@tangle-network/agent-eval 0.102.0 → 0.102.1

package/dist/adapters/http.d.ts

@@ -1,5 +1,5 @@
-import { S as Scenario, D as DispatchFn, b as DispatchContext } from '../types-fWqEJm7h.js';
-import '../run-record-DEwidcqn.js';
+import { S as Scenario, D as DispatchFn, b as DispatchContext } from '../types-DFI_Z-ZL.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';

package/dist/adapters/langchain.d.ts

@@ -1,5 +1,5 @@
-import { S as Scenario, J as JudgeScore, D as DispatchFn, a as JudgeConfig } from '../types-fWqEJm7h.js';
-import '../run-record-DEwidcqn.js';
+import { S as Scenario, J as JudgeScore, D as DispatchFn, a as JudgeConfig } from '../types-DFI_Z-ZL.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';

package/dist/adapters/otel.d.ts

@@ -1,11 +1,11 @@
 import { TraceSpanEvent, HostedClient } from '../hosted/index.js';
-import '../types-fWqEJm7h.js';
-import '../run-record-DEwidcqn.js';
+import '../types-DFI_Z-ZL.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';
-import '../insight-report-C02J3q4T.js';
-import '../summary-report-C4uzRWh8.js';
+import '../insight-report-k0sRTzKg.js';
+import '../summary-report-C0nnxOD8.js';
 import '../failure-cluster-DH9Flgcf.js';
 import '../store-BcFXE6LG.js';
 import '../judge-calibration-0p2QcWNE.js';

package/dist/analyst/index.d.ts

@@ -1,22 +1,22 @@
 import { M as MultiLayerVerifier, V as VerifyOptions, S as Severity } from '../multi-layer-verifier-DUZXrPDA.js';
 import { c as RunCritic, a as RunTrace } from '../run-critic-CmMf05uV.js';
-import { S as SemanticConceptJudgeOptions, a as SemanticConceptJudgeInput, B as BehavioralMetrics } from '../semantic-concept-judge-J8xvjdc3.js';
-export { C as CreateAnalystAiConfig, D as DEFAULT_TRACE_ANALYST_KINDS, b as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, c as FINDING_SUBJECT_GRAMMAR_PROMPT, d as FINDING_SUBJECT_KINDS, e as FindingSubject, f as FindingSubjectKind, g as FindingSubjectStringSchema, h as FindingsDiff, i as FindingsStore, I as IMPROVEMENT_KIND_SPEC, K as KIND_EXPECTED_SUBJECTS, j as KNOWLEDGE_GAP_KIND_SPEC, k as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, l as SKILL_USAGE_ANALYST, m as SkillUsageAnalyst, n as SkillUsageRecord, o as SkillUsageReport, p as SkillUsageScanConfig, q as buildSkillUsageReport, r as createAnalystAi, s as defaultIsMaterial, t as diffFindings, u as emitSkillUsageFindings, v as parseFindingSubject, w as renderFindingSubject } from '../semantic-concept-judge-J8xvjdc3.js';
+import { S as SemanticConceptJudgeOptions, a as SemanticConceptJudgeInput, B as BehavioralMetrics } from '../semantic-concept-judge-D-IlH5v1.js';
+export { C as CreateAnalystAiConfig, D as DEFAULT_TRACE_ANALYST_KINDS, b as DiffPolicy, F as FAILURE_MODE_KIND_SPEC, c as FINDING_SUBJECT_GRAMMAR_PROMPT, d as FINDING_SUBJECT_KINDS, e as FindingSubject, f as FindingSubjectKind, g as FindingSubjectStringSchema, h as FindingsDiff, i as FindingsStore, I as IMPROVEMENT_KIND_SPEC, K as KIND_EXPECTED_SUBJECTS, j as KNOWLEDGE_GAP_KIND_SPEC, k as KNOWLEDGE_POISONING_KIND_SPEC, P as PersistedFinding, l as SKILL_USAGE_ANALYST, m as SkillUsageAnalyst, n as SkillUsageRecord, o as SkillUsageReport, p as SkillUsageScanConfig, q as buildSkillUsageReport, r as createAnalystAi, s as defaultIsMaterial, t as diffFindings, u as emitSkillUsageFindings, v as parseFindingSubject, w as renderFindingSubject } from '../semantic-concept-judge-D-IlH5v1.js';
 import { b as JudgeFn, a as JudgeInput } from '../types-C7DGg5ex.js';
-import { a as Analyst, h as AnalystSeverity, A as AnalystFinding } from '../types-BEzCBMQD.js';
-export { b as AnalystContext, g as AnalystCost, i as AnalystInputKind, j as AnalystRequirements, f as AnalystRunEvent, e as AnalystRunInputs, d as AnalystRunResult, c as AnalystRunSummary, k as ChatCallOpts, C as ChatClient, l as ChatRequest, m as ChatResponse, n as ChatTransport, o as CliBridgeTransportOpts, p as CreateChatClientOpts, D as DirectProviderTransportOpts, E as EvidenceRef, M as MockTransportOpts, R as RouterTransportOpts, S as SandboxSdkTransportOpts, q as computeFindingId, r as createChatClient, s as makeFinding } from '../types-BEzCBMQD.js';
+import { a as Analyst, h as AnalystSeverity, A as AnalystFinding } from '../types-Dz9cKF0g.js';
+export { b as AnalystContext, g as AnalystCost, i as AnalystInputKind, j as AnalystRequirements, f as AnalystRunEvent, e as AnalystRunInputs, d as AnalystRunResult, c as AnalystRunSummary, k as ChatCallOpts, C as ChatClient, l as ChatRequest, m as ChatResponse, n as ChatTransport, o as CliBridgeTransportOpts, p as CreateChatClientOpts, D as DirectProviderTransportOpts, E as EvidenceRef, M as MockTransportOpts, R as RouterTransportOpts, S as SandboxSdkTransportOpts, q as computeFindingId, r as createChatClient, s as makeFinding } from '../types-Dz9cKF0g.js';
 import { TCloud } from '@tangle-network/tcloud';
 import { T as TraceAnalysisStore } from '../store-C1YxJDEK.js';
-export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from '../default-registry-Dhrc__SE.js';
-export { A as ANALYST_SEVERITIES, C as CreateTraceAnalystKindOpts, R as RAW_FINDING_SCHEMA_PROMPT, a as RawAnalystFinding, b as RawAnalystFindingSchema, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, p as parseRawFinding, r as renderPriorFindings } from '../kind-factory-OgqQSvLi.js';
-export { F as FindingToPolicyEditOptions, P as POLICY_EDIT_AXES, a as POLICY_EDIT_TARGET_SURFACES, b as PolicyEdit, c as PolicyEditAdmission, d as PolicyEditAdmissionOptions, e as PolicyEditAxis, f as PolicyEditChange, g as PolicyEditExpectedGain, h as PolicyEditGainDirection, i as PolicyEditGainUnit, j as PolicyEditInit, k as PolicyEditRisk, l as PolicyEditSchemaVersion, m as PolicyEditSource, n as PolicyEditTarget, o as PolicyEditTargetSurface, p as PolicyEditValidationError, q as admitPolicyEdit, r as applyPolicyEditToSurface, s as computePolicyEditId, t as isPolicyEdit, u as makePolicyEdit, v as policyEditFromFinding, w as policyEditsFromFindings, x as scorePolicyEditReadiness, y as validatePolicyEdit } from '../policy-edit-Dccm9tyA.js';
+export { a as AnalystHooks, A as AnalystRegistry, b as AnalystRegistryOptions, B as BudgetPolicy, D as DefaultAnalystRegistryOptions, R as RegistryRunOpts, c as buildDefaultAnalystRegistry } from '../default-registry-ZhqsTr4K.js';
+export { A as ANALYST_SEVERITIES, C as CreateTraceAnalystKindOpts, R as RAW_FINDING_SCHEMA_PROMPT, a as RawAnalystFinding, b as RawAnalystFindingSchema, c as TraceAnalystGolden, T as TraceAnalystKindSpec, d as createTraceAnalystKind, p as parseRawFinding, r as renderPriorFindings } from '../kind-factory-D0nk7AKV.js';
+export { F as FindingToPolicyEditOptions, P as POLICY_EDIT_AXES, a as POLICY_EDIT_TARGET_SURFACES, b as PolicyEdit, c as PolicyEditAdmission, d as PolicyEditAdmissionOptions, e as PolicyEditAxis, f as PolicyEditChange, g as PolicyEditExpectedGain, h as PolicyEditGainDirection, i as PolicyEditGainUnit, j as PolicyEditInit, k as PolicyEditRisk, l as PolicyEditSchemaVersion, m as PolicyEditSource, n as PolicyEditTarget, o as PolicyEditTargetSurface, p as PolicyEditValidationError, q as admitPolicyEdit, r as applyPolicyEditToSurface, s as computePolicyEditId, t as isPolicyEdit, u as makePolicyEdit, v as policyEditFromFinding, w as policyEditsFromFindings, x as scorePolicyEditReadiness, y as validatePolicyEdit } from '../policy-edit-BDQzzsBU.js';
 import { L as LlmClientOptions } from '../llm-client-Bj7g0rqu.js';
 import { AxFunction } from '@ax-llm/ax';
 import '../verdict-C9MlYujm.js';
 import '../schema-m0gsnbt3.js';
 import '../store-BcFXE6LG.js';
 import 'zod';
-import '../run-record-DEwidcqn.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../raw-provider-sink-C46HDghv.js';

package/dist/{analyze-runs-BlJRBniC.d.ts → analyze-runs-Cd-A_K4l.d.ts}

@@ -1,7 +1,7 @@
-import { A as AnalystRegistry } from './default-registry-Dhrc__SE.js';
+import { A as AnalystRegistry } from './default-registry-ZhqsTr4K.js';
 import { a as DatasetScenario } from './dataset-BbGkaN2I.js';
-import { R as RunRecord } from './run-record-DEwidcqn.js';
-import { I as InsightReport } from './insight-report-C02J3q4T.js';
+import { R as RunRecord } from './run-record-MRdJ-Kq2.js';
+import { I as InsightReport } from './insight-report-k0sRTzKg.js';
 
 /**
  * # `analyzeRuns()` — turn a set of agent runs into an actionable decision packet.

package/dist/belief-state/index.d.ts

@@ -1,9 +1,9 @@
 import { c as CalibrationReport } from '../calibration-BPmzuVPk.js';
 import { O as OffPolicyEstimate, a as OffPolicyOptions, b as OffPolicyTrajectory } from '../off-policy-DiwuKKg7.js';
-import { d as CodeAgentSessionSource, a as CodeAgentSessionIntakeOptions, c as CodeAgentSessionMetrics, C as CodeAgentSessionDiagnostic } from '../code-agent-session-B6ZcDwyA.js';
-import { R as RunRecord, b as RunSplitTag } from '../run-record-DEwidcqn.js';
+import { d as CodeAgentSessionSource, a as CodeAgentSessionIntakeOptions, c as CodeAgentSessionMetrics, C as CodeAgentSessionDiagnostic } from '../code-agent-session-Ce-9u7YM.js';
+import { R as RunRecord, b as RunSplitTag } from '../run-record-MRdJ-Kq2.js';
 import { T as TraceStore } from '../store-BcFXE6LG.js';
-import { R as RuntimeTrajectoryRecord, P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection } from '../runtime-trajectory-OJDaTYHN.js';
+import { R as RuntimeTrajectoryRecord, P as ProjectRuntimeTrajectoryEvidenceOptions, a as RuntimeTrajectoryEvidenceProjection } from '../runtime-trajectory-8w0_jmtR.js';
 import '../schema-m0gsnbt3.js';
 import '../outcome-store-rnXLEqSn.js';
 import '@tangle-network/agent-interface';

package/dist/benchmarks/index.d.ts

@@ -1,5 +1,5 @@
-export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as deterministicSplit, e as routing } from '../index-W96macmS.js';
-import '../run-record-DEwidcqn.js';
+export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, c as BenchmarkEvaluation, d as deterministicSplit, e as routing } from '../index-B-bFgiAF.js';
+import '../run-record-MRdJ-Kq2.js';
 import '@tangle-network/agent-interface';
 import '../errors-CzMUYo7b.js';
 import '../schema-m0gsnbt3.js';

package/dist/campaign/index.d.ts

@@ -1,19 +1,19 @@
-import { S as SignedManifest, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-BjGZf9YA.js';
-export { L as LlmJudgeDimension, c as LlmJudgeOptions, l as llmJudge } from '../pre-registration-BjGZf9YA.js';
+import { S as SignedManifest, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-Dzg61IQA.js';
+export { L as LlmJudgeDimension, c as LlmJudgeOptions, l as llmJudge } from '../pre-registration-Dzg61IQA.js';
 import { A as AnalyzeTracesOptions, a as AnalyzeTracesInput, b as AnalyzeTracesResult } from '../analyst-C8HHvfJp.js';
-import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, e as GenerationRecord, g as Gate, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, f as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, y as LabeledScenarioSource, C as CampaignResult, o as CodeSurface } from '../types-fWqEJm7h.js';
-export { k as CampaignAggregates, l as CampaignArtifactWriter, m as CampaignCellResult, n as CampaignCostMeter, z as CampaignTokenUsage, d as CampaignTraceWriter, D as DispatchFn, h as GateContext, j as GateDecision, G as GateResult, p as GenerationCandidate, A as JudgeAggregate, c as JudgeDimension, i as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-fWqEJm7h.js';
-import { e as CampaignRunPlan, P as PlanCampaignRunOptions, C as CampaignStorage, R as RunCampaignOptions, c as RunImprovementLoopOptions } from '../gepa-bxuDoaO9.js';
-export { h as CampaignRunPlanCell, j as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, k as OpenAutoPrResult, b as RunImprovementLoopResult, a as RunOptimizationOptions, l as RunOptimizationResult, m as countSentenceEdits, n as defaultRenderDiff, o as extractH2Sections, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, p as openAutoPr, q as planCampaignRun, r as runCampaign, d as runImprovementLoop, s as runOptimization, t as surfaceHash } from '../gepa-bxuDoaO9.js';
-export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, k as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, l as EmitLoopProvenanceArgs, m as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, n as LoopProvenanceBackend, o as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, R as RunEvalOptions, e as buildEvidenceVector, q as buildLoopProvenanceRecord, f as composeGate, g as defaultProductionGate, s as emitLoopProvenance, h as evolutionaryProposer, i as heldOutGate, t as loopProvenanceSpans, p as paretoPolicy, j as paretoSignificanceGate, u as provenanceRecordPath, v as provenanceSpansPath, r as runEval, w as surfaceContentHash } from '../provenance-BEITkFII.js';
+import { S as Scenario, M as MutableSurface, b as DispatchContext, a as JudgeConfig, e as GenerationRecord, g as Gate, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, f as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, y as LabeledScenarioSource, C as CampaignResult, o as CodeSurface } from '../types-DFI_Z-ZL.js';
+export { k as CampaignAggregates, l as CampaignArtifactWriter, m as CampaignCellResult, n as CampaignCostMeter, z as CampaignTokenUsage, d as CampaignTraceWriter, D as DispatchFn, h as GateContext, j as GateDecision, G as GateResult, p as GenerationCandidate, A as JudgeAggregate, c as JudgeDimension, i as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-DFI_Z-ZL.js';
+import { e as CampaignRunPlan, P as PlanCampaignRunOptions, C as CampaignStorage, R as RunCampaignOptions, c as RunImprovementLoopOptions } from '../gepa-DeyPTlvx.js';
+export { h as CampaignRunPlanCell, j as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, k as OpenAutoPrResult, b as RunImprovementLoopResult, a as RunOptimizationOptions, l as RunOptimizationResult, m as countSentenceEdits, n as defaultRenderDiff, o as extractH2Sections, f as fsCampaignStorage, g as gepaProposer, i as inMemoryCampaignStorage, p as openAutoPr, q as planCampaignRun, r as runCampaign, d as runImprovementLoop, s as runOptimization, t as surfaceHash } from '../gepa-DeyPTlvx.js';
+export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, k as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, l as EmitLoopProvenanceArgs, m as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, n as LoopProvenanceBackend, o as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, P as ParetoSignificanceGateOptions, c as PromotionObjective, d as PromotionPolicy, R as RunEvalOptions, e as buildEvidenceVector, q as buildLoopProvenanceRecord, f as composeGate, g as defaultProductionGate, s as emitLoopProvenance, h as evolutionaryProposer, i as heldOutGate, t as loopProvenanceSpans, p as paretoPolicy, j as paretoSignificanceGate, u as provenanceRecordPath, v as provenanceSpansPath, r as runEval, w as surfaceContentHash } from '../provenance-BhJm32vN.js';
 import { E as EProcessState, a as PairedBootstrapResult } from '../statistics-xP-cWc5k.js';
 import { L as LlmClientOptions } from '../llm-client-Bj7g0rqu.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { A as AgentEvalError } from '../errors-CzMUYo7b.js';
-import { b as RunSplitTag, R as RunRecord } from '../run-record-DEwidcqn.js';
-import { b as PolicyEdit, F as FindingToPolicyEditOptions, d as PolicyEditAdmissionOptions, c as PolicyEditAdmission } from '../policy-edit-Dccm9tyA.js';
-import { T as TraceAnalystKindSpec } from '../kind-factory-OgqQSvLi.js';
-import { A as AnalystFinding } from '../types-BEzCBMQD.js';
+import { b as RunSplitTag, R as RunRecord } from '../run-record-MRdJ-Kq2.js';
+import { b as PolicyEdit, F as FindingToPolicyEditOptions, d as PolicyEditAdmissionOptions, c as PolicyEditAdmission } from '../policy-edit-BDQzzsBU.js';
+import { T as TraceAnalystKindSpec } from '../kind-factory-D0nk7AKV.js';
+import { A as AnalystFinding } from '../types-Dz9cKF0g.js';
 import '@tangle-network/tcloud';
 import '../verdict-C9MlYujm.js';
 import '@ax-llm/ax';
@@ -24,8 +24,8 @@ import '../store-BcFXE6LG.js';
 import '../schema-m0gsnbt3.js';
 import '../pareto-E-pembql.js';
 import '../hosted/index.js';
-import '../insight-report-C02J3q4T.js';
-import '../summary-report-C4uzRWh8.js';
+import '../insight-report-k0sRTzKg.js';
+import '../summary-report-C0nnxOD8.js';
 import '../failure-cluster-DH9Flgcf.js';
 import '../judge-calibration-0p2QcWNE.js';
 import '../types-C7DGg5ex.js';

package/dist/campaign/index.js

@@ -10,8 +10,9 @@ import {
   paretoPolicy,
   paretoSignificanceGate,
   runEval
-} from "../chunk-QUCGGMYM.js";
+} from "../chunk-YLKDN7JV.js";
 import {
+  HARNESS_NATIVE_MODEL,
   agentProfileHash,
   agentProfileId,
   agentProfileModelId,
@@ -41,7 +42,7 @@ import {
   runOptimization,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-52CCCXU3.js";
+} from "../chunk-Z3FLN24V.js";
 import {
   assertRealBackend,
   contentHash,
@@ -52,7 +53,7 @@ import {
   runCampaign,
   summarizeBackendIntegrity,
   tangleTracesRoot
-} from "../chunk-LSCBODPQ.js";
+} from "../chunk-QIT2XZ4E.js";
 import {
   estimateCost,
   isModelPriced
@@ -87,8 +88,9 @@ import {
 } from "../chunk-YBIGNSCZ.js";
 import "../chunk-PC4UYEBM.js";
 import {
+  modelHasSnapshot,
   validateRunRecord
-} from "../chunk-G6S73VA7.js";
+} from "../chunk-2NSLDY4B.js";
 import {
   buildAgentProfileCell
 } from "../chunk-ABOIVNXL.js";
@@ -2031,10 +2033,25 @@ function cellComposite(cell) {
   const composites = Object.values(cell.judgeScores).map((s) => s.composite);
   return composites.length === 0 ? 0 : mean2(composites);
 }
+function requireResolvedModel(cell, profileId) {
+  const resolved = cell.resolvedModel?.trim();
+  if (!resolved) {
+    throw new ProfileMatrixError(
+      `profile '${profileId}' declared the '${HARNESS_NATIVE_MODEL}' runtime-resolved model but its dispatch reported no resolved model for cell '${cell.cellId}' \u2014 report it via ctx.cost.observeModel(<id>) so the RunRecord pins the real model (never records '${HARNESS_NATIVE_MODEL}')`
+    );
+  }
+  if (!modelHasSnapshot(resolved)) {
+    throw new ProfileMatrixError(
+      `profile '${profileId}' resolved to model '${resolved}' for cell '${cell.cellId}', which lacks a snapshot version \u2014 pin it (name@YYYY-MM-DD or name-YYYYMMDD) before reporting it via ctx.cost.observeModel`
+    );
+  }
+  return resolved;
+}
 function buildRunRecord(args) {
   const { cell, profile, profileHash, configHash, experimentId, splitTag, commitSha, matrixId } = args;
   const profileId = agentProfileId(profile);
-  const model = agentProfileModelId(profile);
+  const declaredModel = agentProfileModelId(profile);
+  const model = declaredModel === HARNESS_NATIVE_MODEL ? requireResolvedModel(cell, profileId) : declaredModel;
   const composite = cellComposite(cell);
   const raw = { composite };
   const perJudge = {};
@@ -2120,7 +2137,8 @@ async function runProfileMatrix(opts) {
   for (const profile of opts.profiles) {
     const profileHash = agentProfileHash(profile);
     const profileId = agentProfileId(profile);
-    const model = agentProfileModelId(profile);
+    const declaredModel = agentProfileModelId(profile);
+    const model = declaredModel === HARNESS_NATIVE_MODEL ? `${HARNESS_NATIVE_MODEL}@runtime-resolved` : declaredModel;
     try {
       validateRunRecord({
         runId: `${matrixId}:${profileId}:probe`,
@@ -2149,7 +2167,7 @@ async function runProfileMatrix(opts) {
   for (const profile of opts.profiles) {
     const profileHash = agentProfileHash(profile);
     const profileId = agentProfileId(profile);
-    const model = agentProfileModelId(profile);
+    const declaredModel = agentProfileModelId(profile);
     const configHash = sha({
       profile: profileHash,
       judges: (opts.judges ?? []).map((j) => j.name),
@@ -2173,14 +2191,16 @@ async function runProfileMatrix(opts) {
       runDir: join3(opts.runDir, sanitize(profileId))
     });
     const axis = harnessAxisOf(profile);
-    const agentProfileCell = await buildAgentProfileCell({
+    const buildCellIdentity = (cellModel) => buildAgentProfileCell({
       profileId,
       sourceProfile: { kind: "agent-interface-profile", hash: profileHash },
-      model,
+      model: cellModel,
       ...axis ? { harness: { id: axis.harness } } : {}
     });
+    const sharedCellIdentity = declaredModel === HARNESS_NATIVE_MODEL ? void 0 : await buildCellIdentity(declaredModel);
     const profileRecords = [];
     for (const cell of campaign.cells) {
+      const agentProfileCell = sharedCellIdentity ?? await buildCellIdentity(requireResolvedModel(cell, profileId));
       const record = buildRunRecord({
         cell,
         profile,
@@ -2206,7 +2226,10 @@ async function runProfileMatrix(opts) {
     byProfile[profileId] = {
       profileId,
       profileHash,
-      model,
+      // The declared model, unless it snapped to the sentinel — then the
+      // resolved model the cells actually ran on (all cells of a profile share
+      // one harness, so the first record's model is representative).
+      model: declaredModel === HARNESS_NATIVE_MODEL ? profileRecords[0]?.model ?? declaredModel : declaredModel,
       records: profileRecords.length,
       meanComposite: mean2(profileRecords.map(compositeOf)),
       totalCostUsd: pricedTotalCostUsd,