@tangle-network/agent-eval 0.202.0 → 0.202.2
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.
- package/CHANGELOG.md +11 -0
- package/dist/adapters/http.d.ts +2 -2
- package/dist/analyst/index.d.ts +5 -5
- package/dist/analyst/index.js +1 -1
- package/dist/benchmarks/index.d.ts +4 -4
- package/dist/benchmarks/index.js +1 -1
- package/dist/campaign/index.d.ts +5 -5
- package/dist/campaign/index.js +3 -3
- package/dist/{campaign-XXmOU0R0.js → campaign-Btk1yFfw.js} +130 -20
- package/dist/{campaign-XXmOU0R0.js.map → campaign-Btk1yFfw.js.map} +1 -1
- package/dist/{claude-jsonl-CxZZrDJ3.js → claude-jsonl-DoQJBCVR.js} +49 -15
- package/dist/claude-jsonl-DoQJBCVR.js.map +1 -0
- package/dist/cli.js +1 -1
- package/dist/contract/index.d.ts +7 -7
- package/dist/contract/index.js +1 -1
- package/dist/{default-registry-D4qXx2aG.js → default-registry-6QOiTzqi.js} +2 -2
- package/dist/{default-registry-D4qXx2aG.js.map → default-registry-6QOiTzqi.js.map} +1 -1
- package/dist/{default-registry-D1rNS2yB.d.ts → default-registry-DesnDBvl.d.ts} +3 -3
- package/dist/{default-registry-D1rNS2yB.d.ts.map → default-registry-DesnDBvl.d.ts.map} +1 -1
- package/dist/{define-agent-eval-DQXLPqWc.d.ts → define-agent-eval-BDPV3ieA.d.ts} +5 -5
- package/dist/{define-agent-eval-DQXLPqWc.d.ts.map → define-agent-eval-BDPV3ieA.d.ts.map} +1 -1
- package/dist/{engine-DlKIzDKo.d.ts → engine-neXmlmTu.d.ts} +2 -2
- package/dist/{engine-DlKIzDKo.d.ts.map → engine-neXmlmTu.d.ts.map} +1 -1
- package/dist/evaluation.d.ts +1 -1
- package/dist/experiment/index.d.ts +1 -1
- package/dist/{exporters-XMCYJlqT.d.ts → exporters-DjCu6-lX.d.ts} +2 -2
- package/dist/{exporters-XMCYJlqT.d.ts.map → exporters-DjCu6-lX.d.ts.map} +1 -1
- package/dist/{exporters-Coiv-qO2.js → exporters-ISzpUY3e.js} +10 -4
- package/dist/{exporters-Coiv-qO2.js.map → exporters-ISzpUY3e.js.map} +1 -1
- package/dist/{hf-dataset-kjIfIK_e.js → hf-dataset-DdT7nNkf.js} +3 -3
- package/dist/{hf-dataset-kjIfIK_e.js.map → hf-dataset-DdT7nNkf.js.map} +1 -1
- package/dist/{index-D4pC7Tvd.d.ts → index-BCBpEreJ.d.ts} +18 -14
- package/dist/{index-D4pC7Tvd.d.ts.map → index-BCBpEreJ.d.ts.map} +1 -1
- package/dist/{index-giYkbitB.d.ts → index-BPCdNSMW.d.ts} +20 -5
- package/dist/index-BPCdNSMW.d.ts.map +1 -0
- package/dist/{index-Bi2fLITd.d.ts → index-BtcxmabC.d.ts} +4 -4
- package/dist/{index-Bi2fLITd.d.ts.map → index-BtcxmabC.d.ts.map} +1 -1
- package/dist/{index-D1ZX28rV.d.ts → index-CnKgaHKA.d.ts} +6 -6
- package/dist/{index-D1ZX28rV.d.ts.map → index-CnKgaHKA.d.ts.map} +1 -1
- package/dist/index.d.ts +12 -12
- package/dist/index.js +3 -3
- package/dist/{integrity-uwK7XC-8.js → integrity-atFlABhR.js} +2 -2
- package/dist/{integrity-uwK7XC-8.js.map → integrity-atFlABhR.js.map} +1 -1
- package/dist/{jev-protocol-CpKI7XSu.d.ts → jev-protocol-BQLxwbqD.d.ts} +24 -7
- package/dist/jev-protocol-BQLxwbqD.d.ts.map +1 -0
- package/dist/jev-protocol.d.ts +2 -2
- package/dist/jev-protocol.js +19 -4
- package/dist/jev-protocol.js.map +1 -1
- package/dist/{jev-review-D4ynCORp.d.ts → jev-review-DX6MI_jq.d.ts} +2 -2
- package/dist/{jev-review-D4ynCORp.d.ts.map → jev-review-DX6MI_jq.d.ts.map} +1 -1
- package/dist/jev.d.ts +4 -4
- package/dist/jev.js +2 -2
- package/dist/{matrix--IZKROxD.d.ts → matrix-DRXCV70M.d.ts} +2 -2
- package/dist/{matrix--IZKROxD.d.ts.map → matrix-DRXCV70M.d.ts.map} +1 -1
- package/dist/{mint-C9Lojt1V.d.ts → mint-BFTrlq4t.d.ts} +10 -2
- package/dist/mint-BFTrlq4t.d.ts.map +1 -0
- package/dist/{mint-vlRckJEX.js → mint-DE6nQrX_.js} +26 -7
- package/dist/mint-DE6nQrX_.js.map +1 -0
- package/dist/multishot/golden/index.d.ts +1 -1
- package/dist/multishot/index.d.ts +2 -2
- package/dist/openapi.json +1 -1
- package/dist/rl.d.ts +6 -6
- package/dist/rl.js +3 -3
- package/dist/rollout/index.d.ts +5 -5
- package/dist/rollout/index.js +6 -6
- package/dist/{rollout-BUQixCoc.js → rollout-DTunVdJE.js} +41 -21
- package/dist/{rollout-BUQixCoc.js.map → rollout-DTunVdJE.js.map} +1 -1
- package/dist/{schema-B5LtYvn9.d.ts → schema-Bp8S20Ii.d.ts} +25 -5
- package/dist/{schema-B5LtYvn9.d.ts.map → schema-Bp8S20Ii.d.ts.map} +1 -1
- package/dist/{schema-DAMp77zB.js → schema-D08IIzui.js} +27 -1
- package/dist/{schema-DAMp77zB.js.map → schema-D08IIzui.js.map} +1 -1
- package/dist/search/index.d.ts +2 -2
- package/dist/{search-kernel-C9CXvFR7.d.ts → search-kernel-zI-rsHoG.d.ts} +2 -2
- package/dist/{search-kernel-C9CXvFR7.d.ts.map → search-kernel-zI-rsHoG.d.ts.map} +1 -1
- package/dist/{statistical-heldout-HswMvzbP.d.ts → statistical-heldout-BAbd_oXC.d.ts} +2 -2
- package/dist/{statistical-heldout-HswMvzbP.d.ts.map → statistical-heldout-BAbd_oXC.d.ts.map} +1 -1
- package/dist/{store-tool-spans-Dv9x4XYG.d.ts → store-tool-spans-CKRw7JHa.d.ts} +3 -3
- package/dist/{store-tool-spans-Dv9x4XYG.d.ts.map → store-tool-spans-CKRw7JHa.d.ts.map} +1 -1
- package/dist/supervisor-run/index.d.ts +2 -2
- package/dist/supervisor-run/index.d.ts.map +1 -1
- package/dist/supervisor-run/index.js +22 -14
- package/dist/supervisor-run/index.js.map +1 -1
- package/dist/{tool-groups-BY1MdSIE.d.ts → tool-groups-V23aGP_g.d.ts} +2 -2
- package/dist/tool-groups-V23aGP_g.d.ts.map +1 -0
- package/dist/{trace-questions-DnqdmwyV.d.ts → trace-questions-B-Wf0W3Q.d.ts} +2 -2
- package/dist/{trace-questions-DnqdmwyV.d.ts.map → trace-questions-B-Wf0W3Q.d.ts.map} +1 -1
- package/dist/trace-repair/index.d.ts +1 -1
- package/dist/traces.d.ts +2 -2
- package/dist/{types-DfvQOe91.d.ts → types-BR1LLIxm.d.ts} +2 -2
- package/dist/{types-DfvQOe91.d.ts.map → types-BR1LLIxm.d.ts.map} +1 -1
- package/dist/{types-C4hH69a9.d.ts → types-BbiT5NPH.d.ts} +2 -2
- package/dist/{types-C4hH69a9.d.ts.map → types-BbiT5NPH.d.ts.map} +1 -1
- package/dist/{types-DRNlLbBR.d.ts → types-DUlbOCwL.d.ts} +2 -2
- package/dist/{types-DRNlLbBR.d.ts.map → types-DUlbOCwL.d.ts.map} +1 -1
- package/docs/campaign-coverage.md +20 -0
- package/docs/changes/jev-recorded-evidence.md +24 -0
- package/docs/jev-replay.md +50 -0
- package/docs/retained-dispatch.md +56 -0
- package/docs/rollout.md +3 -1
- package/package.json +1 -1
- package/dist/claude-jsonl-CxZZrDJ3.js.map +0 -1
- package/dist/index-giYkbitB.d.ts.map +0 -1
- package/dist/jev-protocol-CpKI7XSu.d.ts.map +0 -1
- package/dist/mint-C9Lojt1V.d.ts.map +0 -1
- package/dist/mint-vlRckJEX.js.map +0 -1
- package/dist/tool-groups-BY1MdSIE.d.ts.map +0 -1
package/CHANGELOG.md
CHANGED
|
@@ -4,6 +4,17 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
|
|
|
4
4
|
|
|
5
5
|
---
|
|
6
6
|
|
|
7
|
+
## [0.202.2] — 2026-09-29
|
|
8
|
+
|
|
9
|
+
- fix(rollout): preserve attempt evidence and mark lossy projections (#900)
|
|
10
|
+
- feat(campaign): expose native coverage for product adapters (#899)
|
|
11
|
+
- docs(jev): ship executable offline inspection of recorded evaluations (#898)
|
|
12
|
+
- feat(jev): preserve recorded evidence separately from live admission (#897)
|
|
13
|
+
|
|
14
|
+
## [0.202.1] — 2026-09-28
|
|
15
|
+
|
|
16
|
+
- fix(campaign): prevent lost history and changed decoders from resetting retained dispatch (#893)
|
|
17
|
+
|
|
7
18
|
## [0.202.0] — 2026-09-28
|
|
8
19
|
|
|
9
20
|
- feat(meta-eval): audit planning and canonical schemas for evaluator construction (#888)
|
package/dist/adapters/http.d.ts
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { m as CostReceipt } from "../cost-ledger-DbQdN3nO.js";
|
|
2
|
-
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-
|
|
3
|
-
import "../index-
|
|
2
|
+
import { R as Scenario, d as DispatchContext, f as DispatchFn } from "../types-BR1LLIxm.js";
|
|
3
|
+
import "../index-CnKgaHKA.js";
|
|
4
4
|
//#region src/adapters/http.d.ts
|
|
5
5
|
export interface HttpDispatchOptions<TScenario extends Scenario, _TArtifact> {
|
|
6
6
|
/** Static endpoint URL. Mutually exclusive with `resolveUrl`. */
|
package/dist/analyst/index.d.ts
CHANGED
|
@@ -4,13 +4,13 @@ import { A as createChatClient, C as ChatResponse, D as MockTransportOpts, E as
|
|
|
4
4
|
import { A as runAnalystBenchmark, C as AnalystEvidenceResolutionError, D as AnalystLatencyDistribution, E as AnalystIssueExpectation, M as scoreAnalystFindings, O as RunAnalystBenchmarkOptions, S as AnalystEvidenceResolution, T as AnalystFindingScore, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, _ as AnalystBenchmarkProvenance, b as AnalystBenchmarkSummary, d as AnalystBenchmarkDatasetRef, f as AnalystBenchmarkDescriptor, g as AnalystBenchmarkOutput, h as AnalystBenchmarkObservation, j as traceStoreEvidenceResolver, k as registryBenchmarkRunner, m as AnalystBenchmarkLabelState, p as AnalystBenchmarkError, q as ExternalOptimizerRunnerCommand, t as AgentProfile, u as AnalystBenchmarkCase, v as AnalystBenchmarkResult, w as AnalystEvidenceResolver, x as AnalystEvidenceExpectation, y as AnalystBenchmarkRunner } from "../agent-profile-DDfIjdyS.js";
|
|
5
5
|
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-KGxBWH4b.js";
|
|
6
6
|
import { a as ExactAnalystBudgetPolicy, c as RegistryRunOpts, i as BudgetPolicy, n as AnalystRegistry, o as ExactAnalystRunExecutionError, r as AnalystRegistryOptions, s as ExactRegistryRunOpts, t as AnalystHooks } from "../registry-Bu3lNole.js";
|
|
7
|
-
import { a as createTraceAnalyst, c as BehavioralAnalystOptions, i as TraceAnalystDefinition, l as behavioralAnalyst, n as buildDefaultAnalystRegistry, o as renderPriorFindings, r as CreateTraceAnalystOptions, s as runTraceAnalyst, t as DefaultAnalystRegistryOptions, u as deriveEfficiencyFindings } from "../default-registry-
|
|
8
|
-
import { a as TraceAnalystLimits, c as RawAnalystEvidence, d as evidenceRefsFromRawFinding, f as parseRawFinding, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, o as resolveTraceAnalystLimits, r as TraceAnalysisEngineRequest, s as RAW_FINDING_SCHEMA_PROMPT, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, u as RawAnalystFindingSchema } from "../engine-
|
|
9
|
-
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-
|
|
10
|
-
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, E as createChatTraceEngine, S as renderFindingSubject, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, b as findingSubjectGrammarPromptFor, c as emitControlIntegrityFindings, d as FindingsStore, f as PersistedFinding, g as FINDING_SUBJECT_SYNTAX, h as FINDING_SUBJECT_KINDS, i as IMPROVEMENT_KIND_SPEC, j as SemanticConceptJudgeOptions, l as DiffPolicy, m as diffFindings, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as defaultIsMaterial, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine, x as parseFindingSubject, y as KIND_EXPECTED_SUBJECTS } from "../index-
|
|
7
|
+
import { a as createTraceAnalyst, c as BehavioralAnalystOptions, i as TraceAnalystDefinition, l as behavioralAnalyst, n as buildDefaultAnalystRegistry, o as renderPriorFindings, r as CreateTraceAnalystOptions, s as runTraceAnalyst, t as DefaultAnalystRegistryOptions, u as deriveEfficiencyFindings } from "../default-registry-DesnDBvl.js";
|
|
8
|
+
import { a as TraceAnalystLimits, c as RawAnalystEvidence, d as evidenceRefsFromRawFinding, f as parseRawFinding, i as TraceAnalysisEngineResult, l as RawAnalystFinding, n as TraceAnalysisEngine, o as resolveTraceAnalystLimits, r as TraceAnalysisEngineRequest, s as RAW_FINDING_SCHEMA_PROMPT, t as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, u as RawAnalystFindingSchema } from "../engine-neXmlmTu.js";
|
|
9
|
+
import { n as buildTraceToolsForGroup, t as TraceToolGroupName } from "../tool-groups-V23aGP_g.js";
|
|
10
|
+
import { A as SemanticConceptJudgeInput, C as DspyRlmTraceEngineOptions, E as createChatTraceEngine, S as renderFindingSubject, T as ChatTraceEngineOptions, _ as FindingSubject, a as FAILURE_MODE_KIND_SPEC, b as findingSubjectGrammarPromptFor, c as emitControlIntegrityFindings, d as FindingsStore, f as PersistedFinding, g as FINDING_SUBJECT_SYNTAX, h as FINDING_SUBJECT_KINDS, i as IMPROVEMENT_KIND_SPEC, j as SemanticConceptJudgeOptions, l as DiffPolicy, m as diffFindings, n as KNOWLEDGE_POISONING_KIND_SPEC, o as CONTROL_INTEGRITY_ANALYST, p as defaultIsMaterial, r as KNOWLEDGE_GAP_KIND_SPEC, s as ControlIntegrityAnalyst, t as DEFAULT_TRACE_ANALYST_KINDS, u as FindingsDiff, v as FindingSubjectKind, w as createDspyRlmTraceEngine, x as parseFindingSubject, y as KIND_EXPECTED_SUBJECTS } from "../index-BtcxmabC.js";
|
|
11
11
|
import { A as ReplyRowDecoded, C as primeProtocolSha256, D as DecodedReply, E as runPrimeExchange, M as decodeReplyRows, O as ReplyContract, S as normalizePrimeUsage, T as projectPrimeTrajectory, _ as buildPrimePrompt, a as PrimeProjectionOutcome, b as extractPrimeJsonObject, c as PrimeProtocolIdentity, d as PrimeRepairPromptSpec, f as PrimeRepairState, g as analystUsageReceiptFromPrimeUsage, h as PrimeTurnRecord, i as PrimeProjectionDelivery, j as ReplyRowRejection, k as ReplyEnvelope, l as PrimeRawUsage, m as PrimeRowDecoded, n as PrimeExchangeOutcome, o as PrimeProjectionSource, p as PrimeReplyContract, r as PrimeFailure, s as PrimePromptSpec, t as PrimeExchangeOptions, u as PrimeRejectedRow, v as buildPrimeRepairPrompt, w as primeReplyDefect, x as mergePrimeRawUsage, y as emptyPrimeRawUsage } from "../prime-protocol-T0i6xcAs.js";
|
|
12
12
|
import { i as nodeHttpPrimeBridgeTransport, n as PrimeBridgeTransportRequest, r as PrimeBridgeTransportResult, t as PrimeBridgeTransport } from "../prime-bridge-transport-6feEglLf.js";
|
|
13
|
-
import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, o as citedSpansOnly, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-
|
|
13
|
+
import { a as askTraceQuestions, i as VerifiedTraceFinding, n as TraceQuestionOutcome, o as citedSpansOnly, r as TraceQuestionSpec, t as AskTraceQuestionsOptions } from "../trace-questions-B-Wf0W3Q.js";
|
|
14
14
|
import { z } from "zod";
|
|
15
15
|
//#region src/analyst/adapters.d.ts
|
|
16
16
|
interface SemanticConceptJudgeAdapterOpts {
|
package/dist/analyst/index.js
CHANGED
|
@@ -2,7 +2,7 @@ import { i as CostLedger } from "../cost-ledger-_IbkejCF.js";
|
|
|
2
2
|
import { n as assertProposalFindings, r as isProposalFinding } from "../abort-signal-CtzAM_sJ.js";
|
|
3
3
|
import { B as FINDING_SUBJECT_SYNTAX, F as parseRawFinding, G as DEFAULT_TRACE_ANALYST_OUTPUT_TOKENS, H as findingSubjectGrammarPromptFor, I as parseTraceSpanEvidenceUri, K as resolveTraceAnalystLimits, L as coerceJson, M as RAW_FINDING_SCHEMA_PROMPT, N as RawAnalystFindingSchema, P as evidenceRefsFromRawFinding, R as stripCodeFences, U as parseFindingSubject, V as KIND_EXPECTED_SUBJECTS, W as renderFindingSubject, i as buildTraceToolsForGroup, n as renderPriorFindings, r as runTraceAnalyst, t as createTraceAnalyst, z as FINDING_SUBJECT_KINDS } from "../kind-factory-CdPPHtGZ.js";
|
|
4
4
|
import { i as validateUsageSettlementTimeout, n as settleUsageReceiptFromCostLedger } from "../usage-receipt-Cx3xNeyB.js";
|
|
5
|
-
import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as deriveEfficiencyFindings, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-
|
|
5
|
+
import { a as KNOWLEDGE_POISONING_KIND_SPEC, c as FAILURE_MODE_KIND_SPEC, d as emitControlIntegrityFindings, f as behavioralAnalyst, i as DEFAULT_TRACE_ANALYST_KINDS, l as CONTROL_INTEGRITY_ANALYST, n as AnalystRegistry, o as KNOWLEDGE_GAP_KIND_SPEC, p as deriveEfficiencyFindings, r as ExactAnalystRunExecutionError, s as IMPROVEMENT_KIND_SPEC, t as buildDefaultAnalystRegistry, u as ControlIntegrityAnalyst } from "../default-registry-6QOiTzqi.js";
|
|
6
6
|
import { i as makeProposalFinding, n as computeFindingId, r as makeFinding } from "../types-CW5b90JH.js";
|
|
7
7
|
import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-US8YsYGq.js";
|
|
8
8
|
import { a as diffFindings, i as defaultIsMaterial, n as runSemanticConceptJudge, o as createChatTraceEngine, r as FindingsStore, t as SEMANTIC_CONCEPT_JUDGE_VERSION } from "../semantic-concept-judge-BcjYg37z.js";
|
|
@@ -1,8 +1,8 @@
|
|
|
1
1
|
import { c as RunSplitTag } from "../run-record-DF-fuu31.js";
|
|
2
|
-
import { a as CampaignResult } from "../types-
|
|
3
|
-
import "../index-
|
|
4
|
-
import { Yt as CampaignStorage } from "../search-kernel-
|
|
5
|
-
import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-
|
|
2
|
+
import { a as CampaignResult } from "../types-BR1LLIxm.js";
|
|
3
|
+
import "../index-BPCdNSMW.js";
|
|
4
|
+
import { Yt as CampaignStorage } from "../search-kernel-zI-rsHoG.js";
|
|
5
|
+
import { a as BenchmarkFamily, c as BenchmarkSource, i as BenchmarkEvaluation, l as BenchmarkTaskKind, n as BenchmarkAdapter, o as BenchmarkResponder, r as BenchmarkDatasetItem, s as BenchmarkScenario, t as BENCHMARK_SPLIT_SEED, u as deterministicSplit } from "../types-BbiT5NPH.js";
|
|
6
6
|
//#region src/benchmarks/calibration.d.ts
|
|
7
7
|
interface BenchmarkMetricCalibrationOptions<TPayload = unknown, TArtifact = string> {
|
|
8
8
|
adapter: BenchmarkAdapter<BenchmarkDatasetItem<TPayload>, TPayload, TArtifact>;
|
package/dist/benchmarks/index.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { t as __exportAll } from "../rolldown-runtime-8H4AJuhK.js";
|
|
2
2
|
import { rt as runCampaign } from "../llm-judge-C-0P1WiT.js";
|
|
3
3
|
import { S as fsCampaignStorage } from "../external-optimizer-subprocess-CkyUUV7h.js";
|
|
4
|
-
import "../campaign-
|
|
4
|
+
import "../campaign-Btk1yFfw.js";
|
|
5
5
|
import { join, resolve } from "node:path";
|
|
6
6
|
import { readFile, readdir, stat } from "node:fs/promises";
|
|
7
7
|
import { parse } from "smol-toml";
|
package/dist/campaign/index.d.ts
CHANGED
|
@@ -1,10 +1,10 @@
|
|
|
1
1
|
import { c as CostLedgerHandle, w as PendingCostCallView } from "../cost-ledger-DbQdN3nO.js";
|
|
2
2
|
import { _ as ProposalFinding, v as ProposalFindingOrigin, x as makeProposalFinding } from "../types-CeDfCRXp.js";
|
|
3
3
|
import { t as SeriesDistribution } from "../descriptive-B2iPaT9J.js";
|
|
4
|
-
import { $ as paretoSignificanceGate, B as PowerPreflight, G as BuildEvidenceVectorOptions, H as powerPreflight, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, V as PowerPreflightOptions, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, _ as sequentialPairedGate, a as detectScale, c as pairHoldout, d as SequentialDecision, f as SequentialObservation, g as sequentialDecide, h as SequentialStreamState, i as PairedHoldout, l as SequentialDecideFn, m as SequentialPairedGateOptions, n as HeldoutSignificance, o as dimensionRegressions, p as SequentialPairedGate, q as ObjectiveSource, r as HeldoutSignificanceOptions, s as heldoutSignificance, t as DimensionRegression, u as SequentialDecideOptions } from "../statistical-heldout-
|
|
4
|
+
import { $ as paretoSignificanceGate, B as PowerPreflight, G as BuildEvidenceVectorOptions, H as powerPreflight, J as ParetoSignificanceGateOptions, K as EvidenceVector, Q as paretoPolicy, U as AxisEvidence, V as PowerPreflightOptions, W as AxisVerdict, X as PromotionPolicy, Y as PromotionObjective, Z as buildEvidenceVector, _ as sequentialPairedGate, a as detectScale, c as pairHoldout, d as SequentialDecision, f as SequentialObservation, g as sequentialDecide, h as SequentialStreamState, i as PairedHoldout, l as SequentialDecideFn, m as SequentialPairedGateOptions, n as HeldoutSignificance, o as dimensionRegressions, p as SequentialPairedGate, q as ObjectiveSource, r as HeldoutSignificanceOptions, s as heldoutSignificance, t as DimensionRegression, u as SequentialDecideOptions } from "../statistical-heldout-BAbd_oXC.js";
|
|
5
5
|
import { $ as SearchTokenAccounting, A as SearchLedgerHash, B as SearchOperationStartedEvent, C as SearchEdgeOperator, D as SearchFailureReason, E as SearchExecutionIdentity, F as SearchNodeRegisteredEvent, G as SearchSplit, H as SearchProposerKind, I as SearchNodeStatus, J as SearchSurfaceEffect, K as SearchSplitTasks, L as SearchOpenedEvent, M as SearchNodeDecidedEvent, N as SearchNodeDecision, O as SearchLedgerEntry, P as SearchNodeRef, Q as SearchTaskOutcome, R as SearchOperationKind, S as SearchEdgeAttribution, T as SearchEstimateMethod, U as SearchReservation, V as SearchProposer, W as SearchSourceRef, X as SearchSurfaceKind, Y as SearchSurfaceEvidence, Z as SearchTask, _ as SearchClaimPower, a as SearchAttemptAccounting, b as SearchClosedEvent, c as SearchCancelReason, d as SearchCellCancelledEvent, et as SearchTraceRef, f as SearchCellSettledEvent, g as SearchClaimFinalist, h as SearchClaimEstimator, i as SearchArtifactRef, j as SearchModelIdentity, k as SearchLedgerEvent, l as SearchCandidateSurface, m as SearchClaim, n as SEARCH_LEDGER_SCHEMA, o as SearchAudit, p as SearchCellStage, q as SearchSplits, r as SearchArtifactKind, s as SearchBudget, t as NodeEstimate, tt as SearchUnknown, u as SearchCellAllocatedEvent, v as SearchClaimTest, w as SearchEdgeRecordedEvent, x as SearchCostAccounting, y as SearchCloseReason, z as SearchOperationRecordedEvent } from "../search-ledger-types-p2xKfiDJ.js";
|
|
6
|
-
import { $ as SearchNode, A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as labelTrustRank, H as SurfaceProposer, I as ProposedCandidate, J as renderSearchSummary, K as SearchProposerView, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, Q as SearchDecisionRecord, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchCell, Y as searchProposerView, Z as SearchCompletion, _ as GateDecision, a as CampaignResult, at as SearchUnitScore, b as GenerationRecord, c as CampaignTraceWriter, ct as searchTaskSetDigest, d as DispatchContext, et as SearchOperation, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchStateView, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as searchUnitScores, m as GateCheckStatus, n as CampaignArtifactWriter, nt as SearchSpend, o as CampaignScenarioIdentity, ot as searchCellId, p as Gate, q as SearchSummaryOptions, r as CampaignCellResult, rt as SearchState, s as CampaignTokenUsage, st as searchNodeId, t as CampaignAggregates, tt as SearchScoredCell, u as ComponentSurface, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-
|
|
7
|
-
import { $ as ScoreboardRow, $i as RunImprovementLoopOptions, $n as TraceAnalystScenario, $r as GepaOptimizationRecipe, $t as importExternalEvaluations, A as ScenarioSignal, Aa as
|
|
8
|
-
import { A as SearchCellPlan, At as SearchHistoryCoverageRow, B as aide, Bt as FileSearchLedger, C as decideSearchClaim, Ct as searchEdgeId, D as AshaOptions, Dt as SearchHistoryAdmissionOptions, E as verifySearchClaim, Et as CreateSearchHistoryReceiptInput, F as AideOptions, Ft as assertSearchHistoryMatchesState, G as EstimateNodeCellsInput, Gt as openSearchLedger, H as crowdedFrontierParent, Ht as SearchLedger, I as SearchExpansion, It as createSearchHistoryReceipt, J as SearchPosterior, Jt as validateSearchLedgerEvent, K as NodePosterior, Kt as parseSearchLedgerLine, L as SearchPolicy, Lt as searchHistoryCoverageRow, M as SearchUnitExtension, Mt as SearchHistoryReceipt, N as asha, Nt as SearchHistoryRequiredError, O as SearchAllocationView, Ot as SearchHistoryAuditSummary, P as uniform, Pt as assertCompleteSearchHistory, Q as searchPosterior, Qt as inMemoryCampaignStorage, R as SearchPolicyNode, Rt as verifySearchHistoryArtifact, S as SearchClaimVerification, St as developmentClaim, T as searchClaimReserveUsd, Tt as surfaceNode, U as incumbent, Ut as SearchLedgerAppendResult, V as beam, Vt as OpenSearchLedgerOptions, W as incumbentWithOperatorBandit, Wt as SearchLedgerTrustedHeadMode, X as estimateNodeFromCells, Xt as createRunCostLedger, Y as estimateNode, Yt as CampaignStorage, Z as searchCellSetDigest, Zt as fsCampaignStorage, _ as searchPolicyView, _t as SearchOpening, a as SearchCellWork, b as SearchClaimDecision, bt as SearchRunIdentity, c as SearchProposalBlob, d as SearchProposedChild, f as SearchProposerPort, g as searchExpansionIndex, gt as SearchLedgerBinding, h as searchDivergence, ht as RegisterSearchNodeInput, i as SearchCellResult, j as SearchRungDecision, jt as SearchHistoryPolicy, k as SearchAllocator, kt as SearchHistoryCoverage, l as SearchProposalRequest, m as runSearch, mt as RecordSearchEdgeInput, n as SEARCH_KERNEL_SOURCE, o as SearchExecutor, p as SearchRunResult, pt as AllocateSearchCellInput, q as SEARCH_ESTIMATOR, qt as replaySearchLedgerText, r as SearchArtifactCodec, s as SearchLane, t as RunSearchOptions, u as SearchProposalResult, v as PlanSearchClaimInput, vt as SearchRecorder, w as planSearchClaim, wt as surfaceDiff, x as SearchClaimPlan, xt as SettleSearchCellInput, y as SEARCH_CLAIM_RULE, yt as SearchRecorderOptions, z as SearchPolicyView, zt as verifySearchHistoryReceipt } from "../search-kernel-
|
|
6
|
+
import { $ as SearchNode, A as LabeledScenarioWrite, B as ScoredSurfaceOutcome, C as JudgeDimension, D as LabeledScenarioSampleArgs, E as LabeledScenarioRecord, F as ProposeContext, G as labelTrustRank, H as SurfaceProposer, I as ProposedCandidate, J as renderSearchSummary, K as SearchProposerView, L as RedactionStatus, M as OptimizerConfig, N as ParetoParent, O as LabeledScenarioSource, P as ProposalTrackContext, Q as SearchDecisionRecord, R as Scenario, S as JudgeConfig, T as LabelTrust, U as TraceSpan, V as SessionScript, W as isProposedCandidate, X as SearchCell, Y as searchProposerView, Z as SearchCompletion, _ as GateDecision, a as CampaignResult, at as SearchUnitScore, b as GenerationRecord, c as CampaignTraceWriter, ct as searchTaskSetDigest, d as DispatchContext, et as SearchOperation, f as DispatchFn, g as GateContribution, h as GateContext, i as CampaignCostMeter, it as SearchStateView, j as MutableSurface, k as LabeledScenarioStore, l as CodeSurface, lt as searchUnitScores, m as GateCheckStatus, n as CampaignArtifactWriter, nt as SearchSpend, o as CampaignScenarioIdentity, ot as searchCellId, p as Gate, q as SearchSummaryOptions, r as CampaignCellResult, rt as SearchState, s as CampaignTokenUsage, st as searchNodeId, t as CampaignAggregates, tt as SearchScoredCell, u as ComponentSurface, v as GateResult, w as JudgeScore, x as JudgeAggregate, y as GenerationCandidate, z as ScenarioAggregate } from "../types-BR1LLIxm.js";
|
|
7
|
+
import { $ as ScoreboardRow, $i as RunImprovementLoopOptions, $n as TraceAnalystScenario, $r as GepaOptimizationRecipe, $t as importExternalEvaluations, A as ScenarioSignal, Aa as CampaignCoverage, Ai as GepaCandidatePopulationSummary, An as CrossSurfaceComponentEvidence, B as FinalizedProfileMatrixResult, Bn as CrossSurfaceInteractionTask, Bt as FsLabeledScenarioStoreOptions, C as SingleRunLockOptions, Ca as CacheRead, Ci as compareOptimizationMethods, Cn as CrossSurfaceBootstrapPolicy, Cr as campaignMeanComposite, D as searchProposalExecution, Da as cellCachePath, Di as costFromLedgerSummary, Dn as CrossSurfaceCandidateOutcome, Dr as DefaultProductionGateOptions, E as searchModelIdentity, Ea as buildCellSchedule, Ei as combineComparisonCosts, En as CrossSurfaceCandidateEvidence, Er as DefaultProductionGateCheck, F as RetainedDispatchOptions, Fa as campaignSplitDigest, Fi as ExternalOptimizerObservationSummary, Fn as CrossSurfaceIneligibilityReason, G as RunProfileMatrixSegmentOptions, Gn as CrossSurfacePairwiseEntry, Gr as SkillOptOptimizationMethodConfig, Gt as ScoredRollout, H as ProfileMatrixPlan, Hn as CrossSurfacePairCompatibility, Ht as RolloutArgumentDiff, I as RetainedDispatchOutcome, Ia as campaignSplitDigestFromIdentities, Ii as ExternalOptimizerSubmittedCandidate, In as CrossSurfaceInteractionAwareSelection, It as scopedOptimizationMethod, J as runProfileMatrixSegment, Ji as llmJudge, Jn as CrossSurfaceSelectionPolicy, Jr as skillOptOptimizationMethod, Jt as rolloutArgumentDiff, K as createProfileMatrixPlan, Ki as LlmJudgeDimension, Kn as CrossSurfaceRankedSingle, Kr as SkillOptRunnerCommand, Kt as UngroundedLiteralReport, L as createRetainedDispatch, Li as readExternalOptimizerObservationArtifact, Ln as CrossSurfaceInteractionEffect, Lt as sequentialOptimizationMethod, M as selectDiscriminative, Ma as assertCampaignSplitIdentity, Mi as readGepaCandidatePopulationArtifact, Mn as CrossSurfaceDistribution, N as resolveRunDir, Na as campaignCoverage, Ni as ExternalOptimizerExecutionSummary, Nn as CrossSurfaceEligibility, O as searchReceiptAccounting, Oa as FinalEvidencePolicy, Oi as GepaCandidatePopulationArtifact, On as CrossSurfaceCandidateSummary, Or as DefaultProductionRewardHackingOptions, P as tangleTracesRoot, Pa as campaignScenarioIdentity, Pi as ExternalOptimizerObservationArtifact, Pn as CrossSurfaceEvidenceBreakdown, Q as ScoreboardRenderOptions, Qi as transientDispatchFailure, Qn as TraceAnalystArtifact, Qr as GepaOptimizationMethodConfig, Qt as externalSurface, R as CreateProfileMatrixPlanOptions, Rn as CrossSurfaceInteractionPath, Rt as neutralizeText, S as SingleRunLock, Sa as CacheIssueReason, Si as OptimizationTokenUsage, Sn as CrossSurfaceBestSingleSelection, Sr as campaignBreakdown, T as campaignCellSearchResult, Ta as CellScheduleSlot, Ti as ComparisonCost, Tn as CrossSurfaceCandidateComparison, Tr as heldOutGate, U as ProfileMatrixRow, Ui as ReferenceEquivalenceScenario, Un as CrossSurfacePairEvidence, Ut as RolloutArgumentDiffOptions, V as ProfileMatrixCoverage, Vi as ReferenceEquivalenceJudgeOptions, Vn as CrossSurfaceNaiveStackSelection, Vt as LabeledScenarioStoreError, W as ProfileMatrixSegmentResult, Wi as createReferenceEquivalenceJudge, Wn as CrossSurfacePairIncompatibilityReason, Wr as composeGate, Wt as RolloutCall, X as PlaybackDriver, Xi as isTransientTransportFailure, Xn as CrossSurfaceTaskRow, Xr as GepaEngineOptions, Xt as GepaEvaluationImport, Y as PlaybackContext, Yi as TransientFailureOptions, Yn as CrossSurfaceSelections, Yr as GepaAdaptiveEngineRun, Yt as ExternalSearchOptimizer, Z as PlaybackStep, Zi as quotaExhaustedUntil, Zn as BuildTraceAnalystSurfaceDispatchOptions, Zr as GepaEngineRun, Zt as GepaPopulationImport, _ as componentSurfaceIdentityMaterial, _a as runCampaign, _i as OptimizationMethodProvenance, _n as analyzeCrossSurfaceInteractions, _r as provenanceRecordPath, a as WorktreeAdapterError, aa as runOptimization, ai as ExternalTextOptimizationMethodConfig, an as EvalFixture, ar as LoopProvenanceArgsFromResult, at as scoreUserStory, b as surfaceDispatchRef, ba as PlanCampaignRunOptions, bi as OptimizationMethodScore, bn as CrossSurfaceAdditionRejectionReason, br as CampaignBreakdown, c as verifyCodeSurface, ca as SearchLedgerIntegrityError, ci as ExternalOptimizationExample, cn as EvalFixtureRunPlan, cr as LoopProvenanceEvidence, ct as ProfileDispatchFn, d as PhoenixEvaluationResultLike, da as openAutoPr, di as CompareOptimizationMethodsOptions, dn as LoadEvalFixtureScenariosOptions, dr as buildLoopProvenanceRecord, dt as RunProfileMatrixOptions, ea as RunImprovementLoopResult, ei as GepaRunnerCommand, en as importGepaPopulation, er as buildTraceAnalystSurfaceDispatch, et as ScoreboardSummary, f as PhoenixEvaluatorLike, fa as RunEvalOptions, fi as OptimizationMethod, fn as PlanEvalFixtureRunOptions, fr as campaignMeasurementDigest, ft as RunProfileMatrixResult, g as codeSurfaceIdentityMaterial, ga as RunCampaignOptions, gi as OptimizationMethodPairwise, gn as planEvalFixtureRun, gr as loopProvenanceSpans, h as assertCodeSurfaceIdentity, ha as CampaignCellRetryPolicy, hi as OptimizationMethodInput, hn as loadEvalFixtureScenarios, hr as loopProvenanceArgsFromResult, i as WorktreeAdapter, ia as RunOptimizationResult, ii as externalTextOptimizationMethod, in as neutralizationGate, ir as EmitLoopProvenanceResult, it as renderScoreboardMarkdown, j as scoreDiscrimination, ja as assertCampaignDesign, ji as GepaCandidateSelectionScore, jn as CrossSurfaceCompositionStep, k as DiscriminationScore, ka as FinalEvidenceUse, ki as GepaCandidatePopulationCandidate, kn as CrossSurfaceComponent, kr as defaultProductionGate, l as AutoevalsScoreLike, la as OpenAutoPrOptions, li as ExternalTextEvaluationResponse, ln as EvalFixtureScenario, lr as LoopProvenanceOptimizationMethod, lt as ProfileMatrixError, m as phoenixEvaluatorJudge, ma as CampaignCellFailureReceipt, mi as OptimizationMethodComposition, mn as loadEvalFixture, mr as emitLoopProvenance, mt as runProfileMatrix, n as GitWorktreeAdapterOptions, na as PremeasuredOptimizationBaseline, ni as OpenAICompatibleOptimizerModel, nn as recordSkillOptSearch, nr as BuildLoopProvenanceArgs, nt as UserStoryVerdict, o as gitWorktreeAdapter, oa as SearchLedgerConflictError, oi as ExternalTextOptimizerContext, on as EvalFixtureFile, or as LoopProvenanceBackend, ot as scoreboardSummary, p as autoevalsScorerJudge, pa as runEval, pi as OptimizationMethodComparison, pn as discoverEvalFixtures, pr as canonicalDigest, pt as ScenarioRollup, q as finalizeProfileMatrix, qi as LlmJudgeOptions, qn as CrossSurfaceRelativeCost, qr as SkillOptTrainerConfig, qt as classifyUngroundedLiterals, r as Worktree, ra as RunOptimizationOptions, ri as OptimizerModelBudget, rn as NeutralizationGateOptions, rr as EmitLoopProvenanceArgs, rt as makePlaybackDispatch, s as resolveWorktreePath, sa as SearchLedgerError, si as ExternalTextOptimizerResult, sn as EvalFixtureLoadOptions, sr as LoopProvenanceCandidate, st as userStoryScoreboard, t as CodeSurfaceVerification, ta as runImprovementLoop, ti as gepaOptimizationMethod, tn as recordGepaSearch, tr as traceAnalystQualityJudge, tt as UserStory, u as AutoevalsScorerLike, ua as OpenAutoPrResult, ui as decodeExternalTextCandidate, un as EvalFixtureValidationMode, ur as LoopProvenanceRecord, ut as ProfileSummary, v as renderSurfaceDiff, va as CampaignRunPlan, vi as OptimizationMethodResult, vn as AnalyzeCrossSurfaceInteractionsInput, vr as provenanceSpansPath, w as acquireSingleRunLock, wa as readCachedCell, wi as optimizationTokenUsageFromSummary, wn as CrossSurfaceCandidate, wr as HeldOutGateOptions, x as surfaceHash, xa as planCampaignRun, xi as OptimizationPackageSource, xn as CrossSurfaceAttemptCompleteness, y as surfaceContentHash, ya as CampaignRunPlanCell, yi as OptimizationMethodRunOptions, yn as CrossSurfaceAdditionDecision, yr as verifyLoopProvenanceRecord, z as FinalizeProfileMatrixOptions, zn as CrossSurfaceInteractionReport, zt as FsLabeledScenarioStore } from "../index-BPCdNSMW.js";
|
|
8
|
+
import { A as SearchCellPlan, At as SearchHistoryCoverageRow, B as aide, Bt as FileSearchLedger, C as decideSearchClaim, Ct as searchEdgeId, D as AshaOptions, Dt as SearchHistoryAdmissionOptions, E as verifySearchClaim, Et as CreateSearchHistoryReceiptInput, F as AideOptions, Ft as assertSearchHistoryMatchesState, G as EstimateNodeCellsInput, Gt as openSearchLedger, H as crowdedFrontierParent, Ht as SearchLedger, I as SearchExpansion, It as createSearchHistoryReceipt, J as SearchPosterior, Jt as validateSearchLedgerEvent, K as NodePosterior, Kt as parseSearchLedgerLine, L as SearchPolicy, Lt as searchHistoryCoverageRow, M as SearchUnitExtension, Mt as SearchHistoryReceipt, N as asha, Nt as SearchHistoryRequiredError, O as SearchAllocationView, Ot as SearchHistoryAuditSummary, P as uniform, Pt as assertCompleteSearchHistory, Q as searchPosterior, Qt as inMemoryCampaignStorage, R as SearchPolicyNode, Rt as verifySearchHistoryArtifact, S as SearchClaimVerification, St as developmentClaim, T as searchClaimReserveUsd, Tt as surfaceNode, U as incumbent, Ut as SearchLedgerAppendResult, V as beam, Vt as OpenSearchLedgerOptions, W as incumbentWithOperatorBandit, Wt as SearchLedgerTrustedHeadMode, X as estimateNodeFromCells, Xt as createRunCostLedger, Y as estimateNode, Yt as CampaignStorage, Z as searchCellSetDigest, Zt as fsCampaignStorage, _ as searchPolicyView, _t as SearchOpening, a as SearchCellWork, b as SearchClaimDecision, bt as SearchRunIdentity, c as SearchProposalBlob, d as SearchProposedChild, f as SearchProposerPort, g as searchExpansionIndex, gt as SearchLedgerBinding, h as searchDivergence, ht as RegisterSearchNodeInput, i as SearchCellResult, j as SearchRungDecision, jt as SearchHistoryPolicy, k as SearchAllocator, kt as SearchHistoryCoverage, l as SearchProposalRequest, m as runSearch, mt as RecordSearchEdgeInput, n as SEARCH_KERNEL_SOURCE, o as SearchExecutor, p as SearchRunResult, pt as AllocateSearchCellInput, q as SEARCH_ESTIMATOR, qt as replaySearchLedgerText, r as SearchArtifactCodec, s as SearchLane, t as RunSearchOptions, u as SearchProposalResult, v as PlanSearchClaimInput, vt as SearchRecorder, w as planSearchClaim, wt as surfaceDiff, x as SearchClaimPlan, xt as SettleSearchCellInput, y as SEARCH_CLAIM_RULE, yt as SearchRecorderOptions, z as SearchPolicyView, zt as verifySearchHistoryReceipt } from "../search-kernel-zI-rsHoG.js";
|
|
9
9
|
import { B as ExternalOptimizerModelBudget, F as ExternalOptimizerCallbackLimits, G as ExternalOptimizerProcessLimits, H as ExternalOptimizerModelCallRequest, I as ExternalOptimizerChatRequest, L as ExternalOptimizerEndpointFormat, N as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, P as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, Q as resolveExternalOptimizerProcessLimits, R as ExternalOptimizerEvaluationObservation, U as ExternalOptimizerModelCallResult, V as ExternalOptimizerModelCall, W as ExternalOptimizerModelExecutionObservation, Y as ExternalTextCandidate, Z as resolveExternalOptimizerCallbackLimits, q as ExternalOptimizerRunnerCommand, z as ExternalOptimizerEvaluationRefusalReason } from "../agent-profile-DDfIjdyS.js";
|
|
10
|
-
export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
10
|
+
export { type AideOptions, type AllocateSearchCellInput, type AnalyzeCrossSurfaceInteractionsInput, type AshaOptions, type AutoevalsScoreLike, type AutoevalsScorerLike, type AxisEvidence, type AxisVerdict, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type BuildTraceAnalystSurfaceDispatchOptions, type CacheIssueReason, type CacheRead, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellFailureReceipt, type CampaignCellResult, type CampaignCellRetryPolicy, type CampaignCostMeter, type CampaignCoverage, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CellScheduleSlot, type CodeSurface, type CodeSurfaceVerification, type CompareOptimizationMethodsOptions, type ComparisonCost, type ComponentSurface, type CostLedgerHandle, type CreateProfileMatrixPlanOptions, type CreateSearchHistoryReceiptInput, type CrossSurfaceAdditionDecision, type CrossSurfaceAdditionRejectionReason, type CrossSurfaceAttemptCompleteness, type CrossSurfaceBestSingleSelection, type CrossSurfaceBootstrapPolicy, type CrossSurfaceCandidate, type CrossSurfaceCandidateComparison, type CrossSurfaceCandidateEvidence, type CrossSurfaceCandidateOutcome, type CrossSurfaceCandidateSummary, type CrossSurfaceComponent, type CrossSurfaceComponentEvidence, type CrossSurfaceCompositionStep, type CrossSurfaceDistribution, type CrossSurfaceEligibility, type CrossSurfaceEvidenceBreakdown, type CrossSurfaceIneligibilityReason, type CrossSurfaceInteractionAwareSelection, type CrossSurfaceInteractionEffect, type CrossSurfaceInteractionPath, type CrossSurfaceInteractionReport, type CrossSurfaceInteractionTask, type CrossSurfaceNaiveStackSelection, type CrossSurfacePairCompatibility, type CrossSurfacePairEvidence, type CrossSurfacePairIncompatibilityReason, type CrossSurfacePairwiseEntry, type CrossSurfaceRankedSingle, type CrossSurfaceRelativeCost, type CrossSurfaceSelectionPolicy, type CrossSurfaceSelections, type CrossSurfaceTaskRow, DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, type DefaultProductionGateCheck, type DefaultProductionGateOptions, type DefaultProductionRewardHackingOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EstimateNodeCellsInput, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type ExternalOptimizationExample, type ExternalOptimizerCallbackLimits, type ExternalOptimizerChatRequest, type ExternalOptimizerEndpointFormat, type ExternalOptimizerEvaluationObservation, type ExternalOptimizerEvaluationRefusalReason, type ExternalOptimizerExecutionSummary, type ExternalOptimizerModelBudget, type ExternalOptimizerModelCall, type ExternalOptimizerModelCallRequest, type ExternalOptimizerModelCallResult, type ExternalOptimizerModelExecutionObservation, type ExternalOptimizerObservationArtifact, type ExternalOptimizerObservationSummary, type ExternalOptimizerProcessLimits, type ExternalOptimizerRunnerCommand, type ExternalOptimizerSubmittedCandidate, type ExternalSearchOptimizer, type ExternalTextCandidate, type ExternalTextEvaluationResponse, type ExternalTextOptimizationMethodConfig, type ExternalTextOptimizerContext, type ExternalTextOptimizerResult, FileSearchLedger, type FinalEvidencePolicy, type FinalEvidenceUse, type FinalizeProfileMatrixOptions, type FinalizedProfileMatrixResult, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateCheckStatus, type GateContext, type GateContribution, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaAdaptiveEngineRun, type GepaCandidatePopulationArtifact, type GepaCandidatePopulationCandidate, type GepaCandidatePopulationSummary, type GepaCandidateSelectionScore, type GepaEngineOptions, type GepaEngineRun, type GepaEvaluationImport, type GepaOptimizationMethodConfig, type GepaOptimizationRecipe, type GepaPopulationImport, type GepaRunnerCommand, type GitWorktreeAdapterOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, type LlmJudgeDimension, type LlmJudgeOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceOptimizationMethod, type LoopProvenanceRecord, type MutableSurface, type NeutralizationGateOptions, type NodeEstimate, type NodePosterior, type ObjectiveSource, type OpenAICompatibleOptimizerModel, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationMethod, type OptimizationMethodComparison, type OptimizationMethodComposition, type OptimizationMethodInput, type OptimizationMethodPairwise, type OptimizationMethodProvenance, type OptimizationMethodResult, type OptimizationMethodRunOptions, type OptimizationMethodScore, type OptimizationPackageSource, type OptimizationTokenUsage, type OptimizerConfig, type OptimizerModelBudget, type PairedHoldout, type ParetoParent, type ParetoSignificanceGateOptions, type PendingCostCallView, type PhoenixEvaluationResultLike, type PhoenixEvaluatorLike, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlanSearchClaimInput, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PowerPreflight, type PowerPreflightOptions, type PremeasuredOptimizationBaseline, type ProfileDispatchFn, type ProfileMatrixCoverage, ProfileMatrixError, type ProfileMatrixPlan, type ProfileMatrixRow, type ProfileMatrixSegmentResult, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposalFinding, type ProposalFindingOrigin, type ProposalTrackContext, type ProposeContext, type ProposedCandidate, type RecordSearchEdgeInput, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RegisterSearchNodeInput, type RetainedDispatchOptions, type RetainedDispatchOutcome, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunProfileMatrixSegmentOptions, type RunSearchOptions, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAllocationView, type SearchAllocator, type SearchArtifactCodec, type SearchArtifactKind, type SearchArtifactRef, type SearchAttemptAccounting, type SearchAudit, type SearchBudget, type SearchCancelReason, type SearchCandidateSurface, type SearchCell, type SearchCellAllocatedEvent, type SearchCellCancelledEvent, type SearchCellPlan, type SearchCellResult, type SearchCellSettledEvent, type SearchCellStage, type SearchCellWork, type SearchClaim, type SearchClaimDecision, type SearchClaimEstimator, type SearchClaimFinalist, type SearchClaimPlan, type SearchClaimPower, type SearchClaimTest, type SearchClaimVerification, type SearchCloseReason, type SearchClosedEvent, type SearchCompletion, type SearchCostAccounting, type SearchDecisionRecord, type SearchEdgeAttribution, type SearchEdgeOperator, type SearchEdgeRecordedEvent, type SearchEstimateMethod, type SearchExecutionIdentity, type SearchExecutor, type SearchExpansion, type SearchFailureReason, type SearchHistoryAdmissionOptions, type SearchHistoryAuditSummary, type SearchHistoryCoverage, type SearchHistoryCoverageRow, type SearchHistoryPolicy, type SearchHistoryReceipt, SearchHistoryRequiredError, type SearchLane, type SearchLedger, type SearchLedgerAppendResult, type SearchLedgerBinding, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerTrustedHeadMode, type SearchModelIdentity, type SearchNode, type SearchNodeDecidedEvent, type SearchNodeDecision, type SearchNodeRef, type SearchNodeRegisteredEvent, type SearchNodeStatus, type SearchOpenedEvent, type SearchOpening, type SearchOperation, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchOperationStartedEvent, type SearchPolicy, type SearchPolicyNode, type SearchPolicyView, type SearchPosterior, type SearchProposalBlob, type SearchProposalRequest, type SearchProposalResult, type SearchProposedChild, type SearchProposer, type SearchProposerKind, type SearchProposerPort, type SearchProposerView, SearchRecorder, type SearchRecorderOptions, type SearchReservation, type SearchRunIdentity, type SearchRunResult, type SearchRungDecision, type SearchScoredCell, type SearchSourceRef, type SearchSpend, type SearchSplit, type SearchSplitTasks, type SearchSplits, SearchState, SearchStateView, type SearchSummaryOptions, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTask, type SearchTaskOutcome, type SearchTokenAccounting, type SearchTraceRef, type SearchUnitExtension, type SearchUnitScore, type SearchUnknown, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SequentialStreamState, type SeriesDistribution, type SessionScript, type SettleSearchCellInput, type SingleRunLock, type SingleRunLockOptions, type SkillOptOptimizationMethodConfig, type SkillOptRunnerCommand, type SkillOptTrainerConfig, type SurfaceProposer, type TraceAnalystArtifact, type TraceAnalystScenario, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
package/dist/campaign/index.js
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout, r as detectScale } from "../statistical-heldout-CNLOvRq_.js";
|
|
2
2
|
import { A as assertCompleteSearchHistory, B as searchReceiptAccounting, C as incumbentWithOperatorBandit, D as surfaceDiff, E as searchEdgeId, F as verifySearchHistoryArtifact, H as uniform, I as verifySearchHistoryReceipt, K as defaultProductionGate, L as campaignCellSearchResult, M as assertSearchHistoryMatchesState, N as createSearchHistoryReceipt, O as surfaceNode, P as searchHistoryCoverageRow, R as searchModelIdentity, S as incumbent, T as developmentClaim, V as asha, W as openAutoPr, _ as isProposedCandidate, a as transientDispatchFailure, at as resolveRunDir, b as beam, c as compareOptimizationMethods, ct as buildCellSchedule, d as combineComparisonCosts, f as costFromLedgerSummary, g as runOptimization, h as runImprovementLoop, i as quotaExhaustedUntil, it as planCampaignRun, k as SearchHistoryRequiredError, l as optimizationTokenUsageFromSummary, lt as cellCachePath, m as readGepaCandidatePopulationArtifact, nt as runEval, ot as tangleTracesRoot, r as isTransientTransportFailure, rt as runCampaign, st as readCachedCell, t as llmJudge, v as labelTrustRank, w as SearchRecorder, x as crowdedFrontierParent, y as aide, z as searchProposalExecution } from "../llm-judge-C-0P1WiT.js";
|
|
3
|
-
import { D as powerPreflight, F as surfaceContentHash, H as assertCampaignDesign, I as surfaceDispatchRef, J as campaignSplitDigestFromIdentities, K as campaignScenarioIdentity, L as surfaceHash, M as codeSurfaceIdentityMaterial, N as componentSurfaceIdentityMaterial, P as renderSurfaceDiff, U as assertCampaignSplitIdentity, _ as provenanceRecordPath, b as campaignBreakdown, d as buildLoopProvenanceRecord, f as campaignMeasurementDigest, g as loopProvenanceSpans, h as loopProvenanceArgsFromResult, j as assertCodeSurfaceIdentity, m as emitLoopProvenance, p as canonicalDigest, q as campaignSplitDigest, v as provenanceSpansPath, x as campaignMeanComposite, y as verifyLoopProvenanceRecord } from "../campaign-evidence-Dt_rWPk2.js";
|
|
3
|
+
import { D as powerPreflight, F as surfaceContentHash, G as campaignCoverage, H as assertCampaignDesign, I as surfaceDispatchRef, J as campaignSplitDigestFromIdentities, K as campaignScenarioIdentity, L as surfaceHash, M as codeSurfaceIdentityMaterial, N as componentSurfaceIdentityMaterial, P as renderSurfaceDiff, U as assertCampaignSplitIdentity, _ as provenanceRecordPath, b as campaignBreakdown, d as buildLoopProvenanceRecord, f as campaignMeasurementDigest, g as loopProvenanceSpans, h as loopProvenanceArgsFromResult, j as assertCodeSurfaceIdentity, m as emitLoopProvenance, p as canonicalDigest, q as campaignSplitDigest, v as provenanceSpansPath, x as campaignMeanComposite, y as verifyLoopProvenanceRecord } from "../campaign-evidence-Dt_rWPk2.js";
|
|
4
4
|
import { a as SearchLedgerError, i as SearchLedgerConflictError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DY0AXX3W.js";
|
|
5
5
|
import { C as inMemoryCampaignStorage, S as fsCampaignStorage, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, v as resolveExternalOptimizerCallbackLimits, x as createRunCostLedger, y as resolveExternalOptimizerProcessLimits } from "../external-optimizer-subprocess-CkyUUV7h.js";
|
|
6
6
|
import { S as searchCellSetDigest, a as searchPolicyView, b as estimateNode, c as planSearchClaim, i as searchExpansionIndex, l as searchClaimReserveUsd, n as runSearch, o as SEARCH_CLAIM_RULE, r as searchDivergence, s as decideSearchClaim, t as SEARCH_KERNEL_SOURCE, u as verifySearchClaim, w as searchPosterior, x as estimateNodeFromCells, y as SEARCH_ESTIMATOR } from "../search-kernel-Dpniloks.js";
|
|
@@ -9,7 +9,7 @@ import { t as acquireSingleRunLock } from "../single-run-lock-fVu3MZWH.js";
|
|
|
9
9
|
import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-C1nAHgkB.js";
|
|
10
10
|
import { i as makeProposalFinding } from "../types-CW5b90JH.js";
|
|
11
11
|
import { a as importGepaPopulation, c as heldOutGate, d as decodeExternalTextCandidate, f as readExternalOptimizerObservationArtifact, h as createReferenceEquivalenceJudge, i as importExternalEvaluations, l as composeGate, n as gepaOptimizationMethod, o as recordGepaSearch, r as externalSurface, s as recordSkillOptSearch, t as skillOptOptimizationMethod, u as externalTextOptimizationMethod } from "../skillopt-optimization-method-Bls9t-G_.js";
|
|
12
|
-
import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, M as analyzeCrossSurfaceInteractions, N as buildTraceAnalystSurfaceDispatch, O as discoverEvalFixtures, P as traceAnalystQualityJudge, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreUserStory, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as finalizeProfileMatrix, f as runProfileMatrixSegment, g as renderScoreboardMarkdown, h as makePlaybackDispatch, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, l as createRetainedDispatch, m as runProfileMatrix, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as ProfileMatrixError, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as createProfileMatrixPlan, v as scoreboardSummary, w as LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as userStoryScoreboard } from "../campaign-
|
|
12
|
+
import { A as loadEvalFixtureScenarios, C as FsLabeledScenarioStore, D as neutralizationGate, E as rolloutArgumentDiff, M as analyzeCrossSurfaceInteractions, N as buildTraceAnalystSurfaceDispatch, O as discoverEvalFixtures, P as traceAnalystQualityJudge, S as neutralizeText, T as classifyUngroundedLiterals, _ as scoreUserStory, a as autoevalsScorerJudge, b as scopedOptimizationMethod, c as selectDiscriminative, d as finalizeProfileMatrix, f as runProfileMatrixSegment, g as renderScoreboardMarkdown, h as makePlaybackDispatch, i as verifyCodeSurface, j as planEvalFixtureRun, k as loadEvalFixture, l as createRetainedDispatch, m as runProfileMatrix, n as gitWorktreeAdapter, o as phoenixEvaluatorJudge, p as ProfileMatrixError, r as resolveWorktreePath, s as scoreDiscrimination, t as WorktreeAdapterError, u as createProfileMatrixPlan, v as scoreboardSummary, w as LabeledScenarioStoreError, x as sequentialOptimizationMethod, y as userStoryScoreboard } from "../campaign-Btk1yFfw.js";
|
|
13
13
|
import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-BI1RxzwD.js";
|
|
14
14
|
import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-DWn_np6P.js";
|
|
15
|
-
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
15
|
+
export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, FsLabeledScenarioStore, LabeledScenarioStoreError, ProfileMatrixError, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_KERNEL_SOURCE, SEARCH_LEDGER_SCHEMA, SearchHistoryRequiredError, SearchLedgerConflictError, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, WorktreeAdapterError, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertCompleteSearchHistory, assertSearchHistoryMatchesState, autoevalsScorerJudge, beam, buildCellSchedule, buildEvidenceVector, buildLoopProvenanceRecord, buildTraceAnalystSurfaceDispatch, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, combineComparisonCosts, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createReferenceEquivalenceJudge, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, detectScale, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, labelTrustRank, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, makeProposalFinding, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, optimizationTokenUsageFromSummary, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, phoenixEvaluatorJudge, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, provenanceRecordPath, provenanceSpansPath, quotaExhaustedUntil, readCachedCell, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, recordGepaSearch, recordSkillOptSearch, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveExternalOptimizerCallbackLimits, resolveExternalOptimizerProcessLimits, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchExpansionIndex, searchHistoryCoverageRow, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, selectDiscriminative, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, tangleTracesRoot, traceAnalystQualityJudge, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };
|
|
@@ -2389,9 +2389,18 @@ const resultSchema = z.object({
|
|
|
2389
2389
|
receipt: digest,
|
|
2390
2390
|
value: z.unknown()
|
|
2391
2391
|
}).strict();
|
|
2392
|
+
/** The committed byte length and cumulative digest of retained dispatch history. */
|
|
2393
|
+
const anchorEntrySchema = z.object({
|
|
2394
|
+
kind: z.literal("retained-dispatch-anchor-v1"),
|
|
2395
|
+
scope: digest,
|
|
2396
|
+
bytes: z.number().int().positive(),
|
|
2397
|
+
digest
|
|
2398
|
+
}).strict();
|
|
2392
2399
|
/**
|
|
2393
2400
|
* Persist dispatch intent before an external call and its validated result afterward.
|
|
2394
2401
|
* Replays settled calls; never repeats an uncertain call or refunds its allowance.
|
|
2402
|
+
* Every journal event is durably anchored before it takes effect, and history that
|
|
2403
|
+
* cannot be proven equal to its durable anchor fails closed.
|
|
2395
2404
|
* This wraps a campaign's dispatch; runEval still owns scheduling, judging and reports.
|
|
2396
2405
|
* Uses CampaignStorage's existing compare-and-append, not another database or scheduler.
|
|
2397
2406
|
*/
|
|
@@ -2410,12 +2419,17 @@ function createRetainedDispatch(options) {
|
|
|
2410
2419
|
const calls = /* @__PURE__ */ new Map();
|
|
2411
2420
|
const active = /* @__PURE__ */ new Map();
|
|
2412
2421
|
const indexPath = join(options.runDir, "dispatches.jsonl");
|
|
2422
|
+
const scopePath = join(options.runDir, "dispatch-scope.json");
|
|
2423
|
+
const anchorPath = join(options.runDir, "dispatch-anchor.jsonl");
|
|
2424
|
+
const scopeText = canonicalString(header);
|
|
2425
|
+
const scopeDigest = hashCanonical(header);
|
|
2413
2426
|
const receiptsDir = join(options.runDir, "dispatch-results");
|
|
2414
2427
|
const resultPath = (id) => join(receiptsDir, `${id.slice(7)}.json`);
|
|
2415
2428
|
storage.ensureDir(options.runDir);
|
|
2416
2429
|
storage.ensureDir(receiptsDir);
|
|
2417
2430
|
let retainedText = "";
|
|
2418
2431
|
let revision = 0;
|
|
2432
|
+
let chain = null;
|
|
2419
2433
|
let opened = false;
|
|
2420
2434
|
const fail = (message) => {
|
|
2421
2435
|
throw new ValidationError(`retained dispatch: ${message}`);
|
|
@@ -2433,45 +2447,140 @@ function createRetainedDispatch(options) {
|
|
|
2433
2447
|
existing.receipt = event.receipt;
|
|
2434
2448
|
}
|
|
2435
2449
|
};
|
|
2450
|
+
const readAnchor = () => {
|
|
2451
|
+
const content = storage.read(anchorPath);
|
|
2452
|
+
if (content === void 0 && storage.exists(anchorPath)) fail("cannot read durable anchor");
|
|
2453
|
+
const text = content ?? "";
|
|
2454
|
+
if (text && !text.endsWith("\n")) fail("durable anchor is torn");
|
|
2455
|
+
const entries = [];
|
|
2456
|
+
let previous = 0;
|
|
2457
|
+
const lines = text.split("\n");
|
|
2458
|
+
if (lines.at(-1) === "") lines.pop();
|
|
2459
|
+
if (lines.some((line) => !line)) fail("durable anchor contains an empty record");
|
|
2460
|
+
for (const line of lines) {
|
|
2461
|
+
const entry = anchorEntrySchema.parse(JSON.parse(line));
|
|
2462
|
+
if (entry.scope !== scopeDigest) fail("durable anchor scope changed");
|
|
2463
|
+
if (entry.bytes <= previous) fail("durable anchor is not append-only");
|
|
2464
|
+
previous = entry.bytes;
|
|
2465
|
+
entries.push(entry);
|
|
2466
|
+
}
|
|
2467
|
+
return {
|
|
2468
|
+
text,
|
|
2469
|
+
entries
|
|
2470
|
+
};
|
|
2471
|
+
};
|
|
2436
2472
|
const refresh = () => {
|
|
2473
|
+
if (storage.read(scopePath) !== scopeText) fail("scope anchor is missing or changed");
|
|
2437
2474
|
const stored = storage.read(indexPath);
|
|
2438
2475
|
if (stored === void 0 && storage.exists(indexPath)) fail("cannot read existing history");
|
|
2439
2476
|
const text = stored ?? "";
|
|
2440
2477
|
if (!text.startsWith(retainedText) || text && !text.endsWith("\n")) fail("history was truncated, replaced or torn");
|
|
2441
|
-
|
|
2442
|
-
|
|
2443
|
-
|
|
2478
|
+
const { entries } = readAnchor();
|
|
2479
|
+
const bytes = Buffer.byteLength(text);
|
|
2480
|
+
const applied = Buffer.byteLength(retainedText);
|
|
2481
|
+
if (text && !entries.length) fail("durable anchor is missing");
|
|
2482
|
+
const highWater = entries[entries.length - 1];
|
|
2483
|
+
if (highWater && bytes !== highWater.bytes) {
|
|
2484
|
+
if (bytes < highWater.bytes) fail("history was truncated below its durable anchor");
|
|
2485
|
+
fail("history extends beyond its durable anchor");
|
|
2486
|
+
}
|
|
2487
|
+
let offset = 0;
|
|
2488
|
+
let running = null;
|
|
2489
|
+
let entryIndex = 0;
|
|
2490
|
+
let seenHeader = opened;
|
|
2491
|
+
for (const line of text.split("\n")) {
|
|
2492
|
+
if (!line) continue;
|
|
2493
|
+
running = hashCanonical([running, line]);
|
|
2494
|
+
offset += Buffer.byteLength(line) + 1;
|
|
2495
|
+
while (entryIndex < entries.length && entries[entryIndex].bytes === offset) {
|
|
2496
|
+
if (entries[entryIndex].digest !== running) fail("history diverges from its durable anchor");
|
|
2497
|
+
entryIndex += 1;
|
|
2498
|
+
}
|
|
2499
|
+
if (offset <= applied) continue;
|
|
2500
|
+
if (!seenHeader) {
|
|
2501
|
+
const value = JSON.parse(line);
|
|
2444
2502
|
if (hashCanonical(headerSchema.parse(value)) !== hashCanonical(header)) fail("resume changed its scope or allowances");
|
|
2445
|
-
|
|
2446
|
-
} else apply(eventSchema.parse(
|
|
2503
|
+
seenHeader = true;
|
|
2504
|
+
} else apply(eventSchema.parse(JSON.parse(line)));
|
|
2447
2505
|
}
|
|
2506
|
+
if (entries.length && entryIndex !== entries.length) fail("history diverges from its durable anchor");
|
|
2507
|
+
chain = running;
|
|
2448
2508
|
retainedText = text;
|
|
2449
|
-
revision =
|
|
2509
|
+
revision = bytes;
|
|
2510
|
+
opened = seenHeader;
|
|
2450
2511
|
};
|
|
2451
2512
|
const append = (event) => {
|
|
2452
|
-
const
|
|
2513
|
+
const encoded = canonicalString(event);
|
|
2514
|
+
const line = `${encoded}\n`;
|
|
2453
2515
|
const next = storage.append(indexPath, line, revision);
|
|
2454
2516
|
if (next === void 0) {
|
|
2455
2517
|
refresh();
|
|
2456
2518
|
return false;
|
|
2457
2519
|
}
|
|
2458
2520
|
if (next !== revision + Buffer.byteLength(line)) fail("storage returned an invalid append revision");
|
|
2521
|
+
chain = hashCanonical([chain, encoded]);
|
|
2459
2522
|
retainedText += line;
|
|
2460
2523
|
revision = next;
|
|
2461
2524
|
if (event.kind === "retained-dispatch-v1") opened = true;
|
|
2462
2525
|
else apply(event);
|
|
2463
2526
|
return true;
|
|
2464
2527
|
};
|
|
2528
|
+
const advanceAnchor = () => {
|
|
2529
|
+
const currentChain = chain ?? fail("cannot anchor an empty history");
|
|
2530
|
+
if (revision <= 0) fail("cannot anchor an empty history");
|
|
2531
|
+
const { text, entries } = readAnchor();
|
|
2532
|
+
const last = entries[entries.length - 1];
|
|
2533
|
+
if (last) {
|
|
2534
|
+
if (last.bytes > revision) fail("history is behind its durable anchor");
|
|
2535
|
+
if (last.bytes === revision) {
|
|
2536
|
+
if (last.digest !== currentChain) fail("history diverges from its durable anchor");
|
|
2537
|
+
return;
|
|
2538
|
+
}
|
|
2539
|
+
}
|
|
2540
|
+
const line = `${canonicalString({
|
|
2541
|
+
kind: "retained-dispatch-anchor-v1",
|
|
2542
|
+
scope: scopeDigest,
|
|
2543
|
+
bytes: revision,
|
|
2544
|
+
digest: currentChain
|
|
2545
|
+
})}\n`;
|
|
2546
|
+
const next = storage.append(anchorPath, line, Buffer.byteLength(text));
|
|
2547
|
+
if (next === void 0) {
|
|
2548
|
+
const concurrent = readAnchor().entries.at(-1);
|
|
2549
|
+
if (concurrent?.bytes === revision && concurrent.digest === chain) return;
|
|
2550
|
+
fail("durable anchor append is busy; retry with the same run directory");
|
|
2551
|
+
}
|
|
2552
|
+
if (next !== Buffer.byteLength(text) + Buffer.byteLength(line)) fail("storage returned an invalid durable anchor revision");
|
|
2553
|
+
};
|
|
2554
|
+
/** An event takes effect only after both its journal append and its anchor entry are durable. */
|
|
2555
|
+
const commit = (event) => {
|
|
2556
|
+
if (!append(event)) return false;
|
|
2557
|
+
advanceAnchor();
|
|
2558
|
+
return true;
|
|
2559
|
+
};
|
|
2560
|
+
if (options.requireExisting && (!storage.exists(scopePath) || !storage.exists(indexPath) || !storage.exists(anchorPath))) fail("required retained history is missing");
|
|
2561
|
+
if (storage.exists(scopePath)) {
|
|
2562
|
+
if (storage.read(scopePath) !== scopeText) fail("resume changed its scope or allowances");
|
|
2563
|
+
if (!storage.read(indexPath)?.trim()) fail("dispatch history is missing or empty");
|
|
2564
|
+
if (!storage.read(anchorPath)?.trim()) fail("durable anchor is missing");
|
|
2565
|
+
} else {
|
|
2566
|
+
if (storage.exists(indexPath) || storage.exists(anchorPath)) fail("legacy history has no scope anchor; reconcile before migration");
|
|
2567
|
+
const created = storage.append(scopePath, scopeText, 0);
|
|
2568
|
+
if (created === void 0) {
|
|
2569
|
+
if (storage.read(scopePath) !== scopeText || !storage.read(indexPath)?.trim()) fail("scope initialization is busy; retry the same scope");
|
|
2570
|
+
} else if (created !== Buffer.byteLength(scopeText)) fail("storage returned an invalid scope revision");
|
|
2571
|
+
}
|
|
2465
2572
|
refresh();
|
|
2466
|
-
if (!opened && !
|
|
2573
|
+
if (!opened && !commit(header) && !opened) fail("history is busy; retry with the same run directory");
|
|
2467
2574
|
const readResult = (id) => {
|
|
2468
2575
|
const content = storage.read(resultPath(id));
|
|
2469
2576
|
if (content === void 0) return void 0;
|
|
2470
2577
|
const result = resultSchema.parse(JSON.parse(content));
|
|
2471
2578
|
if (result.id !== id) return fail("retained result dispatch id mismatch");
|
|
2472
2579
|
if (hashCanonical(result.value) !== result.receipt) return fail("retained result digest mismatch");
|
|
2580
|
+
const value = parse(result.value);
|
|
2581
|
+
if (hashCanonical(value) !== result.receipt) return fail("result decoder changed retained output");
|
|
2473
2582
|
return {
|
|
2474
|
-
value
|
|
2583
|
+
value,
|
|
2475
2584
|
receipt: result.receipt
|
|
2476
2585
|
};
|
|
2477
2586
|
};
|
|
@@ -2484,7 +2593,7 @@ function createRetainedDispatch(options) {
|
|
|
2484
2593
|
}
|
|
2485
2594
|
if (!existing) fail("settlement has no pending dispatch");
|
|
2486
2595
|
const previousRevision = revision;
|
|
2487
|
-
if (
|
|
2596
|
+
if (commit({
|
|
2488
2597
|
kind: "settled",
|
|
2489
2598
|
id,
|
|
2490
2599
|
receipt
|
|
@@ -2507,16 +2616,17 @@ function createRetainedDispatch(options) {
|
|
|
2507
2616
|
return {
|
|
2508
2617
|
/** The input binds all behavior-affecting request material, not just a caller-selected nonce. */
|
|
2509
2618
|
run(request) {
|
|
2619
|
+
const { lane, dispatch, signal } = request;
|
|
2510
2620
|
const id = hashCanonical([
|
|
2511
2621
|
header.scope,
|
|
2512
|
-
|
|
2622
|
+
lane,
|
|
2513
2623
|
request.input
|
|
2514
2624
|
]);
|
|
2515
2625
|
const pending = active.get(id);
|
|
2516
2626
|
if (pending) return pending;
|
|
2517
2627
|
const work = async () => {
|
|
2518
|
-
if (!limits.has(
|
|
2519
|
-
|
|
2628
|
+
if (!limits.has(lane)) return fail("unknown allowance lane");
|
|
2629
|
+
refresh();
|
|
2520
2630
|
while (true) {
|
|
2521
2631
|
const existing = calls.get(id);
|
|
2522
2632
|
if (existing) {
|
|
@@ -2536,24 +2646,24 @@ function createRetainedDispatch(options) {
|
|
|
2536
2646
|
replayed: true
|
|
2537
2647
|
};
|
|
2538
2648
|
}
|
|
2539
|
-
|
|
2540
|
-
if ((counts.get(
|
|
2649
|
+
signal?.throwIfAborted();
|
|
2650
|
+
if ((counts.get(lane) ?? 0) >= limits.get(lane)) return {
|
|
2541
2651
|
succeeded: false,
|
|
2542
2652
|
reason: "allowance_exhausted",
|
|
2543
2653
|
dispatchId: id
|
|
2544
2654
|
};
|
|
2545
2655
|
const previousRevision = revision;
|
|
2546
|
-
if (
|
|
2656
|
+
if (commit({
|
|
2547
2657
|
kind: "started",
|
|
2548
2658
|
id,
|
|
2549
|
-
lane
|
|
2659
|
+
lane
|
|
2550
2660
|
})) break;
|
|
2551
2661
|
if (revision === previousRevision) return fail("history is busy; retry with the same input");
|
|
2552
2662
|
}
|
|
2553
2663
|
let value;
|
|
2554
2664
|
try {
|
|
2555
|
-
|
|
2556
|
-
value = parse(await
|
|
2665
|
+
signal?.throwIfAborted();
|
|
2666
|
+
value = parse(await dispatch());
|
|
2557
2667
|
} catch (error) {
|
|
2558
2668
|
return {
|
|
2559
2669
|
succeeded: false,
|
|
@@ -3326,4 +3436,4 @@ function resolveWorktreePath(surface, worktreeDir) {
|
|
|
3326
3436
|
//#endregion
|
|
3327
3437
|
export { loadEvalFixtureScenarios as A, FsLabeledScenarioStore as C, neutralizationGate as D, rolloutArgumentDiff as E, analyzeCrossSurfaceInteractions as M, buildTraceAnalystSurfaceDispatch as N, discoverEvalFixtures as O, traceAnalystQualityJudge as P, neutralizeText as S, classifyUngroundedLiterals as T, scoreUserStory as _, autoevalsScorerJudge as a, scopedOptimizationMethod as b, selectDiscriminative as c, finalizeProfileMatrix as d, runProfileMatrixSegment as f, renderScoreboardMarkdown as g, makePlaybackDispatch as h, verifyCodeSurface as i, planEvalFixtureRun as j, loadEvalFixture as k, createRetainedDispatch as l, runProfileMatrix as m, gitWorktreeAdapter as n, phoenixEvaluatorJudge as o, ProfileMatrixError as p, resolveWorktreePath as r, scoreDiscrimination as s, WorktreeAdapterError as t, createProfileMatrixPlan as u, scoreboardSummary as v, LabeledScenarioStoreError as w, sequentialOptimizationMethod as x, userStoryScoreboard as y };
|
|
3328
3438
|
|
|
3329
|
-
//# sourceMappingURL=campaign-
|
|
3439
|
+
//# sourceMappingURL=campaign-Btk1yFfw.js.map
|