@tangle-network/agent-eval 0.144.10 → 0.144.11

package/CHANGELOG.md

@@ -20,6 +20,17 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 - Blind statement-equivalence protocol (`defineEquivalenceCheck`, `buildEquivalenceRecord`, `runEquivalenceCheck`): the two-arm design as a typed primitive with fail-loud refusals (`EquivalenceProtocolError`) — a non-blind arm, a wrong arm count, a refutation without its separating witness, or a mismatched checker strategy throws instead of recording.
 - `docs/verification-strategies.md`: the family, each member's failure mode, and the BCWW (4.6) formalization pilot as the worked example.
 
+## [0.144.11] - 2026-08-10 - GEPA candidate graph
+
+### Added
+
+- `readGepaCandidatePopulationArtifact()` returns GEPA's exact accepted candidates, parent indices, selection scores, and discovery counts from a verified artifact.
+- Direct GEPA method provenance addresses the candidate graph and its configured population bounds, selection identities, and candidate surface kind.
+
+### Fixed
+
+- The Python bridge preserves the official GEPA result graph from both published GEPA 0.1.4 and the pinned source API wrapper.
+
 ## [0.144.10] - 2026-08-10 - Optimizer candidate population
 
 ### Added

package/README.md

@@ -340,6 +340,9 @@ Each official method score records the optimizer and bridge package versions, so
 Call `readExternalOptimizerObservationArtifact()` with `provenance.observations` to read every distinct callback-submitted candidate.
 The reader verifies the artifact digest, canonical rows, sequence, candidate identities, and summary counts before returning candidates.
 This verification proves that the bytes match the supplied summary; use a summary from trusted method provenance when authenticity matters.
+For a direct standard GEPA run, call `readGepaCandidatePopulationArtifact()` with `provenance.gepaCandidatePopulation`.
+It returns GEPA's accepted candidates with exact parent indices, aggregate scores, per-case selection scores, and discovery evaluation counts.
+The callback artifact remains the complete source for rejected or refused proposals that GEPA did not add to its accepted population.
 
 The [optimizer guide](./docs/campaign-proposers.md) covers recipes, budgets, resuming, and data separation.
 The [runnable comparison](./examples/compare-optimization-methods/) can run GEPA, SkillOpt, or both.

package/dist/analyst/index.d.ts

@@ -1274,7 +1274,7 @@ declare const ANALYST_BENCHMARK_HELP = "agent-eval analyst-benchmark\n\nRun the
 declare const ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM = "sha256-canonical-source-manifest";
 declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM = "sha256-canonical-file-manifest";
 declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES: readonly string[];
-declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "951c42191f617a9128ac81ed1af189275397c5a2fe965c8a3c694cf91a26c508";
+declare const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "36d8f26723c19516348db0647de861f0ab1e672cb2da5aac8b3e34a1d6b49bdc";
 /** The published benchmark evidence was produced at this package version, by
  * the retired one-shot direct runner, before trace analysts moved to the
  * recursive DSPy RLM engine. Both evidence digests below are historical facts

package/dist/analyst/index.js

@@ -6,7 +6,7 @@ import { n as isProposalFinding, t as assertProposalFindings } from "../proposal
 import { a as RunCritic, c as buildSkillUsageReport, d as defaultIsMaterial, f as diffFindings, h as defineTraceAnalyst, i as runSemanticConceptJudge, l as emitSkillUsageFindings, m as defineCustomAnalyst, n as SEMANTIC_CONCEPT_JUDGE_VERSION, o as SKILL_USAGE_ANALYST, s as SkillUsageAnalyst, u as FindingsStore } from "../semantic-concept-judge-D1z-KepS.js";
 import { t as createDspyRlmTraceEngine } from "../dspy-rlm-engine-DbTk4JdR.js";
 import { a as scoreAnalystFindings, i as summarizeAnalystBenchmarkRunner, n as runAnalystBenchmark, r as traceStoreEvidenceResolver, t as registryBenchmarkRunner } from "../benchmark-CWeqGl7x.js";
-import { $ as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, A as summarizeCodeTraceCalibration, B as publicBenchmarkSystemPrompt, C as analystDefinitionAsymmetries, D as expandCodeTraceFailureBlocks, E as emptyPublicBenchmarkRunner, F as CODE_TRACE_BENCH_ANALYST_PROMPT, G as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, H as appendVerificationArtifactsToOtlp, I as MAX_INCORRECT_BLOCKS, J as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, K as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, L as MAX_INCORRECT_BLOCK_STEPS, M as analystInstructionsOverrideFromText, N as effectiveAnalystProtocolSha256, O as readAnalystBenchmarkArtifact, P as readAnalystInstructionsOverride, Q as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, R as publicBenchmarkProtocolSha256, S as AnalystExpressivenessError, T as adaptPublicBenchmarkFindings, U as loadCodeTraceVerificationArtifacts, V as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, W as parseVerificationOutcome, X as ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, Y as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, Z as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, _ as runReplVariableAnalystDefinition, a as primeAnalystProtocolSha256, at as ANALYST_BENCHMARK_OBSERVATIONS_FILE, b as runChunkedAnalystDefinition, c as nodeHttpPrimeBridgeTransport, ct as summarizeAgentRxCalibration, d as publicBenchmarkDistributions, dt as agentRxBenchmarkCase, et as analystBenchmarkDependencyLockDigest, f as publicBenchmarkSelectionReport, ft as agentRxPredictionsToFindings, g as rlmEngineLimits, h as publicRlmAnalystDefinition, ht as normalizeBenchmarkLabel, i as createPrimeBenchmarkRunner, it as ANALYST_BENCHMARK_MANIFEST_FILE, j as compareAnalystRunners, k as renderCodeTraceCalibrationMarkdown, l as loadPublicBenchmarkRows, lt as codeTraceBenchCase, m as createPublicBenchmarkRlmRunner, mt as roundAgentRxStep, n as runAnalystBenchmarkCommand, nt as ANALYST_BENCHMARK_COST_LEDGER_FILE, o as primeCodeTraceAnalystDefinition, ot as AGENT_RX_UPSTREAM_REVISION, p as selectPublicBenchmarkRows, pt as normalizeAgentRxCategory, q as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, r as renderAnalystBenchmarkMarkdown, rt as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, s as runInlineAnalystDefinition, st as renderAgentRxCalibrationMarkdown, t as ANALYST_BENCHMARK_HELP, tt as analystBenchmarkImplementationDigest, u as preparePublicAnalystBenchmark, ut as codeTracerPredictionsToFindings, v as createPublicBenchmarkDirectRunner, w as analystDefinitionProtocolSha256, x as decodeReplyRows, y as publicDirectAnalystDefinition, z as publicBenchmarkRlmInstructions } from "../benchmark-command-D3fEnKdf.js";
+import { $ as ANALYST_BENCHMARK_IMPLEMENTATION_SHA256, A as summarizeCodeTraceCalibration, B as publicBenchmarkSystemPrompt, C as analystDefinitionAsymmetries, D as expandCodeTraceFailureBlocks, E as emptyPublicBenchmarkRunner, F as CODE_TRACE_BENCH_ANALYST_PROMPT, G as ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM, H as appendVerificationArtifactsToOtlp, I as MAX_INCORRECT_BLOCKS, J as ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256, K as ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES, L as MAX_INCORRECT_BLOCK_STEPS, M as analystInstructionsOverrideFromText, N as effectiveAnalystProtocolSha256, O as readAnalystBenchmarkArtifact, P as readAnalystInstructionsOverride, Q as ANALYST_BENCHMARK_IMPLEMENTATION_FILES, R as publicBenchmarkProtocolSha256, S as AnalystExpressivenessError, T as adaptPublicBenchmarkFindings, U as loadCodeTraceVerificationArtifacts, V as DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES, W as parseVerificationOutcome, X as ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION, Y as ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256, Z as ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM, _ as runReplVariableAnalystDefinition, a as primeAnalystProtocolSha256, at as ANALYST_BENCHMARK_OBSERVATIONS_FILE, b as runChunkedAnalystDefinition, c as nodeHttpPrimeBridgeTransport, ct as summarizeAgentRxCalibration, d as publicBenchmarkDistributions, dt as agentRxBenchmarkCase, et as analystBenchmarkDependencyLockDigest, f as publicBenchmarkSelectionReport, ft as agentRxPredictionsToFindings, g as rlmEngineLimits, h as publicRlmAnalystDefinition, ht as normalizeBenchmarkLabel, i as createPrimeBenchmarkRunner, it as ANALYST_BENCHMARK_MANIFEST_FILE, j as compareAnalystRunners, k as renderCodeTraceCalibrationMarkdown, l as loadPublicBenchmarkRows, lt as codeTraceBenchCase, m as createPublicBenchmarkRlmRunner, mt as roundAgentRxStep, n as runAnalystBenchmarkCommand, nt as ANALYST_BENCHMARK_COST_LEDGER_FILE, o as primeCodeTraceAnalystDefinition, ot as AGENT_RX_UPSTREAM_REVISION, p as selectPublicBenchmarkRows, pt as normalizeAgentRxCategory, q as ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256, r as renderAnalystBenchmarkMarkdown, rt as ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE, s as runInlineAnalystDefinition, st as renderAgentRxCalibrationMarkdown, t as ANALYST_BENCHMARK_HELP, tt as analystBenchmarkImplementationDigest, u as preparePublicAnalystBenchmark, ut as codeTracerPredictionsToFindings, v as createPublicBenchmarkDirectRunner, w as analystDefinitionProtocolSha256, x as decodeReplyRows, y as publicDirectAnalystDefinition, z as publicBenchmarkRlmInstructions } from "../benchmark-command-BteMFN62.js";
 import { a as extractPrimeJsonObject, c as primeProtocolSha256, d as runPrimeExchange, i as emptyPrimeRawUsage, l as primeReplyDefect, n as buildPrimePrompt, o as mergePrimeRawUsage, r as buildPrimeRepairPrompt, s as normalizePrimeUsage, t as analystUsageReceiptFromPrimeUsage, u as projectPrimeTrajectory } from "../prime-protocol-BfSalTfR.js";
 //#region src/analyst/adapters.ts
 /**

package/dist/{benchmark-command-D3fEnKdf.js → benchmark-command-BteMFN62.js}

@@ -1208,7 +1208,7 @@ const ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES = Object.freeze([
 	"package.json",
 	"pnpm-lock.yaml"
 ]);
-const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "951c42191f617a9128ac81ed1af189275397c5a2fe965c8a3c694cf91a26c508";
+const ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 = "36d8f26723c19516348db0647de861f0ab1e672cb2da5aac8b3e34a1d6b49bdc";
 /** The published benchmark evidence was produced at this package version, by
 * the retired one-shot direct runner, before trace analysts moved to the
 * recursive DSPy RLM engine. Both evidence digests below are historical facts
@@ -6412,4 +6412,4 @@ function shellQuote(value) {
 //#endregion
 export { ANALYST_BENCHMARK_IMPLEMENTATION_SHA256 as $, summarizeCodeTraceCalibration as A, publicBenchmarkSystemPrompt as B, analystDefinitionAsymmetries as C, expandCodeTraceFailureBlocks as D, emptyPublicBenchmarkRunner as E, CODE_TRACE_BENCH_ANALYST_PROMPT as F, ANALYST_BENCHMARK_DEPENDENCY_LOCK_DIGEST_ALGORITHM as G, appendVerificationArtifactsToOtlp as H, MAX_INCORRECT_BLOCKS as I, ANALYST_BENCHMARK_EVIDENCE_DEPENDENCY_LOCK_SHA256 as J, ANALYST_BENCHMARK_DEPENDENCY_LOCK_FILES as K, MAX_INCORRECT_BLOCK_STEPS as L, analystInstructionsOverrideFromText as M, effectiveAnalystProtocolSha256 as N, readAnalystBenchmarkArtifact as O, readAnalystInstructionsOverride as P, ANALYST_BENCHMARK_IMPLEMENTATION_FILES as Q, publicBenchmarkProtocolSha256 as R, AnalystExpressivenessError as S, adaptPublicBenchmarkFindings as T, loadCodeTraceVerificationArtifacts as U, DEFAULT_MAX_VERIFICATION_ARTIFACT_BYTES as V, parseVerificationOutcome as W, ANALYST_BENCHMARK_EVIDENCE_PACKAGE_VERSION as X, ANALYST_BENCHMARK_EVIDENCE_IMPLEMENTATION_SHA256 as Y, ANALYST_BENCHMARK_IMPLEMENTATION_DIGEST_ALGORITHM as Z, runReplVariableAnalystDefinition as _, primeAnalystProtocolSha256 as a, ANALYST_BENCHMARK_OBSERVATIONS_FILE as at, runChunkedAnalystDefinition as b, nodeHttpPrimeBridgeTransport as c, summarizeAgentRxCalibration as ct, publicBenchmarkDistributions as d, agentRxBenchmarkCase as dt, analystBenchmarkDependencyLockDigest as et, publicBenchmarkSelectionReport as f, agentRxPredictionsToFindings as ft, rlmEngineLimits as g, publicRlmAnalystDefinition as h, normalizeBenchmarkLabel as ht, createPrimeBenchmarkRunner as i, ANALYST_BENCHMARK_MANIFEST_FILE as it, compareAnalystRunners as j, renderCodeTraceCalibrationMarkdown as k, loadPublicBenchmarkRows as l, codeTraceBenchCase as lt, createPublicBenchmarkRlmRunner as m, roundAgentRxStep as mt, runAnalystBenchmarkCommand as n, ANALYST_BENCHMARK_COST_LEDGER_FILE as nt, primeCodeTraceAnalystDefinition as o, AGENT_RX_UPSTREAM_REVISION as ot, selectPublicBenchmarkRows as p, normalizeAgentRxCategory as pt, ANALYST_BENCHMARK_DEPENDENCY_LOCK_SHA256 as q, renderAnalystBenchmarkMarkdown as r, ANALYST_BENCHMARK_LOCAL_RECEIPT_FILE as rt, runInlineAnalystDefinition as s, renderAgentRxCalibrationMarkdown as st, ANALYST_BENCHMARK_HELP as t, analystBenchmarkImplementationDigest as tt, preparePublicAnalystBenchmark as u, codeTracerPredictionsToFindings as ut, createPublicBenchmarkDirectRunner as v, analystDefinitionProtocolSha256 as w, decodeReplyRows as x, publicDirectAnalystDefinition as y, publicBenchmarkRlmInstructions as z };
 
-//# sourceMappingURL=benchmark-command-D3fEnKdf.js.map
+//# sourceMappingURL=benchmark-command-BteMFN62.js.map