@tangle-network/agent-eval 0.112.0 → 0.114.0

package/CHANGELOG.md

@@ -4,6 +4,40 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.114.0] — 2026-07-11 — exact directional paired inference
+
+### Added
+
+- `pairedSignTest(differences, alternative)` computes an exact one-sided sign test for paired numeric differences, excludes zero ties, requires a predeclared `greater` or `less` direction, and reports every denominator.
+
+### Changed
+
+- McNemar's exact two-sided calculation now reuses the same log-space binomial-tail implementation without changing its public result contract.
+
+This release is additive.
+Existing consumers do not need to update unless they want the new statistic.
+
+## [0.113.0] — 2026-07-10 — immutable code candidates
+
+### Changed
+
+- `CodeSurface` is now a finalized, content-addressed code candidate.
+  `gitWorktreeAdapter.finalize()` records exact base/candidate commits, the final tree, and the SHA-256 + byte length of the raw binary Git patch; `surfaceHash` and `surfaceContentHash` no longer use filesystem paths.
+  Binary-patch generation runs against an isolated bare repository with fixed diff options, config, attributes, compression, and locale, so ambient repository, global, system, or environment settings cannot change the digest for identical trees.
+- `LoopProvenanceRecord.schema` is now `tangle.loop-provenance.v2`, distinguishing records that use content-addressed code-surface hashes from v1 records whose code hashes included mutable paths.
+- `resolveWorktreePath()` now verifies the candidate before returning its checkout.
+  Dirty or ignored files, moved refs, missing objects, wrong trees, raw byte or executable-mode mismatches, external symlinks, submodules, and patch-byte mismatches fail instead of being evaluated under stale identity.
+  Raw file hashing bypasses Git clean/smudge filters so repository configuration cannot hide different executable bytes.
+  Use `verifyCodeSurface()` when a verification receipt is needed directly.
+
+### Breaking
+
+- `resolveWorktreePath()` no longer returns a best-effort, unchecked locator: it verifies the finalized candidate and throws on any identity mismatch.
+  There is intentionally no lenient fallback on the evaluation path.
+- Path-only `CodeSurface` objects are invalid.
+  Every field in the finalized identity is required.
+  Downstream callers must migrate to adapter-finalized candidates when adopting 0.113.0.
+
 ## [0.112.0] — 2026-07-10 — complete agent-surface findings
 
 ### Added

package/dist/belief-state/index.js

@@ -17,7 +17,7 @@ import {
 } from "../chunk-DTJ6QUQB.js";
 import {
   confidenceInterval
-} from "../chunk-VC43KQCK.js";
+} from "../chunk-PJQFMIOX.js";
 import {
   ValidationError
 } from "../chunk-ONWEPEDO.js";

package/dist/benchmarks/index.d.ts

@@ -1,5 +1,5 @@
-export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, k as BenchmarkDistribution, c as BenchmarkEvaluation, d as BenchmarkFamily, l as BenchmarkMetricCalibrationOptions, m as BenchmarkMetricCalibrationResult, n as BenchmarkReport, e as BenchmarkResponder, o as BenchmarkRunOptions, p as BenchmarkRunResult, f as BenchmarkScenario, q as BenchmarkSliceSummary, g as BenchmarkSource, h as BenchmarkTaskKind, r as BuildStandardRetrievalItemsOptions, R as RetrievalIdAdapterOptions, S as StandardRetrievalArtifact, s as StandardRetrievalDocument, t as StandardRetrievalEvaluationOptions, u as StandardRetrievalPayload, v as StandardRetrievalQrel, w as StandardRetrievalQuery, x as StandardRetrievalResult, y as buildStandardRetrievalItems, z as calibrateBenchmarkMetric, A as createRetrievalIdBenchmarkAdapter, i as deterministicSplit, C as evaluateStandardRetrieval, D as normalizeRetrievedDocumentIds, E as parseBeirCorpusJsonl, F as parseBeirQueriesJsonl, G as parseJsonlRows, H as parseQrels, I as parseTsvRows, J as renderBenchmarkReportMarkdown, K as retrievalMetricsAtCutoff, L as routing, M as runBenchmarkAdapter, N as summarizeBenchmarkCampaign } from '../index-Dc3VLGhp.js';
-import '../types-v--ctu-b.js';
+export { B as BENCHMARK_SPLIT_SEED, a as BenchmarkAdapter, b as BenchmarkDatasetItem, k as BenchmarkDistribution, c as BenchmarkEvaluation, d as BenchmarkFamily, l as BenchmarkMetricCalibrationOptions, m as BenchmarkMetricCalibrationResult, n as BenchmarkReport, e as BenchmarkResponder, o as BenchmarkRunOptions, p as BenchmarkRunResult, f as BenchmarkScenario, q as BenchmarkSliceSummary, g as BenchmarkSource, h as BenchmarkTaskKind, r as BuildStandardRetrievalItemsOptions, R as RetrievalIdAdapterOptions, S as StandardRetrievalArtifact, s as StandardRetrievalDocument, t as StandardRetrievalEvaluationOptions, u as StandardRetrievalPayload, v as StandardRetrievalQrel, w as StandardRetrievalQuery, x as StandardRetrievalResult, y as buildStandardRetrievalItems, z as calibrateBenchmarkMetric, A as createRetrievalIdBenchmarkAdapter, i as deterministicSplit, C as evaluateStandardRetrieval, D as normalizeRetrievedDocumentIds, E as parseBeirCorpusJsonl, F as parseBeirQueriesJsonl, G as parseJsonlRows, H as parseQrels, I as parseTsvRows, J as renderBenchmarkReportMarkdown, K as retrievalMetricsAtCutoff, L as routing, M as runBenchmarkAdapter, N as summarizeBenchmarkCampaign } from '../index-CWr5SIG-.js';
+import '../types-CgSlO6wT.js';
 import '../run-record-DksGsfgv.js';
 import '@tangle-network/agent-interface';
 import '../errors-oeQrLqXC.js';

package/dist/benchmarks/index.js

@@ -16,15 +16,15 @@ import {
   routing_exports,
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
-} from "../chunk-VLNGPUJJ.js";
-import "../chunk-QQVYMBCZ.js";
-import "../chunk-RCXMWGRY.js";
+} from "../chunk-3LXTCTWL.js";
+import "../chunk-VHPD6AXX.js";
+import "../chunk-N6MTC3GK.js";
 import "../chunk-VI2UW6B6.js";
-import "../chunk-XNCF3JU3.js";
+import "../chunk-FAOEFFRT.js";
 import "../chunk-HJWNHCD5.js";
 import "../chunk-DWLIGZBX.js";
-import "../chunk-7QPK6W7R.js";
-import "../chunk-VC43KQCK.js";
+import "../chunk-ARU2PZFM.js";
+import "../chunk-PJQFMIOX.js";
 import "../chunk-RPDDVKI7.js";
 import "../chunk-GGE4NNQT.js";
 import "../chunk-LNQEP766.js";

package/dist/builder-eval/index.js

@@ -5,7 +5,7 @@ import {
 import {
   pearsonR,
   spearmanR
-} from "../chunk-VC43KQCK.js";
+} from "../chunk-PJQFMIOX.js";
 import {
   judgeSpans
 } from "../chunk-MHNQWM4I.js";

package/dist/campaign/index.d.ts

@@ -1,14 +1,14 @@
-import { S as SignedManifest, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-BepVVa6P.js';
-export { L as LlmJudgeDimension, c as LlmJudgeOptions, l as llmJudge } from '../pre-registration-BepVVa6P.js';
+import { S as SignedManifest, B as BackendIntegrityReport, C as CompletionRequirement, R as RuntimeEventLike, a as CompletionVerdict, P as ProducedState, b as CorrectnessChecker } from '../pre-registration-CTQbZbpX.js';
+export { L as LlmJudgeDimension, c as LlmJudgeOptions, l as llmJudge } from '../pre-registration-CTQbZbpX.js';
 import { A as AnalyzeTracesOptions, a as AnalyzeTracesInput, b as AnalyzeTracesResult } from '../analyst-C8HHvfJp.js';
-import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, g as Gate, e as GenerationRecord, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, f as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, y as LabeledScenarioSource, C as CampaignResult, m as CodeSurface } from '../types-v--ctu-b.js';
-export { i as CampaignAggregates, j as CampaignArtifactWriter, k as CampaignCellResult, l as CampaignCostMeter, z as CampaignTokenUsage, d as CampaignTraceWriter, c as DispatchFn, n as GateContext, h as GateDecision, G as GateResult, o as GenerationCandidate, A as JudgeAggregate, a as JudgeDimension, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-v--ctu-b.js';
-import { C as CampaignRunPlan, P as PlanCampaignRunOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions } from '../gepa-T8T215nw.js';
-export { f as CampaignRunPlanCell, h as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, i as OpenAutoPrResult, a as RunImprovementLoopResult, R as RunOptimizationOptions, j as RunOptimizationResult, k as countSentenceEdits, l as defaultRenderDiff, m as extractH2Sections, g as gepaProposer, o as openAutoPr, p as planCampaignRun, r as runCampaign, d as runImprovementLoop, n as runOptimization, s as surfaceHash } from '../gepa-T8T215nw.js';
+import { S as Scenario, M as MutableSurface, D as DispatchContext, b as JudgeConfig, g as Gate, e as GenerationRecord, J as JudgeScore, L as LabeledScenarioStore, s as LabeledScenarioWrite, t as LabeledScenarioSampleArgs, u as LabeledScenarioRecord, v as LabelTrust, f as SurfaceProposer, w as ProposedCandidate, x as ProposeContext, m as CodeSurface, y as LabeledScenarioSource, C as CampaignResult } from '../types-CgSlO6wT.js';
+export { i as CampaignAggregates, j as CampaignArtifactWriter, k as CampaignCellResult, l as CampaignCostMeter, z as CampaignTokenUsage, d as CampaignTraceWriter, c as DispatchFn, n as GateContext, h as GateDecision, G as GateResult, o as GenerationCandidate, A as JudgeAggregate, a as JudgeDimension, p as Mutator, O as OptimizationProposer, q as OptimizerConfig, P as ParetoParent, R as RedactionStatus, B as ScenarioAggregate, r as SessionScript, T as TraceSpan, E as isProposedCandidate, F as labelTrustRank } from '../types-CgSlO6wT.js';
+import { C as CampaignRunPlan, P as PlanCampaignRunOptions, b as RunCampaignOptions, c as RunImprovementLoopOptions } from '../gepa-DolL_Fko.js';
+export { f as CampaignRunPlanCell, h as GepaProposerConstraints, G as GepaProposerOptions, O as OpenAutoPrOptions, i as OpenAutoPrResult, a as RunImprovementLoopResult, R as RunOptimizationOptions, j as RunOptimizationResult, k as countSentenceEdits, l as defaultRenderDiff, m as extractH2Sections, g as gepaProposer, o as openAutoPr, p as planCampaignRun, r as runCampaign, d as runImprovementLoop, n as runOptimization } from '../gepa-DolL_Fko.js';
 import { C as CampaignStorage } from '../storage-Dw_f7WMt.js';
 export { f as fsCampaignStorage, i as inMemoryCampaignStorage } from '../storage-Dw_f7WMt.js';
-export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, l as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, m as EmitLoopProvenanceArgs, n as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, o as LoopProvenanceBackend, q as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, c as ParetoSignificanceGateOptions, P as PowerPreflight, s as PowerPreflightOptions, d as PromotionObjective, e as PromotionPolicy, R as RunEvalOptions, f as buildEvidenceVector, t as buildLoopProvenanceRecord, g as composeGate, h as defaultProductionGate, u as emitLoopProvenance, i as evolutionaryProposer, j as heldOutGate, v as loopProvenanceSpans, p as paretoPolicy, k as paretoSignificanceGate, w as powerPreflight, x as provenanceRecordPath, y as provenanceSpansPath, r as runEval, z as surfaceContentHash } from '../provenance-KhY8ESVM.js';
-import { E as EProcessState, a as PairedBootstrapResult } from '../statistics-CDfpwIdX.js';
+export { A as AxisEvidence, a as AxisVerdict, B as BuildEvidenceVectorOptions, l as BuildLoopProvenanceArgs, D as DefaultProductionGateOptions, m as EmitLoopProvenanceArgs, n as EmitLoopProvenanceResult, E as EvidenceVector, b as EvolutionaryProposerOptions, H as HeldOutGateOptions, o as LoopProvenanceBackend, q as LoopProvenanceCandidate, L as LoopProvenanceRecord, O as ObjectiveSource, c as ParetoSignificanceGateOptions, P as PowerPreflight, s as PowerPreflightOptions, d as PromotionObjective, e as PromotionPolicy, R as RunEvalOptions, f as buildEvidenceVector, t as buildLoopProvenanceRecord, g as composeGate, h as defaultProductionGate, u as emitLoopProvenance, i as evolutionaryProposer, j as heldOutGate, v as loopProvenanceSpans, p as paretoPolicy, k as paretoSignificanceGate, w as powerPreflight, x as provenanceRecordPath, y as provenanceSpansPath, r as runEval } from '../provenance-BZmpWmn4.js';
+import { E as EProcessState, a as PairedBootstrapResult } from '../statistics-oUbOJe-S.js';
 import { L as LlmClientOptions } from '../llm-client-DyqEH4jH.js';
 import { AgentProfile } from '@tangle-network/agent-interface';
 import { A as AgentEvalError } from '../errors-oeQrLqXC.js';
@@ -1013,6 +1013,17 @@ interface ParameterSweepProposerOptions {
 /** Config/parameter-level proposer for FAPO's middle escalation level. */
 declare function parameterSweepProposer(opts: ParameterSweepProposerOptions): SurfaceProposer;
 
+/** Fail when a code surface does not carry a complete immutable identity. */
+declare function assertCodeSurfaceIdentity(surface: unknown): asserts surface is CodeSurface;
+/** Canonical, location-independent identity of a finalized code candidate.
+ *  Commit metadata is excluded: two commits with the same base, final tree,
+ *  and patch bytes are the same executable candidate. */
+declare function codeSurfaceIdentityMaterial(surface: CodeSurface): string;
+/** Full SHA-256 content identity for a prompt or finalized code surface. */
+declare function surfaceContentHash(surface: MutableSurface): `sha256:${string}`;
+/** Short loop key derived from the same content identity as provenance. */
+declare function surfaceHash(surface: MutableSurface): string;
+
 /**
  * `compareProposers` — a head-to-head lift benchmark across surface proposers
  * on ONE corpus. This is the forcing function: optimizer quality (GEPA
@@ -2208,21 +2219,28 @@ declare function isTransientTransportFailure(message: string | null | undefined,
  * VCS-pluggable worktree adapter. One improvement = one worktree, PR-like
  * (multiple commits allowed). A code-tier proposer's `propose()` creates a
  * worktree, an agent commits the change into it, and `finalize()` returns a
- * `CodeSurface{ worktreeRef }` the measurement checks out to run the worker
- * against the changed code. On promotion the worktree becomes the PR branch.
+ * content-addressed `CodeSurface` the measurement verifies before running.
+ * On promotion the worktree becomes the PR branch.
  *
  * The interface is VCS-agnostic so a future `jj` ([jj-vcs](https://github.com/jj-vcs/jj))
  * adapter can slot in without touching proposer code. Only the git adapter
  * ships today. See `docs/design/loop-taxonomy.md`.
  */
 
+type GitOutput = string | Uint8Array;
+type GitEnvironment = Readonly<Record<string, string>>;
+type GitRunner = (args: string[], cwd: string, env?: GitEnvironment) => GitOutput;
 interface Worktree {
     /** Absolute path to the checked-out worktree directory. */
-    path: string;
+    readonly path: string;
     /** The branch the worktree is on (becomes the PR branch on promotion). */
-    branch: string;
+    readonly branch: string;
     /** The ref the worktree was forked from. */
-    baseRef: string;
+    readonly baseRef: string;
+    /** Exact commit `baseRef` resolved to before the worktree was created. */
+    readonly baseCommit: string;
+    /** Exact tree object for `baseCommit`. */
+    readonly baseTree: string;
 }
 interface WorktreeAdapter {
     /** Create an isolated worktree on a fresh branch off `baseRef`. */
@@ -2230,9 +2248,8 @@ interface WorktreeAdapter {
         baseRef: string;
         label: string;
     }): Promise<Worktree>;
-    /** Commit any pending changes in the worktree, then return a CodeSurface
-     *  pointing at it. The agent has already written its change into
-     *  `worktree.path` by the time this is called. */
+    /** Commit pending changes, freeze the exact Git objects + binary patch, and
+     *  verify the worktree still matches that identity. */
     finalize(worktree: Worktree, summary: string): Promise<CodeSurface>;
     /** Remove the worktree (and its branch) — called for losing candidates. */
     discard(worktree: Worktree): Promise<void>;
@@ -2249,16 +2266,32 @@ interface GitWorktreeAdapterOptions {
     worktreeDir?: string;
     /** Branch-name prefix. Default: `improve`. */
     branchPrefix?: string;
-    /** Test seam — defaults to a real `git` runner. */
-    git?: (args: string[], cwd: string) => string;
+    /** Test seam — defaults to a real `git` runner. The return value must contain
+     *  stdout verbatim, and runners that execute Git must forward the optional
+     *  environment overrides used to isolate patch generation. */
+    git?: GitRunner;
+}
+interface CodeSurfaceVerification {
+    /** Verified worktree path. */
+    path: string;
+    /** Git's canonical root for the verified checkout. */
+    repoRoot: string;
+    /** Recomputed full content identity. */
+    contentHash: `sha256:${string}`;
+    /** Exact verified binary-patch bytes. Candidate-bundle builders encode this
+     *  directly instead of reproducing Git diff options. */
+    patchBytes: Uint8Array;
 }
 /**
  * Git-backed `WorktreeAdapter`: creates isolated worktrees on fresh branches, commits agent changes, and discards losers.
  */
 declare function gitWorktreeAdapter(opts: GitWorktreeAdapterOptions): WorktreeAdapter;
-/** Resolve a `CodeSurface`'s worktreeRef to a directory the measurement can
- *  run the worker in. A path ref is returned as-is; anything else is treated
- *  as a ref under the adapter's worktree dir. */
+/** Verify a finalized code surface against its current checkout. This rejects
+ *  dirty/ignored files, moved refs, missing Git objects, raw byte/mode
+ *  mismatches, external symlinks, and submodules. */
+declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string): CodeSurfaceVerification;
+/** Resolve a code candidate for evaluation only after verifying its immutable
+ *  identity against the checkout at `worktreeRef`. */
 declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
 
-export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type ApplySkillPatchResult, type BuildAnalystSurfaceDispatchOptions, type CampaignBreakdown, CampaignResult, CampaignRunPlan, CampaignStorage, CodeSurface, type CompareProposersOptions, type CompositeProposerOptions, type DimensionRegression, type DiscriminationScore, DispatchContext, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoEntryConfig, type FapoFailureCluster, type FapoOptimizationLevel, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, Gate, GenerationRecord, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPrimitive, type JsonValue, JudgeConfig, JudgeScore, LabelTrust, LabeledScenarioRecord, LabeledScenarioSampleArgs, LabeledScenarioSource, LabeledScenarioStore, LabeledScenarioStoreError, LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, type LoadEvalFixtureScenariosOptions, type MemoryCurationProposerOptions, MutableSurface, type NeutralizationGateOptions, type OptimizerEntryConfig, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditProposerOptions, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, ProposeContext, type ProposePatchesArgs, ProposedCandidate, type ProposerComparison, type ProposerEntry, type ProposerPairwise, type ProposerScore, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, RunCampaignOptions, RunImprovementLoopOptions, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, Scenario, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, SurfaceProposer, type SurfaceScore, type TraceAnalystProposerOptions, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, applySkillPatch, buildAnalystSurfaceDispatch, callbackGovernor, campaignBreakdown, campaignMeanComposite, classifyUngroundedLiterals, compareProposers, compositeProposer, detectScale, dimensionRegressions, discoverEvalFixtures, extractFapoAttributionSignals, failureModeRecallJudge, fapoEscalationEntry, fapoProposer, fsLineageStore, gepaParetoEntry, gepaReflectionEntry, gitWorktreeAdapter, haloProposer, heldoutSignificance, heuristicGovernor, isTransientTransportFailure, lineageNodeId, loadEvalFixture, loadEvalFixtureScenarios, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, pairHoldout, parameterSweepProposer, parseSkillPatchResponse, patchEditCount, planEvalFixtureRun, policyEditProposer, renderScoreboardMarkdown, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runLineage, runLineageLoop, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptEntry, skillOptProposer, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard };
+export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type ApplySkillPatchResult, type BuildAnalystSurfaceDispatchOptions, type CampaignBreakdown, CampaignResult, CampaignRunPlan, CampaignStorage, CodeSurface, type CodeSurfaceVerification, type CompareProposersOptions, type CompositeProposerOptions, type DimensionRegression, type DiscriminationScore, DispatchContext, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoEntryConfig, type FapoFailureCluster, type FapoOptimizationLevel, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, Gate, GenerationRecord, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPrimitive, type JsonValue, JudgeConfig, JudgeScore, LabelTrust, LabeledScenarioRecord, LabeledScenarioSampleArgs, LabeledScenarioSource, LabeledScenarioStore, LabeledScenarioStoreError, LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, type LoadEvalFixtureScenariosOptions, type MemoryCurationProposerOptions, MutableSurface, type NeutralizationGateOptions, type OptimizerEntryConfig, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditProposerOptions, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, ProposeContext, type ProposePatchesArgs, ProposedCandidate, type ProposerComparison, type ProposerEntry, type ProposerPairwise, type ProposerScore, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, RunCampaignOptions, RunImprovementLoopOptions, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, Scenario, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, SurfaceProposer, type SurfaceScore, type TraceAnalystProposerOptions, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, applySkillPatch, assertCodeSurfaceIdentity, buildAnalystSurfaceDispatch, callbackGovernor, campaignBreakdown, campaignMeanComposite, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareProposers, compositeProposer, detectScale, dimensionRegressions, discoverEvalFixtures, extractFapoAttributionSignals, failureModeRecallJudge, fapoEscalationEntry, fapoProposer, fsLineageStore, gepaParetoEntry, gepaReflectionEntry, gitWorktreeAdapter, haloProposer, heldoutSignificance, heuristicGovernor, isTransientTransportFailure, lineageNodeId, loadEvalFixture, loadEvalFixtureScenarios, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, pairHoldout, parameterSweepProposer, parseSkillPatchResponse, patchEditCount, planEvalFixtureRun, policyEditProposer, renderScoreboardMarkdown, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runLineage, runLineageLoop, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, sequentialDecide, sequentialPairedGate, skillOptEntry, skillOptProposer, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard, verifyCodeSurface };

package/dist/campaign/index.js

@@ -55,13 +55,16 @@ import {
   skillOptEntry,
   skillOptProposer,
   traceAnalystProposer,
-  userStoryScoreboard
-} from "../chunk-QQVYMBCZ.js";
+  userStoryScoreboard,
+  verifyCodeSurface
+} from "../chunk-VHPD6AXX.js";
 import {
+  assertCodeSurfaceIdentity,
   buildEvidenceVector,
   buildLoopProvenanceRecord,
   campaignBreakdown,
   campaignMeanComposite,
+  codeSurfaceIdentityMaterial,
   composeGate,
   countSentenceEdits,
   defaultProductionGate,
@@ -89,7 +92,7 @@ import {
   runOptimization,
   surfaceContentHash,
   surfaceHash
-} from "../chunk-RCXMWGRY.js";
+} from "../chunk-N6MTC3GK.js";
 import "../chunk-VI2UW6B6.js";
 import {
   fsCampaignStorage,
@@ -98,11 +101,11 @@ import {
   resolveRunDir,
   runCampaign,
   tangleTracesRoot
-} from "../chunk-XNCF3JU3.js";
+} from "../chunk-FAOEFFRT.js";
 import "../chunk-HJWNHCD5.js";
 import "../chunk-DWLIGZBX.js";
-import "../chunk-7QPK6W7R.js";
-import "../chunk-VC43KQCK.js";
+import "../chunk-ARU2PZFM.js";
+import "../chunk-PJQFMIOX.js";
 import "../chunk-RPDDVKI7.js";
 import "../chunk-GGE4NNQT.js";
 import "../chunk-LNQEP766.js";
@@ -123,6 +126,7 @@ export {
   aceProposer,
   acquireSingleRunLock,
   applySkillPatch,
+  assertCodeSurfaceIdentity,
   buildAnalystSurfaceDispatch,
   buildEvidenceVector,
   buildLoopProvenanceRecord,
@@ -130,6 +134,7 @@ export {
   campaignBreakdown,
   campaignMeanComposite,
   classifyUngroundedLiterals,
+  codeSurfaceIdentityMaterial,
   compareProposers,
   composeGate,
   compositeProposer,
@@ -207,6 +212,7 @@ export {
   surfaceHash,
   tangleTracesRoot,
   traceAnalystProposer,
-  userStoryScoreboard
+  userStoryScoreboard,
+  verifyCodeSurface
 };
 //# sourceMappingURL=index.js.map

package/dist/{chunk-VLNGPUJJ.js → chunk-3LXTCTWL.js}

@@ -1,7 +1,7 @@
 import {
   fsCampaignStorage,
   runCampaign
-} from "./chunk-XNCF3JU3.js";
+} from "./chunk-FAOEFFRT.js";
 import {
   __export
 } from "./chunk-PZ5AY32C.js";
@@ -763,4 +763,4 @@ export {
   retrievalMetricsAtCutoff,
   benchmarks_exports
 };
-//# sourceMappingURL=chunk-VLNGPUJJ.js.map
+//# sourceMappingURL=chunk-3LXTCTWL.js.map

package/dist/{chunk-7QPK6W7R.js → chunk-ARU2PZFM.js}

@@ -1,6 +1,6 @@
 import {
   pearsonR
-} from "./chunk-VC43KQCK.js";
+} from "./chunk-PJQFMIOX.js";
 
 // src/rl/verifiable-reward.ts
 var DEFAULT_DETERMINISTIC_LAYERS = /* @__PURE__ */ new Set([
@@ -309,4 +309,4 @@ export {
   filterDeterministicallyRewarded,
   detectRewardHacking
 };
-//# sourceMappingURL=chunk-7QPK6W7R.js.map
+//# sourceMappingURL=chunk-ARU2PZFM.js.map

package/dist/{chunk-EMKORATZ.js → chunk-DPZAEKA6.js}

@@ -5,7 +5,7 @@ import {
   pairedBootstrap,
   pairedMde,
   wilcoxonSignedRank
-} from "./chunk-VC43KQCK.js";
+} from "./chunk-PJQFMIOX.js";
 import {
   canonicalize,
   hashJson
@@ -877,4 +877,4 @@ export {
   RESEARCH_REPORT_HARD_PAIR_FLOOR,
   researchReport
 };
-//# sourceMappingURL=chunk-EMKORATZ.js.map
+//# sourceMappingURL=chunk-DPZAEKA6.js.map

package/dist/{chunk-XNCF3JU3.js → chunk-FAOEFFRT.js}

@@ -1,6 +1,6 @@
 import {
   confidenceInterval
-} from "./chunk-VC43KQCK.js";
+} from "./chunk-PJQFMIOX.js";
 import {
   AgentEvalError
 } from "./chunk-ONWEPEDO.js";
@@ -787,4 +787,4 @@ export {
   runCampaign,
   planCampaignRun
 };
-//# sourceMappingURL=chunk-XNCF3JU3.js.map
+//# sourceMappingURL=chunk-FAOEFFRT.js.map

package/dist/{chunk-WZOIFFG2.js → chunk-LOBMT6SB.js}

@@ -1,6 +1,6 @@
 import {
   researchReport
-} from "./chunk-EMKORATZ.js";
+} from "./chunk-DPZAEKA6.js";
 import {
   RunIntegrityError,
   assertRunCaptured
@@ -354,4 +354,4 @@ function defaultRunId(params) {
 export {
   runEvalCampaign
 };
-//# sourceMappingURL=chunk-WZOIFFG2.js.map
+//# sourceMappingURL=chunk-LOBMT6SB.js.map

package/dist/{chunk-RT2AFUXM.js → chunk-MOXWMGPC.js}

@@ -1,6 +1,6 @@
 import {
   summaryTable
-} from "./chunk-EMKORATZ.js";
+} from "./chunk-DPZAEKA6.js";
 import {
   VerificationError
 } from "./chunk-ONWEPEDO.js";
@@ -574,4 +574,4 @@ export {
   judgeReplayGate,
   renderReleaseReport
 };
-//# sourceMappingURL=chunk-RT2AFUXM.js.map
+//# sourceMappingURL=chunk-MOXWMGPC.js.map

package/dist/{chunk-RCXMWGRY.js → chunk-N6MTC3GK.js}

@@ -1,13 +1,13 @@
 import {
   runCampaign,
   summarizeBackendIntegrity
-} from "./chunk-XNCF3JU3.js";
+} from "./chunk-FAOEFFRT.js";
 import {
   detectRewardHacking
-} from "./chunk-7QPK6W7R.js";
+} from "./chunk-ARU2PZFM.js";
 import {
   pairedBootstrap
-} from "./chunk-VC43KQCK.js";
+} from "./chunk-PJQFMIOX.js";
 import {
   DEFAULT_REDACTION_RULES
 } from "./chunk-GGE4NNQT.js";
@@ -1928,8 +1928,66 @@ function campaignBreakdown(campaign) {
   return { dimensions, scenarios };
 }
 
-// src/campaign/presets/run-optimization.ts
+// src/campaign/surface-identity.ts
 import { createHash } from "crypto";
+var GIT_OBJECT_ID = /^(?:[a-f0-9]{40}|[a-f0-9]{64})$/;
+var SHA256 = /^sha256:[a-f0-9]{64}$/;
+function assertCodeSurfaceIdentity(surface) {
+  if (!surface || typeof surface !== "object") {
+    throw new TypeError("CodeSurface must be an object");
+  }
+  const candidate = surface;
+  if (candidate.kind !== "code") throw new TypeError('CodeSurface.kind must be "code"');
+  if (typeof candidate.worktreeRef !== "string" || candidate.worktreeRef.trim().length === 0) {
+    throw new TypeError("CodeSurface.worktreeRef must be a non-empty locator");
+  }
+  if (typeof candidate.baseRef !== "string" || candidate.baseRef.trim().length === 0) {
+    throw new TypeError("CodeSurface.baseRef must be a non-empty ref label");
+  }
+  for (const [field, value] of [
+    ["baseCommit", candidate.baseCommit],
+    ["baseTree", candidate.baseTree],
+    ["candidateCommit", candidate.candidateCommit],
+    ["candidateTree", candidate.candidateTree]
+  ]) {
+    if (typeof value !== "string" || !GIT_OBJECT_ID.test(value)) {
+      throw new TypeError(`CodeSurface.${field} must be a full Git object id`);
+    }
+  }
+  const patch = candidate.patch;
+  if (!patch || typeof patch !== "object" || patch.format !== "git-diff-binary") {
+    throw new TypeError('CodeSurface.patch.format must be "git-diff-binary"');
+  }
+  if (typeof patch.sha256 !== "string" || !SHA256.test(patch.sha256)) {
+    throw new TypeError("CodeSurface.patch.sha256 must be a sha256 digest");
+  }
+  if (!Number.isSafeInteger(patch.byteLength) || patch.byteLength < 0) {
+    throw new TypeError("CodeSurface.patch.byteLength must be a non-negative safe integer");
+  }
+}
+function codeSurfaceIdentityMaterial(surface) {
+  assertCodeSurfaceIdentity(surface);
+  return JSON.stringify({
+    schema: "tangle.code-surface.v1",
+    baseCommit: surface.baseCommit,
+    baseTree: surface.baseTree,
+    candidateTree: surface.candidateTree,
+    patch: {
+      format: surface.patch.format,
+      sha256: surface.patch.sha256,
+      byteLength: surface.patch.byteLength
+    }
+  });
+}
+function surfaceContentHash(surface) {
+  const material = typeof surface === "string" ? surface : codeSurfaceIdentityMaterial(surface);
+  return `sha256:${createHash("sha256").update(material).digest("hex")}`;
+}
+function surfaceHash(surface) {
+  return surfaceContentHash(surface).slice("sha256:".length, "sha256:".length + 16);
+}
+
+// src/campaign/presets/run-optimization.ts
 async function runOptimization(opts) {
   const { proposer } = opts;
   const promoteTopK = opts.promoteTopK ?? 2;
@@ -2098,14 +2156,6 @@ function computeParetoFrontier(scored) {
   }));
   return paretoFrontier(scored, objectives).frontier;
 }
-function surfaceHash(surface) {
-  const material = typeof surface === "string" ? surface : JSON.stringify({
-    kind: surface.kind,
-    worktreeRef: surface.worktreeRef,
-    baseRef: surface.baseRef ?? null
-  });
-  return createHash("sha256").update(material).digest("hex").slice(0, 16);
-}
 
 // src/campaign/presets/run-improvement-loop.ts
 var DEFAULT_DISPATCH_TIMEOUT_MS = 6e5;
@@ -2135,7 +2185,7 @@ async function runImprovementLoop(opts) {
   const dispatchTimeoutMs = opts.dispatchTimeoutMs ?? DEFAULT_DISPATCH_TIMEOUT_MS;
   const optimization = await runOptimization({ ...opts, dispatchTimeoutMs });
   const winnerIsBaseline = optimization.winnerSurfaceHash === surfaceHash(opts.baselineSurface);
-  const { runCampaign: runCampaign2 } = await import("./run-campaign-HJQEDCJQ.js");
+  const { runCampaign: runCampaign2 } = await import("./run-campaign-UADIM77S.js");
   const baselineOnHoldout = await runCampaign2({
     ...opts,
     dispatchTimeoutMs,
@@ -2273,14 +2323,6 @@ function evolutionaryProposer(opts) {
 // src/campaign/provenance.ts
 import { createHash as createHash2 } from "crypto";
 import { join as join2 } from "path";
-function surfaceContentHash(surface) {
-  const material = typeof surface === "string" ? surface : JSON.stringify({
-    kind: surface.kind,
-    worktreeRef: surface.worktreeRef,
-    baseRef: surface.baseRef ?? null
-  });
-  return `sha256:${createHash2("sha256").update(material).digest("hex")}`;
-}
 function meanHoldoutComposite(campaign) {
   const xs = [];
   for (const cell of campaign.cells) {
@@ -2314,7 +2356,7 @@ function buildLoopProvenanceRecord(args) {
   const baselineHoldoutComposite = meanHoldoutComposite(args.baselineOnHoldout);
   const winnerHoldoutComposite = meanHoldoutComposite(args.winnerOnHoldout);
   const record = {
-    schema: "tangle.loop-provenance.v1",
+    schema: "tangle.loop-provenance.v2",
     runId: args.runId,
     runDir: args.runDir,
     timestamp: args.timestamp,
@@ -2576,17 +2618,19 @@ export {
   countSentenceEdits,
   campaignMeanComposite,
   campaignBreakdown,
-  runOptimization,
+  assertCodeSurfaceIdentity,
+  codeSurfaceIdentityMaterial,
+  surfaceContentHash,
   surfaceHash,
+  runOptimization,
   runImprovementLoop,
   defaultRenderDiff,
   runEval,
   evolutionaryProposer,
-  surfaceContentHash,
   buildLoopProvenanceRecord,
   loopProvenanceSpans,
   provenanceRecordPath,
   provenanceSpansPath,
   emitLoopProvenance
 };
-//# sourceMappingURL=chunk-RCXMWGRY.js.map
+//# sourceMappingURL=chunk-N6MTC3GK.js.map