@tangle-network/agent-eval 0.143.0 → 0.144.0

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (64) hide show
  1. package/CHANGELOG.md +22 -0
  2. package/README.md +21 -8
  3. package/dist/analyst/index.d.ts +4 -4
  4. package/dist/analyst/index.d.ts.map +1 -1
  5. package/dist/analyst/index.js +3 -3
  6. package/dist/{benchmark-command-CSddvlq7.js → benchmark-command-4c7N_rlw.js} +7 -6
  7. package/dist/benchmark-command-4c7N_rlw.js.map +1 -0
  8. package/dist/benchmarks/index.d.ts +1 -1
  9. package/dist/benchmarks/index.js +1 -1
  10. package/dist/{benchmarks-Dg0B56tK.js → benchmarks-05UGZ8sZ.js} +4 -4
  11. package/dist/{benchmarks-Dg0B56tK.js.map → benchmarks-05UGZ8sZ.js.map} +1 -1
  12. package/dist/campaign/index.d.ts +4 -3
  13. package/dist/campaign/index.js +4 -4
  14. package/dist/{campaign-C7ZGiZYG.js → campaign-BKOtvRAB.js} +11 -9
  15. package/dist/campaign-BKOtvRAB.js.map +1 -0
  16. package/dist/cli.js +1 -1
  17. package/dist/contract/index.d.ts +1 -1
  18. package/dist/contract/index.js +3 -3
  19. package/dist/{default-registry-BupFSXSN.js → default-registry-DfHJEwYh.js} +2 -2
  20. package/dist/{default-registry-BupFSXSN.js.map → default-registry-DfHJEwYh.js.map} +1 -1
  21. package/dist/{dspy-rlm-engine-CJ4sN8Wx.js → dspy-rlm-engine-CBFwlyaY.js} +89 -7
  22. package/dist/dspy-rlm-engine-CBFwlyaY.js.map +1 -0
  23. package/dist/external-optimizer-contracts-iK0yu4AR.d.ts +128 -0
  24. package/dist/external-optimizer-contracts-iK0yu4AR.d.ts.map +1 -0
  25. package/dist/{index-BKjcUH13.d.ts → index-BZUe-ODI.d.ts} +41 -5
  26. package/dist/index-BZUe-ODI.d.ts.map +1 -0
  27. package/dist/{index-DY83M1WL.d.ts → index-DgbFl4cv.d.ts} +9 -8
  28. package/dist/{index-DY83M1WL.d.ts.map → index-DgbFl4cv.d.ts.map} +1 -1
  29. package/dist/{index-CaOjqCli.d.ts → index-DtMpBKVF.d.ts} +3 -3
  30. package/dist/{index-CaOjqCli.d.ts.map → index-DtMpBKVF.d.ts.map} +1 -1
  31. package/dist/index.d.ts +7 -7
  32. package/dist/index.js +9 -8
  33. package/dist/index.js.map +1 -1
  34. package/dist/{integrity-hruKe6oe.js → integrity-DY6tIbl0.js} +48 -18
  35. package/dist/integrity-DY6tIbl0.js.map +1 -0
  36. package/dist/openapi.json +1 -1
  37. package/dist/{single-run-lock-Do-8vBIJ.js → single-run-lock-t1si1ob7.js} +370 -152
  38. package/dist/single-run-lock-t1si1ob7.js.map +1 -0
  39. package/dist/{skill-usage-CRqaHi0R.d.ts → skill-usage-BiVEU0QY.d.ts} +3 -3
  40. package/dist/{skill-usage-CRqaHi0R.d.ts.map → skill-usage-BiVEU0QY.d.ts.map} +1 -1
  41. package/dist/{skillopt-optimization-method-CwRvi-rL.d.ts → skillopt-optimization-method-B7o01OdX.d.ts} +43 -9
  42. package/dist/skillopt-optimization-method-B7o01OdX.d.ts.map +1 -0
  43. package/dist/{skillopt-optimization-method-Di8Xkalt.js → skillopt-optimization-method-Ds8J1_K8.js} +289 -222
  44. package/dist/skillopt-optimization-method-Ds8J1_K8.js.map +1 -0
  45. package/dist/supervisor-run/index.d.ts +3 -3
  46. package/dist/supervisor-run/index.js +3 -3
  47. package/dist/{supervisor-run-BEns49Nx.js → supervisor-run-DiyQVczd.js} +301 -25
  48. package/dist/supervisor-run-DiyQVczd.js.map +1 -0
  49. package/dist/{types-Dea6tiVI.d.ts → types-yLK8gXE9.d.ts} +15 -2
  50. package/dist/types-yLK8gXE9.d.ts.map +1 -0
  51. package/docs/campaign-proposers.md +11 -7
  52. package/package.json +1 -1
  53. package/dist/benchmark-command-CSddvlq7.js.map +0 -1
  54. package/dist/campaign-C7ZGiZYG.js.map +0 -1
  55. package/dist/dspy-rlm-engine-CJ4sN8Wx.js.map +0 -1
  56. package/dist/external-optimizer-contracts-nb7c_WAR.d.ts +0 -42
  57. package/dist/external-optimizer-contracts-nb7c_WAR.d.ts.map +0 -1
  58. package/dist/index-BKjcUH13.d.ts.map +0 -1
  59. package/dist/integrity-hruKe6oe.js.map +0 -1
  60. package/dist/single-run-lock-Do-8vBIJ.js.map +0 -1
  61. package/dist/skillopt-optimization-method-CwRvi-rL.d.ts.map +0 -1
  62. package/dist/skillopt-optimization-method-Di8Xkalt.js.map +0 -1
  63. package/dist/supervisor-run-BEns49Nx.js.map +0 -1
  64. package/dist/types-Dea6tiVI.d.ts.map +0 -1
package/CHANGELOG.md CHANGED
@@ -4,6 +4,28 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
4
4
 
5
5
  ---
6
6
 
7
+ ## [0.144.0] - 2026-08-03 - caller-owned optimizer execution
8
+
9
+ ### Changed
10
+
11
+ - **Breaking:** official GEPA and SkillOpt optimizer models replace direct `baseUrl` and `apiKey` fields with a caller-owned `call` callback and stable `callRef` identity.
12
+ Agent Eval no longer accepts provider credentials or performs provider retries for this path; the package that owns execution, such as Runtime with one exact `AgentProfile`, makes each admitted call and reports its own retry history.
13
+ - Every invoked optimizer-model call must return a typed success or failure, a canonical measured cost receipt, and finite JSON execution evidence.
14
+ Rejected callbacks, malformed outcomes, missing receipts, missing evidence, and response/receipt usage disagreements fail the optimizer attempt.
15
+ - Official optimizer runs persist append-only candidate, evaluation, refusal, and model-execution records with counts and SHA-256 identities in method provenance.
16
+ - Comparison costs now carry explicit observed, estimated, or uncaptured provenance, so an unknown bill remains unknown even when a known subtotal exists.
17
+ - Optimizer pricing and dollar limits are optional.
18
+ When billed USD is unknown, callers omit both instead of presenting a catalog estimate or zero as observed billing.
19
+ - GEPA `maxProposerCostUsd` is optional while evaluation, request, byte, output-token, reasoning-token, and timeout limits remain caller-controlled.
20
+
21
+ ### Fixed
22
+
23
+ - Runtime file-backed run contexts are now read by the supervisor-run parser instead of being mistaken for missing inline journals.
24
+ - Agent-profile matrix expansion preserves the canonical `harness` field on every generated profile.
25
+ - OpenAI-style nested prompt-cache fields and Anthropic-style separate cache-read and cache-creation fields are preserved and reconciled against the execution owner's receipt.
26
+ - Negative, fractional, and unsafe reasoning-token allowances are rejected before they can understate a cost reservation.
27
+ - The official optimizer model callback is invoked exactly once per admitted request; Agent Eval no longer hides a second retry policy inside its proxy.
28
+
7
29
  ## [0.143.0] - 2026-08-03 - exact campaign cost provenance
8
30
 
9
31
  ### Changed
package/README.md CHANGED
@@ -174,6 +174,13 @@ Agent Eval supplies the train and selection cases, executes candidates, records
174
174
  ```ts
175
175
  import { gepaOptimizationMethod } from '@tangle-network/agent-eval/campaign'
176
176
 
177
+ // Supplied by the package that owns model execution. Discovery uses Runtime
178
+ // with one exact AgentProfile; Agent Eval receives no provider credential.
179
+ declare const optimizerExecution: {
180
+ call: import('@tangle-network/agent-eval/campaign').ExternalOptimizerModelCall
181
+ callRef: string
182
+ }
183
+
177
184
  const optimizerPricing = {
178
185
  inputUsdPerMillion: Number(process.env.OPTIMIZER_INPUT_USD_PER_MILLION),
179
186
  outputUsdPerMillion: Number(process.env.OPTIMIZER_OUTPUT_USD_PER_MILLION),
@@ -192,8 +199,7 @@ const gepa = gepaOptimizationMethod<MyCase, MyArtifact>({
192
199
  },
193
200
  optimizer: {
194
201
  model: 'gpt-4.1-mini',
195
- baseUrl: 'https://api.openai.com/v1',
196
- apiKey: process.env.OPENAI_API_KEY!,
202
+ ...optimizerExecution,
197
203
  budget: {
198
204
  maxCostUsd: 5,
199
205
  maxRequests: 100,
@@ -209,7 +215,8 @@ const gepa = gepaOptimizationMethod<MyCase, MyArtifact>({
209
215
  ```
210
216
 
211
217
  GEPA also supports official sequential, adaptive, best-of, vote, and Omni recipes through the same factory.
212
- When every recipe stage uses the standard GEPA engine, Agent Eval keeps the provider key outside Python and records exact reflection usage through a local proxy.
218
+ When every recipe stage uses the standard GEPA engine, Agent Eval gives Python only a loopback address and asks the caller-owned execution package to make every model call.
219
+ Every call must return measured usage and finite JSON execution evidence; Agent Eval stores both and fails the attempt when either is missing.
213
220
  Other official GEPA engines can still run through `engineConfig`, but their model spend remains incomplete unless the engine reports it.
214
221
  Custom engines can register through GEPA's official registry by listing their Python modules in `engineModules`.
215
222
 
@@ -218,6 +225,7 @@ Run the repository example:
218
225
  ```sh
219
226
  OPTIMIZERS=gepa \
220
227
  LLM_API_KEY="$OPENAI_API_KEY" \
228
+ OPTIMIZER_EXECUTION_OWNER_MODULE=@acme/runtime-optimizer-owner \
221
229
  GEPA_PRICE_IN_PER_M=0.4 \
222
230
  GEPA_PRICE_OUT_PER_M=1.6 \
223
231
  pnpm tsx examples/compare-optimization-methods/index.ts
@@ -233,6 +241,11 @@ SkillOpt receives train and selection cases but never receives final cases.
233
241
  ```ts
234
242
  import { skillOptOptimizationMethod } from '@tangle-network/agent-eval/campaign'
235
243
 
244
+ declare const optimizerExecution: {
245
+ call: import('@tangle-network/agent-eval/campaign').ExternalOptimizerModelCall
246
+ callRef: string
247
+ }
248
+
236
249
  const optimizerPricing = {
237
250
  inputUsdPerMillion: Number(process.env.OPTIMIZER_INPUT_USD_PER_MILLION),
238
251
  outputUsdPerMillion: Number(process.env.OPTIMIZER_OUTPUT_USD_PER_MILLION),
@@ -247,8 +260,7 @@ const skillopt = skillOptOptimizationMethod<MyCase, MyArtifact>({
247
260
  },
248
261
  optimizer: {
249
262
  model: 'gpt-4.1-mini',
250
- baseUrl: 'https://api.openai.com/v1',
251
- apiKey: process.env.OPENAI_API_KEY!,
263
+ ...optimizerExecution,
252
264
  budget: {
253
265
  maxCostUsd: 5,
254
266
  maxRequests: 100,
@@ -265,8 +277,9 @@ const skillopt = skillOptOptimizationMethod<MyCase, MyArtifact>({
265
277
  ```
266
278
 
267
279
  Replace the example rates with the exact rates for your endpoint.
268
- Agent Eval places a local OpenAI-compatible proxy between each standard optimizer and the provider.
269
- The proxy enforces request, byte, token, and dollar limits before forwarding calls, records exact usage from provider responses, and does not pass the provider key to the optimizer process.
280
+ Agent Eval places a local OpenAI-compatible proxy between each standard optimizer and the caller-owned execution package.
281
+ The proxy enforces request, byte, and token limits before one callback invocation, records the caller's measured receipt and execution evidence, and does not accept a provider URL or credential.
282
+ Dollar limits and pricing are optional; omit both when billed USD is unknown instead of supplying a catalog estimate as if it were observed billing.
270
283
 
271
284
  ## Optimize A DSPy Program
272
285
 
@@ -426,7 +439,7 @@ See [concepts](./docs/concepts.md), [customer paths](./docs/customer-journeys.md
426
439
  | `@tangle-network/agent-eval/reporting` | Statistical comparisons and report rendering. |
427
440
  | `@tangle-network/agent-eval/analyst` | Built-in and custom trace analysts, labeled comparison, costs, and reports. |
428
441
  | `@tangle-network/agent-eval/traces` | Store, replay, and inspect structured traces. |
429
- | `@tangle-network/agent-eval/supervisor-run` | Read and analyze recursive supervisor runs without collapsing missing measurements to zero. |
442
+ | `@tangle-network/agent-eval/supervisor-run` | Read loops or agent-runtime recursive run directories without collapsing missing measurements to zero. |
430
443
  | `@tangle-network/agent-eval/benchmarks` | Benchmark adapters and retrieval metrics. |
431
444
  | `@tangle-network/agent-eval/rl` | Export rewards, preferences, and training rows. |
432
445
  | `@tangle-network/agent-eval/wire` | HTTP and RPC schemas for other languages. |
@@ -1,5 +1,5 @@
1
1
  import { a as MultiLayerVerifier, l as VerifyOptions, o as Severity } from "../multi-layer-verifier-BHY1gWAc.js";
2
- import { A as parseFindingSubject, C as FINDING_SUBJECT_KINDS, D as FindingSubjectStringSchema, E as FindingSubjectKind, F as DefineExactCustomAnalystOptions, G as SemanticConceptJudgeOptions, I as defineCustomAnalyst, L as defineTraceAnalyst, M as DspyRlmTraceEngineOptions, N as createDspyRlmTraceEngine, O as KIND_EXPECTED_SUBJECTS, P as DefineCustomAnalystOptions, S as FINDING_SUBJECT_GRAMMAR_PROMPT, T as FindingSubject, W as SemanticConceptJudgeInput, Y as RunCritic, Z as RunTrace, _ as FindingsDiff, a as SkillUsageScanConfig, b as defaultIsMaterial, c as DEFAULT_TRACE_ANALYST_KINDS, d as IMPROVEMENT_KIND_SPEC, f as FAILURE_MODE_KIND_SPEC, g as DiffPolicy, h as emitControlIntegrityFindings, i as SkillUsageReport, j as renderFindingSubject, k as findingSubjectGrammarPromptFor, l as KNOWLEDGE_POISONING_KIND_SPEC, m as ControlIntegrityAnalyst, n as SkillUsageAnalyst, o as buildSkillUsageReport, p as CONTROL_INTEGRITY_ANALYST, r as SkillUsageRecord, s as emitSkillUsageFindings, t as SKILL_USAGE_ANALYST, u as KNOWLEDGE_GAP_KIND_SPEC, v as FindingsStore, w as FINDING_SUBJECT_SYNTAX, x as diffFindings, y as PersistedFinding } from "../skill-usage-CRqaHi0R.js";
2
+ import { A as parseFindingSubject, C as FINDING_SUBJECT_KINDS, D as FindingSubjectStringSchema, E as FindingSubjectKind, F as DefineExactCustomAnalystOptions, G as SemanticConceptJudgeOptions, I as defineCustomAnalyst, L as defineTraceAnalyst, M as DspyRlmTraceEngineOptions, N as createDspyRlmTraceEngine, O as KIND_EXPECTED_SUBJECTS, P as DefineCustomAnalystOptions, S as FINDING_SUBJECT_GRAMMAR_PROMPT, T as FindingSubject, W as SemanticConceptJudgeInput, Y as RunCritic, Z as RunTrace, _ as FindingsDiff, a as SkillUsageScanConfig, b as defaultIsMaterial, c as DEFAULT_TRACE_ANALYST_KINDS, d as IMPROVEMENT_KIND_SPEC, f as FAILURE_MODE_KIND_SPEC, g as DiffPolicy, h as emitControlIntegrityFindings, i as SkillUsageReport, j as renderFindingSubject, k as findingSubjectGrammarPromptFor, l as KNOWLEDGE_POISONING_KIND_SPEC, m as ControlIntegrityAnalyst, n as SkillUsageAnalyst, o as buildSkillUsageReport, p as CONTROL_INTEGRITY_ANALYST, r as SkillUsageRecord, s as emitSkillUsageFindings, t as SKILL_USAGE_ANALYST, u as KNOWLEDGE_GAP_KIND_SPEC, v as FindingsStore, w as FINDING_SUBJECT_SYNTAX, x as diffFindings, y as PersistedFinding } from "../skill-usage-BiVEU0QY.js";
3
3
  import { b as CustomTokenPricing, c as CostLedgerHandle } from "../cost-ledger-FuQvHxPm.js";
4
4
  import { A as ChatClient, B as SandboxSdkTransportOpts, F as CreateChatClientOpts, I as CustomTransportOpts, L as DirectProviderTransportOpts, M as ChatResponse, N as ChatTransport, P as CliBridgeTransportOpts, R as MockTransportOpts, V as createChatClient, j as ChatRequest, k as ChatCallOpts, m as JudgeInput, p as JudgeFn, z as RouterTransportOpts } from "../types-BjMFz88h.js";
5
5
  import { _ as makeFinding, a as AnalystInputKind, c as AnalystRunInputs, d as AnalystSeverity, f as AnalystUsageReceipt, g as computeFindingId, h as ProposalFindingOrigin, i as AnalystFinding, l as AnalystRunResult, m as ProposalFinding, n as AnalystContext, o as AnalystRequirements, p as EvidenceRef, r as AnalystCost, s as AnalystRunEvent, t as Analyst, u as AnalystRunSummary, v as makeProposalFinding, x as TraceAnalysisStore } from "../types-D3jh6F98.js";
@@ -7,7 +7,7 @@ import { a as createTraceAnalyst, c as runTraceAnalyst, d as deriveEfficiencyFin
7
7
  import { a as ExactAnalystRunPolicySnapshot, c as ExactAnalystSnapshot, d as ExactExecutionComponentSnapshot, i as ExactAnalystRunEvent, l as ExactCapableAnalyst, n as ExactAnalystExecutionPlanSnapshot, o as ExactAnalystRunResult, r as ExactAnalystRunCompletion, s as ExactAnalystRunSummary, t as ExactAnalystBudgetSnapshot, u as ExactExecutionComponentIdentity } from "../exact-types-B0lJV3tu.js";
8
8
  import { A as ExactAnalystRunExecutionError, D as AnalystRegistryOptions, E as AnalystRegistry, M as RegistryRunOpts, N as assertExactRegistryRunOpts, O as BudgetPolicy, T as AnalystHooks, j as ExactRegistryRunOpts, k as ExactAnalystBudgetPolicy } from "../completion-verifier-D15NHYSk.js";
9
9
  import { C as scoreAnalystFindings, S as traceStoreEvidenceResolver, _ as AnalystIssueExpectation, a as AnalystBenchmarkLabelState, b as registryBenchmarkRunner, c as AnalystBenchmarkProvenance, d as AnalystBenchmarkSummary, f as AnalystEvidenceExpectation, g as AnalystFindingScore, h as AnalystEvidenceResolver, i as AnalystBenchmarkError, l as AnalystBenchmarkResult, m as AnalystEvidenceResolutionError, n as AnalystBenchmarkDatasetRef, o as AnalystBenchmarkObservation, p as AnalystEvidenceResolution, r as AnalystBenchmarkDescriptor, s as AnalystBenchmarkOutput, t as AnalystBenchmarkCase, u as AnalystBenchmarkRunner, v as AnalystLatencyDistribution, x as runAnalystBenchmark, y as RunAnalystBenchmarkOptions } from "../benchmark-DxaZfy0w.js";
10
- import { r as ExternalOptimizerRunnerCommand } from "../external-optimizer-contracts-nb7c_WAR.js";
10
+ import { l as ExternalOptimizerRunnerCommand } from "../external-optimizer-contracts-iK0yu4AR.js";
11
11
  import { a as TraceAnalysisEngineRequest, c as resolveTraceAnalystLimits, d as RawAnalystEvidence, f as RawAnalystEvidenceSchema, g as parseRawFinding, h as evidenceRefsFromRawFinding, i as TraceAnalysisEngine, l as ANALYST_SEVERITIES, m as RawAnalystFindingSchema, n as buildTraceToolsForGroup, o as TraceAnalysisEngineResult, p as RawAnalystFinding, r as DEFAULT_TRACE_ANALYST_LIMITS, s as TraceAnalystLimits, t as TraceToolGroupName, u as RAW_FINDING_SCHEMA_PROMPT } from "../tool-groups-Bqy4A3QB.js";
12
12
  //#region src/analyst/adapters.d.ts
13
13
  declare function liftSeverity(s: Severity): AnalystSeverity;
@@ -741,7 +741,7 @@ declare const ANALYST_BENCHMARK_HELP = "agent-eval analyst-benchmark\n\nRun the
741
741
  declare const ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM = "sha256-canonical-source-manifest";
742
742
  declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM = "sha256-canonical-file-manifest";
743
743
  declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES: readonly string[];
744
- declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "2867725e505a600f4f1f9d934fb797f5fb94a71c906577d05009aecc254f85f3";
744
+ declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "6327c48fa39d9fc6da2fc1dda5d89b08e16dd1de717c2eaf18c855ad5a4282c5";
745
745
  /** The published benchmark evidence was produced at this package version, by
746
746
  * the retired one-shot direct runner, before trace analysts moved to the
747
747
  * recursive DSPy RLM engine. Both evidence digests below are historical facts
@@ -753,7 +753,7 @@ declare const ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION = "0.137.0";
753
753
  declare const ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 = "1e03f2daed356d60316aabefb407ec1e437ac94d408d61eea4ae096e9c6fbb5b";
754
754
  declare const ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 = "4dba263b6256a30d56c7fdb2d992d3a953c0035d731f359b704db806f68f75ac";
755
755
  declare const ANALYST_BENCHMARK_IMPLEMENTATION_FILES: readonly string[];
756
- declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "b8b6d4f7a01f8341a0565159806f93ff9ca71c88322e63e4e52a86eed2c1d9c7";
756
+ declare const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "02ffe561616f79e62fcc3292ccd17cf55d5b385b8749367ce6b64e9f40586ac5";
757
757
  declare function analystBenchmarkImplementationDigest(): string;
758
758
  declare function analystBenchmarkDependencyLockDigest(): string;
759
759
  //#endregion
@@ -1 +1 @@
1
- {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;iBA4CgB,aAAa,GAAG,WAAgB;UAe/B,oBAAoB;EACnC;EACA;EACA,UAAU,mBAAmB;;;;;EAK7B,UAAU,KAAK,cAAc;;iBAGf,sBAAsB,KAAK,MAAM,oBAAoB,OAAO,QAAQ;UAwEnE;EACf;EACA;EACA,SAAS;;EAET;;iBAGc,uBAAuB,OAAM,uBAA4B,QAAQ;UAiEhE;EACf;EACA;EACA,OAAO;;EAEP,MAAM;;EAEN,OAAO;;EAEP;;iBAGc,mBAAmB,MAAM,mBAAmB,QAAQ;UAiDnD;EACf;EACA;;EAEA,UAAU,KAAK;;EAEf;;iBAGc,kCACd,OAAM,kCACL,QAAQ;;;cC5RE;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;iBAGK,4BACd,QAAQ,wBACR,2BACC;iBAkBa,iCAAiC,SAAS;;;KCtD9C;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;;;iBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;iBAoGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;iBA6Ea,yBAAyB;;iBAoNzB,iBAAiB;;;iBC7YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;iBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;iBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;iBA8Ec,yBACd,gBAAgB,2BACf;;;cChGU;KAED;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;iBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;iBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KChLU;;;;;;;;UASK;;WAEN;;WAEA;;UAGM;EACf;EACA;EACA;EACA;EACA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;EAGF,mBAAmB;;UAGJ;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;;;UCzEK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;iBAGO,8BAA8B,uBAAuB;iBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;iBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;iBCtQQ,wBACpB,eACC,QAAQ,MAAM;iBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;iBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;iBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;iBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;cCxKC;;;;cAKA;cAMA;;iBAiGG,4BAA4B,SAAS;;iBAerC,+BAA+B,SAAS;;;;iBAaxC,8BAA8B,SAAS;;;;iBChFvC,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;;;iBChCV,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;;KCnCd;UAoBK;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UCnEc;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBAGK,8BACd,QAAQ,yBACP;iBAca,mCAAmC,SAAS;;;UCrC3C;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;;IAEF;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;IACA;IACA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;cAGW;cACA;cACA;cACA;;;iBChIS,6BACpB,eACC,QAAQ;;;UCiEM;EACf,uBACE,SAAS,+BACT,QAAQ,sCACL,uBAAuB;;;;;;;;;;KAWlB;UAEK;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;;iBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;cAiTU;;;cCrbA;cAEA;cAEA;cAOA;;;;;;;;cAUA;cAEA;cAGA;cAGA;cAsFA;iBAGG;iBAIA;;;;iBCjHA,oCAAoC,eAAe;;iBAQnD,gCAAgC,eAAe;;;;;;;;;;;iBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;iBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;iBCER,gCACd,kBACA,uBAAuB,gCACtB;;;;;;;;;;;;;;;iBCGa,gBAAgB;;;;;iBAgBhB,WAAW;;;;;;;iBAeX,oBAAoB;;;;iBCXpB,kBAAkB,mBAAmB,WAAW;;;;;;iBAShD,uBACd,mBACA,mBACC,cAAc"}
1
+ {"version":3,"file":"index.d.ts","names":[],"sources":["../../src/analyst/adapters.ts","../../src/analyst/benchmark-agentrx-calibration.ts","../../src/analyst/benchmark-dataset-types.ts","../../src/analyst/benchmark-dataset-agentrx.ts","../../src/analyst/benchmark-dataset-codetrace.ts","../../src/analyst/benchmark-dataset-utils.ts","../../src/analyst/benchmark-verification-outcome.ts","../../src/analyst/benchmark-verification-artifacts.ts","../../src/analyst/benchmark-public-types.ts","../../src/analyst/benchmark-public-adapters.ts","../../src/analyst/benchmark-public-data.ts","../../src/analyst/benchmark-public-prompt.ts","../../src/analyst/benchmark-public-model.ts","../../src/analyst/benchmark-public-rlm.ts","../../src/analyst/benchmark-comparison.ts","../../src/analyst/benchmark-public-calibration.ts","../../src/analyst/benchmark-command-artifact.ts","../../src/analyst/benchmark-command-result.ts","../../src/analyst/benchmark-command.ts","../../src/analyst/benchmark-implementation.ts","../../src/analyst/benchmark-instructions-override.ts","../../src/analyst/benchmark-report.ts","../../src/analyst/benchmark-summary.ts","../../src/analyst/parse-tolerant.ts","../../src/analyst/proposal-findings.ts"],"mappings":";;;;;;;;;;;;iBA4CgB,aAAa,GAAG,WAAgB;UAe/B,oBAAoB;EACnC;EACA;EACA,UAAU,mBAAmB;;;;;EAK7B,UAAU,KAAK,cAAc;;iBAGf,sBAAsB,KAAK,MAAM,oBAAoB,OAAO,QAAQ;UAwEnE;EACf;EACA;EACA,SAAS;;EAET;;iBAGc,uBAAuB,OAAM,uBAA4B,QAAQ;UAiEhE;EACf;EACA;EACA,OAAO;;EAEP,MAAM;;EAEN,OAAO;;EAEP;;iBAGc,mBAAmB,MAAM,mBAAmB,QAAQ;UAiDnD;EACf;EACA;;EAEA,UAAU,KAAK;;EAEf;;iBAGc,kCACd,OAAM,kCACL,QAAQ;;;cC5RE;UAEI;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,SAAS;;iBAGK,4BACd,QAAQ,wBACR,2BACC;iBAkBa,iCAAiC,SAAS;;;KCtD9C;UAEK;EACf,YAAY;EACZ;EACA;EACA;EACA;EACA;;UAGe;EACf,eAAe;EACf,mBAAmB;EACnB;IAAe,YAAY;IAAY;;EACvC,wBAAwB;EACxB;EACA;EACA;;UAGe;EACf,UAAU;EACV;EACA;EACA;EACA;;UAGe;EACf,UAAU;EACV,mBAAmB;EACnB;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA;EACA,iBAAiB;;KAGP,0CAEC,sCACA,iCACA;UAEI;EACf;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA,oCAAoC;;UAGrB;EACf,eAAe;EACf,WAAW,sBAAsB;;KAGvB;UAEK;;;;;EAKf,WAAW;;UAGI,kCACP,yBACN;UAEa,oCAAoC;EACnD;;EAEA;;UAGe,yCAAyC;EACxD;EACA;EACA;EACA;;UAGe,2CACP,kCACN;;;iBCtGY,qBAAqB,QACnC,KAAK,YACL,OAAO,QACP,UAAS,8BACR,qBAAqB;;iBAoGR,6BACd,mBAAmB,YACnB,iBACA,UAAS,mCACR;iBA6Ea,yBAAyB;;iBAoNzB,iBAAiB;;;iBC7YjB,mBAAmB,QACjC,KAAK,mBACL,OAAO,QACP,UAAS,4BACR,qBAAqB;;iBAwFR,gCACd,2BACA,aAAa,uBACb,UAAS,qCACR;;;iBClHa,wBAAwB;;;KCA5B;UAEK;EACf;EACA;EACA,QAAQ;;UAGO;EACf,QAAQ;EACR;EAKA;IAAe;IAAe;;EAC9B,SAAS;EACT;EACA;EACA;EACA;;UAGe;EACf;EACA;;iBA8Ec,yBACd,gBAAgB,2BACf;;;cChGU;KAED;UAEK;EACf,MAAM;EACN;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;EACT;EACA;EACA;EACA;EACA;EACA,OAAO;EACP,cAAc;EACd,UAAU,OAAO;;UAGF;EACf,UAAU;EACV,SAAS;EACT,OAAO,MAAM;IAA6B;;;iBAatB,mCAAmC;EACvD;EACA,KAAK;EACL;IACE,QAAQ;iBAwHI,kCACd,kBACA,iBACA,WAAW,6BACX;;;KChLU;;;;;;;;UASK;;WAEN;;WAEA;;UAGM;EACf;EACA;EACA;EACA;EACA;;EAEA,UAAU;;EAEV;;EAEA,uBAAuB;EACvB;IACE,SAAS;IACT;IACA;IACA;IACA;;;;;;IAMA;;EAEF,aAAa;EACb;IACE;IACA;;;EAGF,mBAAmB;;UAGJ;EACf,OAAO,qBAAqB;EAC5B;EACA;EACA;EACA,YAAY;IACV;IACA;IACA;;EAEF,uBAAuB;EACvB,WAAW;;UAGI;EACf;EACA;EACA,QAAQ;;UAGO;EACf,OAAO;EACP,OAAO;EACP,OAAO;EACP,YAAY;EACZ,QAAQ;;UAGO;EACf;EACA;EACA;EACA;EACA;EACA;EACA,QAAQ;EACR,UAAU;;;;;;;;;;UCzEK;EACf;EACA;;EAEA;EACA;EACA,UAAU;EACV;EACA;EACA;EACA;EACA,WAAW;;;;;;;;;UAUI;EACf;EACA;;EAEA,kCAAkC;;EAElC;;EAEA;;EAEA;;EAEA;;EAEA;;;;;;;;UASe;EACf;EACA,OAAO;;iBAGO,8BAA8B,uBAAuB;iBAiB/C,6BAA6B;EACjD,SAAS;EACT;EACA,mBAAmB;EACnB;EACA,OAAO;EACP,SAAS;IACP;EACF,UAAU;EACV,aAAa;;EAEb,aAAa;;;;;;;;;iBA4MO,6BAA6B;EACjD;EACA,iBAAiB;EACjB,OAAO;EACP;EACA;EACA,SAAS;IACP;EACF,UAAU;EACV,aAAa;EACb,YAAY;;;;iBCtQQ,wBACpB,eACC,QAAQ,MAAM;iBA2BD,0BACd,SAAS,+BACT,eAAe,2BACf;EAAW;EAAe;IACzB,MAAM;iBAwBO,6BACd,SAAS,+BACT,eAAe,4BACd;iBAyCa,+BACd,SAAS,+BACT,iBAAiB,2BACjB,mBAAmB,2BACnB,eACC;iBAcmB,8BAA8B;EAClD,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;IACE,QAAQ;;;;;;;cCxKC;;;;cAKA;cAMA;;iBAiGG,4BAA4B,SAAS;;iBAerC,+BAA+B,SAAS;;;;iBAaxC,8BAA8B,SAAS;;;;iBChFvC,kCACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;;;iBChCV,+BACd,SAAS,+BACT,QAAQ,oCACP,uBAAuB;;;KCnCd;UAoBK;EACf,QAAQ;EACR;;EAEA;;EAEA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBASK,sBACd,QAAQ,wBACR;EACE;EACA;EACA;EACA;EACA;IAED;;;UCnEc;EACf;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;EACA;;UAGe;EACf;EACA;EACA,SAAS;;iBAGK,8BACd,QAAQ,yBACP;iBAca,mCAAmC,SAAS;;;UCrC3C;EACf;EACA;EACA;IACE,SAAS;IACT;IACA;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D,uBAAuB;IACvB,0BAA0B;IAC1B;MACE;MACA;MACA;MACA,QAAQ;;IAEV;MACE;MACA;;MAEA;MACA;MACA;MACA;MACA;MACA;MACA;;MAEA;MACA;MACA;;;EAGJ,QAAQ;EACR,aAAa;EACb,uBAAuB;EACvB,qBAAqB;;UAGN;EACf;EACA;EACA;EACA;IACE;IACA;IACA;;;UAIa;EACf;IACE,SAAS;IACT;IACA;IACA;MACE;MACA;MACA;;IAEF;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;IAEA;IACA;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA,YAAY;MAAQ;MAAiB;MAAsB;;IAC3D;IACA;;;UAIa;EACf;EACA;EACA;EACA;EACA,UAAU;;UAGK;EACf;EACA;EACA;EACA;IACE;IACA;IACA;IACA;IACA;IACA;;EAEF;EACA;IACE;IACA;IACA;;EAEF;IACE;IACA;IACA;IACA;IACA;IACA;;;UAIa;EACf;EACA;EACA;EACA,aAAa;EACb;;cAGW;cACA;cACA;cACA;;;iBChIS,6BACpB,eACC,QAAQ;;;UCiEM;EACf,uBACE,SAAS,+BACT,QAAQ,sCACL,uBAAuB;;;;;;;;;;KAWlB;UAEK;EACf,SAAS;EACT,SAAS;EACT;EACA;EACA;EACA;EACA;EACA;EACA,OAAO;EACP;EACA;EACA;EACA;;EAEA;EACA;EACA;EACA;EACA;EACA;;iBAKoB,2BACpB,yBACA,MAAK,OAAO,YACZ,eAAc,sCACb;cAiTU;;;cCrbA;cAEA;cAEA;cAOA;;;;;;;;cAUA;cAEA;cAGA;cAGA;cAuFA;iBAGG;iBAIA;;;;iBClHA,oCAAoC,eAAe;;iBAQnD,gCAAgC,eAAe;;;;;;;;;;;iBAyB/C,+BACd,SAAS,+BACT,WAAW,KAAK;;;iBCzCF,+BACd,QAAQ,wBACR,uBAAsB;;;iBCER,gCACd,kBACA,uBAAuB,gCACtB;;;;;;;;;;;;;;;iBCGa,gBAAgB;;;;;iBAgBhB,WAAW;;;;;;;iBAeX,oBAAoB;;;;iBCXpB,kBAAkB,mBAAmB,WAAW;;;;;;iBAShD,uBACd,mBACA,mBACC,cAAc"}
@@ -1,12 +1,12 @@
1
1
  import { i as CostLedger } from "../cost-ledger-DMFxsLKr.js";
2
- import { C as createChatClient, _ as CONTROL_INTEGRITY_ANALYST, b as behavioralAnalyst, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, t as buildDefaultAnalystRegistry, v as ControlIntegrityAnalyst, x as deriveEfficiencyFindings, y as emitControlIntegrityFindings } from "../default-registry-BupFSXSN.js";
2
+ import { C as createChatClient, _ as CONTROL_INTEGRITY_ANALYST, b as behavioralAnalyst, f as DEFAULT_TRACE_ANALYST_KINDS, g as FAILURE_MODE_KIND_SPEC, h as IMPROVEMENT_KIND_SPEC, i as assertExactRegistryRunOpts, m as KNOWLEDGE_GAP_KIND_SPEC, n as AnalystRegistry, p as KNOWLEDGE_POISONING_KIND_SPEC, r as ExactAnalystRunExecutionError, t as buildDefaultAnalystRegistry, v as ControlIntegrityAnalyst, x as deriveEfficiencyFindings, y as emitControlIntegrityFindings } from "../default-registry-DfHJEwYh.js";
3
3
  import { A as coerceJson, B as renderFindingSubject, D as RawAnalystFindingSchema, E as RawAnalystEvidenceSchema, F as FINDING_SUBJECT_SYNTAX, G as resolveTraceAnalystLimits, I as FindingSubjectStringSchema, L as KIND_EXPECTED_SUBJECTS, M as stripCodeFences, N as FINDING_SUBJECT_GRAMMAR_PROMPT, O as evidenceRefsFromRawFinding, P as FINDING_SUBJECT_KINDS, R as findingSubjectGrammarPromptFor, T as RAW_FINDING_SCHEMA_PROMPT, W as DEFAULT_TRACE_ANALYST_LIMITS, a as buildTraceToolsForGroup, i as runTraceAnalyst, j as coerceToFindingRows, k as parseRawFinding, n as renderPriorFindings, r as renderUpstreamFindings, t as createTraceAnalyst, w as ANALYST_SEVERITIES, z as parseFindingSubject } from "../kind-factory-Bvwe3pup.js";
4
4
  import { a as computeFindingId, i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger, o as makeFinding, s as makeProposalFinding } from "../usage-receipt-CgxMEBZq.js";
5
5
  import { n as isProposalFinding, t as assertProposalFindings } from "../proposal-findings-2GIUo1et.js";
6
6
  import { a as RunCritic, c as buildSkillUsageReport, d as defaultIsMaterial, f as diffFindings, h as defineTraceAnalyst, i as runSemanticConceptJudge, l as emitSkillUsageFindings, m as defineCustomAnalyst, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, s as SkillUsageAnalyst, u as FindingsStore } from "../semantic-concept-judge-DJQtFr95.js";
7
- import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-CJ4sN8Wx.js";
7
+ import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-CBFwlyaY.js";
8
8
  import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-CYtcIF2V.js";
9
- import { $ as normalizeAgentRxCategory, A as parseVerificationOutcome, B as analystBenchmarkDependencyLockDigest, C as MAX_INCORRECT_BLOCK_STEPS, D as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, E as publicBenchmarkSystemPrompt, F as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, G as ANALYST_BENCHMARK_OBSERVATIONS_FILE, H as ANALYST_BENCHMARK_COST_LEDGER_FILE, I as ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, J as summarizeAgentRxCalibration, K as AGENT_RX_UPSTREAM_REVISION, L as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, M as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, N as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, O as appendVerificationArtifactsToOtlp, P as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, Q as agentRxPredictionsToFindings, R as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, S as MAX_INCORRECT_BLOCKS, T as publicBenchmarkRlmInstructions, U as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, V as analystBenchmarkImplementationDigest, W as ANALYST_BENCHMARK_MANIFEST_FILE, X as codeTracerPredictionsToFindings, Y as codeTraceBenchCase, Z as agentRxBenchmarkCase, _ as compareAnalystRunners, a as preparePublicAnalystBenchmark, b as readAnalystInstructionsOverride, c as selectPublicBenchmarkRows, d as adaptPublicBenchmarkFindings, et as roundAgentRxStep, f as emptyPublicBenchmarkRunner, g as summarizeCodeTraceCalibration, h as renderCodeTraceCalibrationMarkdown, i as loadPublicBenchmarkRows, j as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, k as loadCodeTraceVerificationArtifacts, l as createPublicBenchmarkRlmRunner, m as readAnalystBenchmarkArtifact, n as runAnalystBenchmarkCommand, o as publicBenchmarkDistributions, p as expandCodeTraceFailureBlocks, q as renderAgentRxCalibrationMarkdown, r as renderAnalystBenchmarkMarkdown, s as publicBenchmarkSelectionReport, t as ANALYST_BENCHMARK_HELP, tt as normalizeBenchmarkLabel, u as createPublicBenchmarkDirectRunner, v as analystInstructionsOverrideFromText, w as publicBenchmarkProtocolSha256, x as CODE_TRACE_BENCH_ANALYST_PROMPT, y as effectiveAnalystProtocolSha256, z as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 } from "../benchmark-command-CSddvlq7.js";
9
+ import { $ as normalizeAgentRxCategory, A as parseVerificationOutcome, B as analystBenchmarkDependencyLockDigest, C as MAX_INCORRECT_BLOCK_STEPS, D as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, E as publicBenchmarkSystemPrompt, F as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, G as ANALYST_BENCHMARK_OBSERVATIONS_FILE, H as ANALYST_BENCHMARK_COST_LEDGER_FILE, I as ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, J as summarizeAgentRxCalibration, K as AGENT_RX_UPSTREAM_REVISION, L as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, M as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, N as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, O as appendVerificationArtifactsToOtlp, P as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, Q as agentRxPredictionsToFindings, R as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, S as MAX_INCORRECT_BLOCKS, T as publicBenchmarkRlmInstructions, U as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, V as analystBenchmarkImplementationDigest, W as ANALYST_BENCHMARK_MANIFEST_FILE, X as codeTracerPredictionsToFindings, Y as codeTraceBenchCase, Z as agentRxBenchmarkCase, _ as compareAnalystRunners, a as preparePublicAnalystBenchmark, b as readAnalystInstructionsOverride, c as selectPublicBenchmarkRows, d as adaptPublicBenchmarkFindings, et as roundAgentRxStep, f as emptyPublicBenchmarkRunner, g as summarizeCodeTraceCalibration, h as renderCodeTraceCalibrationMarkdown, i as loadPublicBenchmarkRows, j as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, k as loadCodeTraceVerificationArtifacts, l as createPublicBenchmarkRlmRunner, m as readAnalystBenchmarkArtifact, n as runAnalystBenchmarkCommand, o as publicBenchmarkDistributions, p as expandCodeTraceFailureBlocks, q as renderAgentRxCalibrationMarkdown, r as renderAnalystBenchmarkMarkdown, s as publicBenchmarkSelectionReport, t as ANALYST_BENCHMARK_HELP, tt as normalizeBenchmarkLabel, u as createPublicBenchmarkDirectRunner, v as analystInstructionsOverrideFromText, w as publicBenchmarkProtocolSha256, x as CODE_TRACE_BENCH_ANALYST_PROMPT, y as effectiveAnalystProtocolSha256, z as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 } from "../benchmark-command-4c7N_rlw.js";
10
10
  //#region src/analyst/adapters.ts
11
11
  /**
12
12
  * Adapter factories — lift each existing agent-eval primitive into the
@@ -5,8 +5,8 @@ import { c as callLlmJson, f as maximumChargeForLlmRequest, l as costReceiptFrom
5
5
  import { O as evidenceRefsFromRawFinding, h as TRACE_ANALYSIS_LIMITS, i as runTraceAnalyst } from "./kind-factory-Bvwe3pup.js";
6
6
  import { o as makeFinding, r as usageReceiptFromCostLedger } from "./usage-receipt-CgxMEBZq.js";
7
7
  import { i as hashCanonical, r as canonicalString } from "./canonical-D011XM8r.js";
8
- import { n as createRunCostLedger, r as fsCampaignStorage, t as acquireSingleRunLock } from "./single-run-lock-Do-8vBIJ.js";
9
- import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-CJ4sN8Wx.js";
8
+ import { n as createRunCostLedger, r as fsCampaignStorage, t as acquireSingleRunLock } from "./single-run-lock-t1si1ob7.js";
9
+ import { t as createDspyRlmTraceEngine } from "./dspy-rlm-engine-CBFwlyaY.js";
10
10
  import { c as writeLedgerFileAtomically, s as withLedgerFileLock } from "./ledger-core-Dxz0Rkwa.js";
11
11
  import { E as pairedBootstrap } from "./statistics-ByxzSiOM.js";
12
12
  import { i as otlpTextToTraceAnalysisStore, r as createOtlpBufferTraceStore, t as DEFAULT_MAX_TRACE_FILE_BYTES } from "./store-otlp-D4I90_vR.js";
@@ -1173,7 +1173,7 @@ const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES = Object.freeze([
1173
1173
  "package.json",
1174
1174
  "pnpm-lock.yaml"
1175
1175
  ]);
1176
- const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "2867725e505a600f4f1f9d934fb797f5fb94a71c906577d05009aecc254f85f3";
1176
+ const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "6327c48fa39d9fc6da2fc1dda5d89b08e16dd1de717c2eaf18c855ad5a4282c5";
1177
1177
  /** The published benchmark evidence was produced at this package version, by
1178
1178
  * the retired one-shot direct runner, before trace analysts moved to the
1179
1179
  * recursive DSPy RLM engine. Both evidence digests below are historical facts
@@ -1267,9 +1267,10 @@ const ANALYST_BENCHMARK_IMPLEMENTATION_FILES = Object.freeze([
1267
1267
  "src/trace-analyst/types.ts",
1268
1268
  "src/trace/attribute-vocabulary.ts",
1269
1269
  "src/trace/otlp-attributes.ts",
1270
- "src/trace/raw-provider-sink.ts"
1270
+ "src/trace/raw-provider-sink.ts",
1271
+ "src/verdict-cache.ts"
1271
1272
  ]);
1272
- const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "b8b6d4f7a01f8341a0565159806f93ff9ca71c88322e63e4e52a86eed2c1d9c7";
1273
+ const ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 = "02ffe561616f79e62fcc3292ccd17cf55d5b385b8749367ce6b64e9f40586ac5";
1273
1274
  function analystBenchmarkImplementationDigest() {
1274
1275
  return ANALYST_BENCHMARK_IMPLEMENTATION_SHA256;
1275
1276
  }
@@ -5247,4 +5248,4 @@ function shellQuote(value) {
5247
5248
  //#endregion
5248
5249
  export { normalizeAgentRxCategory as $, parseVerificationOutcome as A, analystBenchmarkDependencyLockDigest as B, MAX_INCORRECT_BLOCK_STEPS as C, DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES as D, publicBenchmarkSystemPrompt as E, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 as F, ANALYST_BENCHMARK_OBSERVATIONS_FILE as G, ANALYST_BENCHMARK_COST_LEDGER_FILE as H, ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION as I, summarizeAgentRxCalibration as J, AGENT_RX_UPSTREAM_REVISION as K, ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM as L, ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES as M, ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 as N, appendVerificationArtifactsToOtlp as O, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 as P, agentRxPredictionsToFindings as Q, ANALYST_BENCHMARK_IMPLEMENTATION_FILES as R, MAX_INCORRECT_BLOCKS as S, publicBenchmarkRlmInstructions as T, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE as U, analystBenchmarkImplementationDigest as V, ANALYST_BENCHMARK_MANIFEST_FILE as W, codeTracerPredictionsToFindings as X, codeTraceBenchCase as Y, agentRxBenchmarkCase as Z, compareAnalystRunners as _, preparePublicAnalystBenchmark as a, readAnalystInstructionsOverride as b, selectPublicBenchmarkRows as c, adaptPublicBenchmarkFindings as d, roundAgentRxStep as et, emptyPublicBenchmarkRunner as f, summarizeCodeTraceCalibration as g, renderCodeTraceCalibrationMarkdown as h, loadPublicBenchmarkRows as i, ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM as j, loadCodeTraceVerificationArtifacts as k, createPublicBenchmarkRlmRunner as l, readAnalystBenchmarkArtifact as m, runAnalystBenchmarkCommand as n, publicBenchmarkDistributions as o, expandCodeTraceFailureBlocks as p, renderAgentRxCalibrationMarkdown as q, renderAnalystBenchmarkMarkdown as r, publicBenchmarkSelectionReport as s, ANALYST_BENCHMARK_HELP as t, normalizeBenchmarkLabel as tt, createPublicBenchmarkDirectRunner as u, analystInstructionsOverrideFromText as v, publicBenchmarkProtocolSha256 as w, CODE_TRACE_BENCH_ANALYST_PROMPT as x, effectiveAnalystProtocolSha256 as y, ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 as z };
5249
5250
 
5250
- //# sourceMappingURL=benchmark-command-CSddvlq7.js.map
5251
+ //# sourceMappingURL=benchmark-command-4c7N_rlw.js.map