@tangle-network/agent-eval 0.222.0 → 0.223.0

package/dist/benchmarks/index.js

@@ -1,9 +1,9 @@
 import { t as __exportAll } from "../rolldown-runtime-BJohB4LM.js";
 import { o as summarizeNumberSeries } from "../descriptive-cPwITVfy.js";
-import { t as mean } from "../mean-CdDYG-us.js";
-import { Z as runCampaign } from "../llm-judge-cgbhTKQp.js";
+import { t as mean } from "../mean-BPuUNznQ.js";
+import { Z as runCampaign } from "../llm-judge-D6JtIqiU.js";
 import { M as fsCampaignStorage } from "../external-optimizer-subprocess-U616hV_l.js";
-import "../campaign-ddr1N3yD.js";
+import "../campaign-DdOgQfp8.js";
 import { join, resolve } from "node:path";
 import { readFile, readdir, stat } from "node:fs/promises";
 import { parse } from "smol-toml";

package/dist/campaign/index.js

@@ -1,16 +1,16 @@
-import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-TP6ndL0B.js";
-import { $ as resolveRunDir, A as campaignCellSearchResult, C as surfaceDiff, D as createSearchHistoryReceipt, E as assertSearchHistoryMatchesState, F as uniform, M as searchProposalExecution, N as searchReceiptAccounting, O as verifySearchHistoryArtifact, P as asha, Q as planCampaignRun, S as searchEdgeId, X as runEval, Z as runCampaign, _ as crowdedFrontierParent, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, et as buildCellSchedule, f as runImprovementLoop, g as beam, h as aide, j as searchModelIdentity, k as verifySearchHistoryReceipt, m as isProposedCandidate, n as isTransientTransportFailure, p as runOptimization, r as transientDispatchFailure, t as llmJudge, tt as cellCachePath, u as costFromLedgerSummary, v as incumbent, w as surfaceNode, x as developmentClaim, y as incumbentWithOperatorBandit, z as defaultProductionGate } from "../llm-judge-cgbhTKQp.js";
-import { A as campaignSplitDigest, O as campaignCoverage, T as assertCampaignDesign, f as campaignBreakdown, j as campaignSplitDigestFromIdentities, k as campaignScenarioIdentity, l as canonicalDigest, p as campaignMeanComposite, u as emitLoopProvenance, v as powerPreflight } from "../campaign-evidence-DN8rWuBj.js";
+import { a as heldoutSignificance, i as dimensionRegressions, o as pairHoldout } from "../statistical-heldout-C9hz_AF9.js";
+import { $ as resolveRunDir, A as campaignCellSearchResult, C as surfaceDiff, D as createSearchHistoryReceipt, E as assertSearchHistoryMatchesState, F as uniform, M as searchProposalExecution, N as searchReceiptAccounting, O as verifySearchHistoryArtifact, P as asha, Q as planCampaignRun, S as searchEdgeId, X as runEval, Z as runCampaign, _ as crowdedFrontierParent, b as SearchRecorder, c as readGepaCandidatePopulationArtifact, et as buildCellSchedule, f as runImprovementLoop, g as beam, h as aide, j as searchModelIdentity, k as verifySearchHistoryReceipt, m as isProposedCandidate, n as isTransientTransportFailure, p as runOptimization, r as transientDispatchFailure, t as llmJudge, tt as cellCachePath, u as costFromLedgerSummary, v as incumbent, w as surfaceNode, x as developmentClaim, y as incumbentWithOperatorBandit, z as defaultProductionGate } from "../llm-judge-D6JtIqiU.js";
+import { A as campaignSplitDigest, O as campaignCoverage, T as assertCampaignDesign, f as campaignBreakdown, j as campaignSplitDigestFromIdentities, k as campaignScenarioIdentity, l as canonicalDigest, p as campaignMeanComposite, u as emitLoopProvenance, v as powerPreflight } from "../campaign-evidence-MjCjkb5a.js";
 import { a as SearchLedgerError, o as SearchLedgerIntegrityError } from "../search-ledger-file-DTBD6X-y.js";
 import { M as fsCampaignStorage, N as inMemoryCampaignStorage, j as createRunCostLedger, l as DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, u as DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS } from "../external-optimizer-subprocess-U616hV_l.js";
 import { a as componentSurfaceIdentityMaterial, c as surfaceDispatchRef, i as codeSurfaceIdentityMaterial, l as surfaceHash, o as renderSurfaceDiff, r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "../json-recovery-DK4MDm5r.js";
 import { a as makeProposalFinding } from "../types-BK0blWhe.js";
-import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDigest, O as searchPosterior, T as estimateNodeFromCells, _ as planSearchClaim, a as searchDivergence, b as verifySearchClaim, g as decideSearchClaim, h as SEARCH_CLAIM_RULE, i as runSearch, o as searchPolicyView, v as searchClaimDecision, w as estimateNode, y as searchClaimReserveUsd } from "../pareto-B9udAt2i.js";
+import { C as SEARCH_ESTIMATOR, D as searchEstimateMethod, E as searchCellSetDigest, O as searchPosterior, T as estimateNodeFromCells, _ as planSearchClaim, a as searchDivergence, b as verifySearchClaim, g as decideSearchClaim, h as SEARCH_CLAIM_RULE, i as runSearch, o as searchPolicyView, v as searchClaimDecision, w as estimateNode, y as searchClaimReserveUsd } from "../pareto-7_LEmYGG.js";
 import { a as validateSearchLedgerEvent, c as SearchStateView, d as searchNodeId, f as searchTaskSetDigest, i as replaySearchLedgerText, l as searchUnitScores, n as openSearchLedger, o as SEARCH_LEDGER_SCHEMA, r as parseSearchLedgerLine, s as SearchState, t as FileSearchLedger, u as searchCellId } from "../search-ledger-BsVi-Ngv.js";
 import { t as acquireSingleRunLock } from "../single-run-lock-BGlW3OSk.js";
-import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-Ct7SrmVc.js";
-import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-ddr1N3yD.js";
-import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-qB1M554K.js";
-import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-CBYeD4zi.js";
-import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-B1I0tRfL.js";
+import { n as searchProposerView, t as renderSearchSummary } from "../search-summary-Bt3uuK53.js";
+import { C as measureCampaignCost, S as analyzeCrossSurfaceInteractions, _ as neutralizationGate, a as createRetainedDispatch, b as loadEvalFixtureScenarios, c as runProfileMatrixSegment, d as scoreUserStory, f as scoreboardSummary, g as neutralizeText, h as sequentialOptimizationMethod, i as scoreDiscrimination, l as runProfileMatrix, m as scopedOptimizationMethod, n as resolveWorktreePath, o as createProfileMatrixPlan, p as userStoryScoreboard, r as verifyCodeSurface, s as finalizeProfileMatrix, t as gitWorktreeAdapter, u as renderScoreboardMarkdown, v as discoverEvalFixtures, x as planEvalFixtureRun, y as loadEvalFixture } from "../campaign-DdOgQfp8.js";
+import { a as importExternalEvaluations, c as composeGate, d as readExternalOptimizerObservationArtifact, i as externalSurface, l as externalTextOptimizationMethod, n as compareOptimizationMethods, o as importGepaPopulation, r as gepaOptimizationMethod, s as heldOutGate, t as skillOptOptimizationMethod, u as decodeExternalTextCandidate } from "../skillopt-optimization-method-BKFbJEyz.js";
+import { n as paretoPolicy, r as paretoSignificanceGate, t as buildEvidenceVector } from "../promotion-policy-D7B4_jEZ.js";
+import { n as sequentialPairedGate, t as sequentialDecide } from "../sequential-De7BTAlV.js";
 export { DEFAULT_EXTERNAL_OPTIMIZER_CALLBACK_LIMITS, DEFAULT_EXTERNAL_OPTIMIZER_PROCESS_LIMITS, FileSearchLedger, SEARCH_CLAIM_RULE, SEARCH_ESTIMATOR, SEARCH_LEDGER_SCHEMA, SearchLedgerError, SearchLedgerIntegrityError, SearchRecorder, SearchState, SearchStateView, acquireSingleRunLock, aide, analyzeCrossSurfaceInteractions, asha, assertCampaignDesign, assertCodeSurfaceIdentity, assertSearchHistoryMatchesState, beam, buildCellSchedule, buildEvidenceVector, campaignBreakdown, campaignCellSearchResult, campaignCoverage, campaignMeanComposite, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, canonicalDigest, cellCachePath, codeSurfaceIdentityMaterial, compareOptimizationMethods, componentSurfaceIdentityMaterial, composeGate, costFromLedgerSummary, createProfileMatrixPlan, createRetainedDispatch, createRunCostLedger, createSearchHistoryReceipt, crowdedFrontierParent, decideSearchClaim, decodeExternalTextCandidate, defaultProductionGate, developmentClaim, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, estimateNode, estimateNodeFromCells, externalSurface, externalTextOptimizationMethod, finalizeProfileMatrix, fsCampaignStorage, gepaOptimizationMethod, gitWorktreeAdapter, heldOutGate, heldoutSignificance, importExternalEvaluations, importGepaPopulation, inMemoryCampaignStorage, incumbent, incumbentWithOperatorBandit, isProposedCandidate, isTransientTransportFailure, llmJudge, loadEvalFixture, loadEvalFixtureScenarios, makeProposalFinding, measureCampaignCost, neutralizationGate, neutralizeText, openSearchLedger, pairHoldout, paretoPolicy, paretoSignificanceGate, parseSearchLedgerLine, planCampaignRun, planEvalFixtureRun, planSearchClaim, powerPreflight, readExternalOptimizerObservationArtifact, readGepaCandidatePopulationArtifact, renderScoreboardMarkdown, renderSearchSummary, renderSurfaceDiff, replaySearchLedgerText, resolveRunDir, resolveWorktreePath, runCampaign, runEval, runImprovementLoop, runOptimization, runProfileMatrix, runProfileMatrixSegment, runSearch, scopedOptimizationMethod, scoreDiscrimination, scoreUserStory, scoreboardSummary, searchCellId, searchCellSetDigest, searchClaimDecision, searchClaimReserveUsd, searchDivergence, searchEdgeId, searchEstimateMethod, searchModelIdentity, searchNodeId, searchPolicyView, searchPosterior, searchProposalExecution, searchProposerView, searchReceiptAccounting, searchTaskSetDigest, searchUnitScores, sequentialDecide, sequentialOptimizationMethod, sequentialPairedGate, skillOptOptimizationMethod, surfaceContentHash, surfaceDiff, surfaceDispatchRef, surfaceHash, surfaceNode, transientDispatchFailure, uniform, userStoryScoreboard, validateSearchLedgerEvent, verifyCodeSurface, verifySearchClaim, verifySearchHistoryArtifact, verifySearchHistoryReceipt };

package/dist/{campaign-ddr1N3yD.js → campaign-DdOgQfp8.js}

@@ -3,24 +3,25 @@ import { a as hashCanonical, i as compareCodeUnits, r as canonicalString } from
 import { a as agentProfileModelId, i as agentProfileId, r as agentProfileHash, s as harnessAxisOf } from "./agent-profile-Dngk3LZ0.js";
 import { i as buildAgentProfileCell } from "./agent-profile-cell-CT5JYueM.js";
 import { c as validateRunRecord, i as modelHasSnapshot, n as UNKNOWN_MODEL, o as runTaskScore } from "./run-record-DEjx43cI.js";
-import { t as comparePairedArms } from "./paired-arms-D-54hO-k.js";
-import { i as pairedBootstrap, t as mean } from "./mean-CdDYG-us.js";
-import { o as pairHoldout } from "./statistical-heldout-TP6ndL0B.js";
+import { t as comparePairedArms } from "./paired-arms-D5H_58e8.js";
+import { r as pairedBootstrap } from "./paired-tests-C2zUqgUW.js";
+import { n as meanOrNull, t as mean } from "./mean-BPuUNznQ.js";
+import { o as pairHoldout } from "./statistical-heldout-C9hz_AF9.js";
 import { d as mapConcurrent } from "./journal-C1uLJi0w.js";
 import { n as contentHash, t as canonicalJson } from "./verdict-cache-BVH-fl18.js";
-import { $ as resolveRunDir, Q as planCampaignRun, Z as runCampaign, l as combineComparisonCosts, nt as computeManifestHash, o as executeOptimizationMethod, tt as cellCachePath } from "./llm-judge-cgbhTKQp.js";
-import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-CZy4-TnZ.js";
-import { C as assertRealBackend, T as assertCampaignDesign, k as campaignScenarioIdentity, w as summarizeBackendIntegrity } from "./campaign-evidence-DN8rWuBj.js";
+import { $ as resolveRunDir, Q as planCampaignRun, Z as runCampaign, l as combineComparisonCosts, nt as computeManifestHash, o as executeOptimizationMethod, tt as cellCachePath } from "./llm-judge-D6JtIqiU.js";
+import { n as campaignCellToRunRecord, r as projectCampaignCellQuality, t as campaignCellCostProvenance } from "./run-record-BUm-Jggo.js";
+import { C as assertRealBackend, T as assertCampaignDesign, k as campaignScenarioIdentity, w as summarizeBackendIntegrity } from "./campaign-evidence-MjCjkb5a.js";
 import { M as fsCampaignStorage } from "./external-optimizer-subprocess-U616hV_l.js";
 import { r as assertCodeSurfaceIdentity, s as surfaceContentHash } from "./json-recovery-DK4MDm5r.js";
 import "./types-BK0blWhe.js";
-import "./pareto-B9udAt2i.js";
+import "./pareto-7_LEmYGG.js";
 import "./search-ledger-BsVi-Ngv.js";
-import "./search-summary-Ct7SrmVc.js";
+import "./search-summary-Bt3uuK53.js";
 import { r as verifyCompletion } from "./completion-verifier-rqbMh5-4.js";
 import "./external-optimizer-process-BRVmaDFe.js";
-import "./skillopt-optimization-method-qB1M554K.js";
-import "./sequential-B1I0tRfL.js";
+import "./skillopt-optimization-method-BKFbJEyz.js";
+import "./sequential-De7BTAlV.js";
 import { createHash } from "node:crypto";
 import { z } from "zod";
 import { closeSync, constants, existsSync, fstatSync, lstatSync, mkdirSync, mkdtempSync, openSync, readFileSync, readSync, readdirSync, readlinkSync, realpathSync, rmSync, statSync } from "node:fs";
@@ -531,9 +532,6 @@ function distribution(values) {
 		total
 	};
 }
-function meanOrNull$1(values) {
-	return values.length === 0 ? null : values.reduce((sum, value) => sum + value, 0) / values.length;
-}
 function unique(values) {
 	return [...new Set(values)];
 }
@@ -637,8 +635,8 @@ function interactionEffect(context, baseline, singles, composition) {
 		perTask,
 		n: complete.length,
 		nMissing: perTask.length - complete.length,
-		meanPassInteraction: meanOrNull$1(passInteractions),
-		meanScoreInteraction: meanOrNull$1(scoreInteractions),
+		meanPassInteraction: meanOrNull(passInteractions),
+		meanScoreInteraction: meanOrNull(scoreInteractions),
 		passBootstrap: bootstrapInteraction(passInteractions, context),
 		scoreBootstrap: bootstrapInteraction(scoreInteractions, context)
 	};
@@ -1689,9 +1687,6 @@ async function runProfileMatrix(opts) {
 function scoreOf(r) {
 	return runTaskScore(r);
 }
-function meanOrNull(values) {
-	return values.length === 0 ? null : mean(values);
-}
 function rollup(records, keyOf) {
 	const groups = /* @__PURE__ */ new Map();
 	for (const r of records) {
@@ -3101,4 +3096,4 @@ function resolveWorktreePath(surface, worktreeDir) {
 }
 export { measureCampaignCost as C, analyzeCrossSurfaceInteractions as S, neutralizationGate as _, createRetainedDispatch as a, loadEvalFixtureScenarios as b, runProfileMatrixSegment as c, scoreUserStory as d, scoreboardSummary as f, neutralizeText as g, sequentialOptimizationMethod as h, scoreDiscrimination as i, runProfileMatrix as l, scopedOptimizationMethod as m, resolveWorktreePath as n, createProfileMatrixPlan as o, userStoryScoreboard as p, verifyCodeSurface as r, finalizeProfileMatrix as s, gitWorktreeAdapter as t, renderScoreboardMarkdown as u, discoverEvalFixtures as v, planEvalFixtureRun as x, loadEvalFixture as y };
 
-//# sourceMappingURL=campaign-ddr1N3yD.js.map
+//# sourceMappingURL=campaign-DdOgQfp8.js.map