD;EACf,OAAO;;EAEP,eAAe;;EAEf;;;;;;EAMA,eAAe,eAAe,gBAAgB;;iBAGhC,cAAc,MAAM,uBAAuB"}

package/dist/contract/index.js

@@ -1,12 +1,12 @@
 import { s as ValidationError } from "../errors-Dngq5h35.js";
 import { r as pairedBootstrap } from "../paired-tests-BHIhYVdu.js";
-import { a as summarizeExecution, i as analyzeRuns, n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-cSMEFzVu.js";
+import { a as summarizeExecution, i as analyzeRuns, n as SelfImproveRunError, r as selfImprove, t as defineAgentEval } from "../define-agent-eval-BXusiVTQ.js";
 import { a as parseRunRecordSafe } from "../run-record-D2lDdSAz.js";
-import { G as runEval, K as runCampaign, f as runImprovementLoop, nt as campaignSplitDigest, t as llmJudge, x as compareOptimizationMethods, z as defaultProductionGate } from "../llm-judge-DhiJqSjB.js";
+import { G as runEval, K as runCampaign, f as runImprovementLoop, nt as campaignSplitDigest, t as llmJudge, x as compareOptimizationMethods, z as defaultProductionGate } from "../llm-judge-DqAkIQA1.js";
 import { i as isModelPriced, n as estimateCost } from "../metrics-Cl0L1KUy.js";
 import { i as CostLedger } from "../cost-ledger-BSe92yAV.js";
-import { d as mapConcurrentRange } from "../ledger-core-BmZt19oQ.js";
-import { S as inMemoryCampaignStorage, x as fsCampaignStorage } from "../external-optimizer-subprocess-BhKYK0Jv.js";
+import { d as mapConcurrentRange } from "../ledger-core-DTae9rv_.js";
+import { S as inMemoryCampaignStorage, x as fsCampaignStorage } from "../external-optimizer-subprocess-x5AbhAYX.js";
 import { a as heldoutSignificance, s as decidePairedPromotion } from "../power-preflight-DEw-uC7q.js";
 import { r as makeProposalFinding } from "../types-BI4fT3HN.js";
 import { $ as classifyOtlpSpanRole, et as isOtlpModelCall } from "../kind-factory-CPmSd58s.js";
@@ -15,10 +15,11 @@ import { LLM_MODEL_ATTR_KEYS, SPAN_KIND_ATTR_KEYS } from "../trace-attributes.js
 import { t as extractUsage } from "../extract-usage-BrQ8mCLX.js";
 import { n as InMemoryOutcomeStore, t as FileSystemOutcomeStore } from "../outcome-store-ChBKlTd_.js";
 import { i as summarizeTraceErrors, n as recordAggregateMeasurements, r as summarizeExecutionMeasurements, t as readTaskFailureLabels } from "../task-failure-attributes-CpQ4y5RD.js";
-import { a as externalTextOptimizationMethod, c as REFERENCE_EQUIVALENCE_INPUT_LIMITS, d as runReferenceEquivalenceJudge, i as composeGate, l as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, r as heldOutGate, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-C6JSfYxT.js";
+import { a as externalTextOptimizationMethod, c as REFERENCE_EQUIVALENCE_INPUT_LIMITS, d as runReferenceEquivalenceJudge, i as composeGate, l as REFERENCE_EQUIVALENCE_JUDGE_VERSION, n as gepaOptimizationMethod, r as heldOutGate, t as skillOptOptimizationMethod, u as createReferenceEquivalenceJudge } from "../skillopt-optimization-method-CcWZBaSz.js";
 import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-xzA40Evo.js";
 import { createHash } from "node:crypto";
 import { agentCandidateBenchmarkSuiteSchema, agentCandidateBenchmarkTaskSchema, agentCandidateBundleSchema, agentCandidateEvaluationPolicySchema, agentCandidateExperimentSchema, agentImprovementMeasuredComparisonSchema, agentProfileImprovementExperimentSchema, agentProfileImprovementMeasuredComparisonSchema, agentProfileImprovementRunCellSchema, agentProfileImprovementRunReceiptSchema, agentProfileImprovementSuiteInputsSchema, agentProfileImprovementSuiteSchema, agentProfileImprovementTaskSchema, candidateExecutionEvidenceSchema, canonicalCandidateDigest, canonicalCandidateJson, numbersApproximatelyEqual, omitTopLevelDigest } from "@tangle-network/agent-interface";
+import { createReadStream } from "node:fs";
 import { dirname, join } from "node:path";
 import { mkdir, readFile, readdir, stat, writeFile } from "node:fs/promises";
 //#region src/contract/fixed-spend.ts
@@ -1558,6 +1559,7 @@ function projectionFor(source, entries) {
 function codexProjection(entries) {
 	const actions = [];
 	const calls = /* @__PURE__ */ new Map();
+	const transcriptStream = hasCodexTranscriptStream(entries);
 	let finalText = null;
 	let terminal = "unknown";
 	let explicitTerminal = false;
@@ -1566,8 +1568,9 @@ function codexProjection(entries) {
 		const payload = record$1(entry.payload) ?? {};
 		const payloadType = stringField$1(payload, "type");
 		const timestampMs = timestamp(entry.timestamp);
-		const item = record$1(entry.item);
-		if (entryType === "item.started" && item) {
+		const itemEvent = admittedCodexItem(entry, transcriptStream);
+		const item = itemEvent?.item ?? null;
+		if (itemEvent?.eventType === "item.started" && item) {
 			const itemType = stringField$1(item, "type");
 			if (isCodexActionItem(itemType)) {
 				const id = stringField$1(item, "id") ?? `item-${actions.length}`;
@@ -1580,7 +1583,7 @@ function codexProjection(entries) {
 					status: "started",
 					timestampMs,
 					metadata: compactMetadata({
-						sourceEventType: entryType,
+						sourceEventType: itemEvent.eventType,
 						itemType
 					})
 				});
@@ -1588,7 +1591,7 @@ function codexProjection(entries) {
 				actions.push(action);
 			}
 		}
-		if (entryType === "item.completed" && item) {
+		if (itemEvent?.eventType === "item.completed" && item) {
 			const itemType = stringField$1(item, "type");
 			if (itemType === "agent_message" || itemType === "message") finalText = nonEmpty(stringField$1(item, "text")) ?? finalText;
 			if (isCodexActionItem(itemType)) {
@@ -1606,7 +1609,7 @@ function codexProjection(entries) {
 						status,
 						timestampMs,
 						metadata: compactMetadata({
-							sourceEventType: entryType,
+							sourceEventType: itemEvent.eventType,
 							itemType
 						})
 					});
@@ -1986,6 +1989,79 @@ function completeClaudeTools(content, calls) {
 		if (action) action.status = part.is_error === true ? "failed" : "completed";
 	}
 }
+/**
+* Codex writes the same session items on two transports. The app-server stream
+* of `codex exec --json` puts an item at the top level under `item.started` or
+* `item.completed` and names its type in snake_case. A rollout file under
+* `~/.codex/sessions` wraps the same item in an `event_msg` entry under
+* `payload.item_started` or `payload.item_completed` and names its type in
+* PascalCase. Read both here so one branch serves both transports.
+*/
+function codexItemEvent(entry) {
+	const entryType = stringField$1(entry, "type");
+	if (entryType === "item.started" || entryType === "item.completed") {
+		const item = record$1(entry.item);
+		return item ? {
+			eventType: entryType,
+			item
+		} : null;
+	}
+	if (entryType !== "event_msg") return null;
+	const payload = record$1(entry.payload);
+	const payloadType = payload ? stringField$1(payload, "type") : void 0;
+	if (payloadType !== "item_started" && payloadType !== "item_completed") return null;
+	const item = record$1(payload?.item);
+	if (!item) return null;
+	const itemType = codexItemType(stringField$1(item, "type"));
+	return {
+		eventType: payloadType === "item_started" ? "item.started" : "item.completed",
+		item: itemType === void 0 ? item : {
+			...item,
+			type: itemType
+		}
+	};
+}
+/**
+* A rollout item type differs from its app-server name by case alone, except
+* `CollabAgentToolCall`, which the app-server calls `collab_tool_call`. Convert
+* the case mechanically so a new item type still arrives under a stable name.
+*/
+function codexItemType(itemType) {
+	if (itemType === void 0 || !/^[A-Z]/.test(itemType)) return itemType;
+	if (itemType === "CollabAgentToolCall") return "collab_tool_call";
+	return itemType.replace(/([a-z0-9])([A-Z])/g, "$1_$2").toLowerCase();
+}
+function hasCodexTranscriptStream(entries) {
+	return entries.some((entry) => stringField$1(entry, "type") === "response_item");
+}
+/**
+* A rollout file records each turn twice: `response_item` holds the model
+* transcript and the item stream holds the execution record. The app-server
+* transport sends no `response_item` entry at all. So `response_item` keeps the
+* tool call, reasoning, and assistant message accounting wherever it is
+* present, and the item stream adds only the facts a `response_item` cannot
+* express. Counting both streams together doubles the totals: a 186-file
+* rollout corpus holds 58,752 `Reasoning` items against 58,893 `reasoning`
+* response items, and 49,694 `CommandExecution` items against 52,183 response
+* item tool calls.
+*
+* A collaboration or subagent item states no fact of its own here. Its id is
+* either the id of a response item that the same file already reports, or the
+* id of a command that ran inside the sandbox, which the enclosing response
+* item already counts. Over the same corpus, all 320 collaboration items and
+* all 523 subagent items that match a response item resolve to an action that
+* `surfaceForTool` already marks `subagent`, and the 1,434 that match nothing
+* would each add a second action for one operation.
+*/
+function isCodexExecutionOnlyItem(itemType) {
+	return itemType === "file_change" || itemType === "user_message" || itemType === "context_compaction";
+}
+function admittedCodexItem(entry, transcriptStream) {
+	const event = codexItemEvent(entry);
+	if (!event) return null;
+	if (!transcriptStream) return event;
+	return isCodexExecutionOnlyItem(stringField$1(event.item, "type")) ? event : null;
+}
 function isCodexActionItem(itemType) {
 	return itemType === "command_execution" || itemType === "mcp_tool_call" || itemType === "collab_tool_call" || itemType === "web_search" || itemType === "file_change";
 }
@@ -2058,13 +2134,25 @@ function actionFor(input) {
 function surfaceForTool(name) {
 	const normalized = name.toLowerCase();
 	if (normalized.startsWith("mcp__") || normalized.includes("mcp_tool")) return "mcp";
-	if (normalized === "task" || normalized === "agent" || normalized.includes("subagent") || normalized.includes("spawn_agent") || normalized.includes("collab") || normalized.startsWith("multi_agent")) return "subagent";
+	if (normalized === "task" || normalized === "agent" || normalized.includes("subagent") || normalized.includes("spawn_agent") || normalized.includes("collab") || normalized.startsWith("multi_agent") || isCodexCollaborationTool(normalized)) return "subagent";
 	if (normalized === "skill" || normalized.includes("skill")) return "skill";
 	if (normalized.includes("hook")) return "hook";
 	if (normalized.includes("web_search") || normalized.includes("webfetch") || normalized === "web") return "web";
 	if (normalized === "edit" || normalized === "write" || normalized.includes("patch") || normalized.includes("file_change")) return "code";
 	return "tool";
 }
+/**
+* Codex serves its subagent tools from the `collaboration` and `multi_agent_v1`
+* namespaces. The namespace does not reach this function, so the tool name must
+* carry the decision. Codex names every lifecycle verb `<verb>_agent` or
+* `<verb>_agents`, and adds three message verbs that do not use that suffix.
+* A namespaced name such as `collaboration.wait_agent` keeps the verb last, so
+* only the final segment is tested.
+*/
+function isCodexCollaborationTool(normalized) {
+	const verb = normalized.slice(normalized.lastIndexOf(".") + 1);
+	return /^[a-z_]+_agents?$/.test(verb) || verb === "send_message" || verb === "send_input" || verb === "followup_task";
+}
 function statusFrom(value) {
 	const status = (stringField$1(value, "status") ?? "").toLowerCase();
 	if (status === "completed" || status === "success" || status === "succeeded") return "completed";
@@ -2127,18 +2215,80 @@ function timestamp(value) {
 }
 //#endregion
 //#region src/contract/intake/code-agent-session.ts
+/** The single per-line rule. Both the string path and the file path call this,
+*  so malformed-line handling and entry validation cannot drift apart. */
+function readCodeAgentJsonlLine(line, lineNumber) {
+	const trimmed = line.trim();
+	if (!trimmed) return void 0;
+	try {
+		return {
+			kind: "entry",
+			lineNumber,
+			entry: JSON.parse(trimmed)
+		};
+	} catch {
+		return {
+			kind: "malformed",
+			lineNumber
+		};
+	}
+}
 function parseCodeAgentJsonl(jsonl) {
 	const entries = [];
 	let malformedLines = 0;
+	let lineNumber = 0;
 	for (const line of jsonl.split("\n")) {
-		const trimmed = line.trim();
-		if (!trimmed) continue;
-		try {
-			entries.push(JSON.parse(trimmed));
-		} catch {
-			malformedLines += 1;
+		lineNumber += 1;
+		const read = readCodeAgentJsonlLine(line, lineNumber);
+		if (!read) continue;
+		if (read.kind === "malformed") malformedLines += 1;
+		else entries.push(read.entry);
+	}
+	return {
+		entries,
+		malformedLines
+	};
+}
+/** Reads a transcript one line at a time and never holds the file as a single
+*  string. `parseCodeAgentJsonl` needs the whole file in one string, so a
+*  session above V8's ~512MB string ceiling throws `ERR_STRING_TOO_LONG` and
+*  cannot be ingested at all; the largest real Codex rollout on record is 695MB.
+*
+*  Lines break on `\n` only, which is what the string path's `split('\n')` does.
+*  `node:readline` also breaks on a bare `\r`, so it is deliberately not used
+*  here: a lone carriage return inside a line must stay inside that line for the
+*  two paths to report the same malformed count. */
+async function* streamCodeAgentJsonlFile(path) {
+	const stream = createReadStream(path, { encoding: "utf8" });
+	let pending = "";
+	let lineNumber = 0;
+	try {
+		for await (const chunk of stream) {
+			pending += chunk;
+			let start = 0;
+			for (let at = pending.indexOf("\n"); at !== -1; at = pending.indexOf("\n", start)) {
+				lineNumber += 1;
+				const read = readCodeAgentJsonlLine(pending.slice(start, at), lineNumber);
+				if (read) yield read;
+				start = at + 1;
+			}
+			pending = pending.slice(start);
 		}
+	} finally {
+		stream.destroy();
 	}
+	const last = readCodeAgentJsonlLine(pending, lineNumber + 1);
+	if (last) yield last;
+}
+/** Streaming counterpart to `parseCodeAgentJsonl` for a transcript on disk.
+*  It returns the same shape, so a caller that holds every entry keeps working
+*  above the string ceiling. The entry array still grows with the transcript;
+*  consume `streamCodeAgentJsonlFile` directly when memory must stay flat. */
+async function parseCodeAgentJsonlFile(path) {
+	const entries = [];
+	let malformedLines = 0;
+	for await (const read of streamCodeAgentJsonlFile(path)) if (read.kind === "malformed") malformedLines += 1;
+	else entries.push(read.entry);
 	return {
 		entries,
 		malformedLines
@@ -2332,6 +2482,7 @@ function metricsFor(source, entries) {
 function codexMetrics(entries) {
 	const metrics = emptyMetrics(entries.length);
 	const startedToolIds = /* @__PURE__ */ new Set();
+	const transcriptStream = hasCodexTranscriptStream(entries);
 	let startedAt;
 	let completedAt;
 	for (const entry of entries) {
@@ -2350,8 +2501,8 @@ function codexMetrics(entries) {
 		}
 		if (entryType === "turn.failed") metrics.turnsAborted += 1;
 		if (entryType === "error") metrics.unclassifiedErrors += 1;
-		const item = record(entry.item);
-		if (item && (entryType === "item.started" || entryType === "item.completed")) addCodexExecItem(metrics, startedToolIds, entryType, item);
+		const itemEvent = admittedCodexItem(entry, transcriptStream);
+		if (itemEvent) addCodexExecItem(metrics, startedToolIds, itemEvent.eventType, itemEvent.item, transcriptStream);
 		if (entryType === "response_item") {
 			if (payloadType === "function_call" || payloadType === "custom_tool_call") metrics.toolCalls += 1;
 			if (payloadType === "function_call_output" || payloadType === "custom_tool_call_output") metrics.toolOutputs += 1;
@@ -2396,10 +2547,10 @@ function codexMetrics(entries) {
 	metrics.processScore = codexProcessScore(metrics);
 	return metrics;
 }
-function addCodexExecItem(metrics, startedToolIds, eventType, item) {
+function addCodexExecItem(metrics, startedToolIds, eventType, item, transcriptStream) {
 	const itemType = stringField(item, "type");
 	const itemId = stringField(item, "id");
-	const isTool = itemType === "command_execution" || itemType === "mcp_tool_call" || itemType === "collab_tool_call" || itemType === "web_search";
+	const isTool = !transcriptStream && (itemType === "command_execution" || itemType === "mcp_tool_call" || itemType === "collab_tool_call" || itemType === "web_search");
 	if (eventType === "item.started" && isTool) {
 		metrics.toolCalls += 1;
 		if (itemId) startedToolIds.add(itemId);
@@ -2410,7 +2561,9 @@ function addCodexExecItem(metrics, startedToolIds, eventType, item) {
 	}
 	if (eventType !== "item.completed") return;
 	if (itemType === "agent_message" || itemType === "message") metrics.assistantMessages += 1;
+	if (itemType === "user_message") metrics.userMessages += 1;
 	if (itemType === "reasoning") metrics.reasoningItems += 1;
+	if (itemType === "context_compaction") metrics.contextCompactions += 1;
 	if (itemType === "file_change") {
 		metrics.patchAttempts += 1;
 		const status = stringField(item, "status");
@@ -2838,10 +2991,24 @@ function cwdFromEntries(entries) {
 		if (cwd) return cwd;
 	}
 }
+/** Join the `text` of every part in a `[{ type: 'text', text }]` content list. */
+function textFromContentParts(content) {
+	if (!Array.isArray(content)) return void 0;
+	const text = content.map((part) => {
+		const obj = record(part);
+		return obj ? stringField(obj, "text") : void 0;
+	}).filter((part) => part !== void 0).join("\n");
+	return text.length > 0 ? text : void 0;
+}
 function firstUserText(entries) {
 	for (const entry of entries) {
 		const payload = record(entry.payload);
 		if ((payload ? stringField(payload, "type") : void 0) === "user_message") return stringField(payload, "message");
+		const itemEvent = payload ? admittedCodexItem(entry, false) : null;
+		if (itemEvent && stringField(itemEvent.item, "type") === "user_message") {
+			const text = textFromContentParts(itemEvent.item.content);
+			if (text) return text;
+		}
 		if (stringField(entry, "role") === "user") {
 			const content = entry.content;
 			if (typeof content === "string") return content;
@@ -2853,13 +3020,8 @@ function firstUserText(entries) {
 		if (message && stringField(message, "role") === "user") {
 			const content = message.content;
 			if (typeof content === "string") return content;
-			if (Array.isArray(content)) {
-				const text = content.map((part) => {
-					const obj = record(part);
-					return obj ? stringField(obj, "text") : void 0;
-				}).filter((part) => part !== void 0).join("\n");
-				if (text) return text;
-			}
+			const text = textFromContentParts(content);
+			if (text) return text;
 		}
 	}
 }
@@ -3259,6 +3421,6 @@ function collectNumericAttrs(spans) {
 	return raw;
 }
 //#endregion
-export { FileSystemOutcomeStore, InMemoryOutcomeStore, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, SelfImproveRunError, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
+export { FileSystemOutcomeStore, InMemoryOutcomeStore, REFERENCE_EQUIVALENCE_INPUT_LIMITS, REFERENCE_EQUIVALENCE_JUDGE_VERSION, SelfImproveRunError, analyzeRuns, buildDefaultAnalystRegistry, buildEvidenceVector, campaignSplitDigest, compareOptimizationMethods, composeGate, createChatClient, createReferenceEquivalenceJudge, defaultProductionGate, defineAgentEval, diffGenerations, diffRunBaselineToWinner, diffRuns, evalReportingSuite, evaluatePairedMeasurements, externalTextOptimizationMethod, fromClaudeCodeSession, fromCodexSession, fromFeedbackTable, fromKimiCodeSession, fromOpenCodeSession, fromOtelSpans, fromPiSession, fromPigraphSession, fromRunRecordDir, fsCampaignStorage, gepaOptimizationMethod, heldOutGate, inMemoryCampaignStorage, llmJudge, makeProposalFinding, measuredComparisonFromAgentProfileImprovementExperiment, measuredComparisonFromCandidateExperiment, observeCodeAgentSession, paretoPolicy, paretoSignificanceGate, parseAgentTrace, parseCodeAgentJsonl, parseCodeAgentJsonlFile, partitionRunsByAuthoringModel, runAgentProfileImprovementExperiment, runCampaign, runCandidateExperiment, runEval, runImprovementLoop, runReferenceEquivalenceJudge, sealAgentProfileImprovementExperiment, sealAgentProfileImprovementSuite, sealAgentProfileImprovementTask, sealCandidateBenchmarkSuite, sealCandidateBenchmarkTask, sealCandidateExperiment, selfImprove, skillOptOptimizationMethod, streamCodeAgentJsonlFile, summarizeExecution, verifyAgentProfileImprovementExperiment, verifyAgentProfileImprovementExperimentComparison, verifyAgentProfileImprovementSuiteInputs, verifyAgentProfileImprovementTask, verifyCandidateBenchmarkSuite, verifyCandidateBenchmarkSuiteInputs, verifyCandidateBenchmarkTask, verifyCandidateExperiment, verifyCandidateExperimentComparison };
 
 //# sourceMappingURL=index.js.map