@tangle-network/agent-eval 0.119.1 → 0.120.0

package/CHANGELOG.md

@@ -4,6 +4,23 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 
 ---
 
+## [0.120.0] — 2026-07-14 — exact measured provenance
+
+### Changed
+
+- Give each campaign a canonical split digest over the full scenario payload and replicate count, and require the exported benchmark to identify that exact heldout design.
+- Emit self-addressed `tangle.loop-provenance` records from one shared loop-result translator, and derive prompt or code diffs from the exact measured surfaces instead of accepting caller-authored text.
+- Retain one strict hash per scenario so consumers can verify the tested split without persisting customer task payloads.
+
+### Breaking
+
+- Require the exact baseline surface and bind portable evidence to every search, winner, heldout, and neutralized campaign; the full decision; the reconstructed receipt ledger; duration; power analysis; and validated candidate history.
+- Require every scored row to link to a successful model receipt; reject contradictory provenance, duplicate scenarios, broken cell identity or pairing, failed judges, and invalid cost, latency, or token measurements before products can publish an improvement proposal.
+- Stop optimization when a proposer returns no candidates, reject empty or non-contiguous generation records, and refuse a shipped no-op.
+- Remove the duplicate worker-record collector and caller-authored diff hook; settled receipts and exact surfaces are now the only evidence path.
+- Remove `@tangle-network/agent-eval/primeintellect`, which generated a legacy single-turn Verifiers package with unsafe substring scoring. Use `@tangle-network/agent-runtime/primeintellect` for task packaging, real runtime execution, and full trace import.
+- Verify the same npm archive implementation used by the release workflow.
+
 ## [0.119.1] — 2026-07-14 — portable improvement evidence
 
 ### Added
@@ -127,8 +144,7 @@ All notable changes to `@tangle-network/agent-eval` and its sibling `agent-eval-
 - `runOptimization({ promoteTopK })` accepts only `1`; multiple concurrent incumbents were never represented by the optimizer state and now fail before dispatch.
 - `ScoredSurfaceOutcome` requires `split: 'search'` and the actual `generation` that measured the surface.
 - `llmPolicyEditProposer()` requires explicit raw-score objectives and `PolicyEditFindingInput` rows whose source is either an exact measured surface-generation pair or an explicitly global finding.
-- `LoopProvenanceRecord.schema` is now `tangle.loop-provenance.v3`.
-  Consumers of v2 records must migrate to the v3 baseline score, parent chain, coverage, eligibility, and exact surface fields; no compatibility shim is provided.
+- Loop provenance preserves complete candidate measurement lineage: baseline score, parent chain, coverage, eligibility, and exact surface fields.
 
 ## [0.115.3] — 2026-07-12 — fail-closed structured output parsing
 
@@ -192,7 +208,7 @@ Existing consumers do not need to update unless they want the new statistic.
 - `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.
+- Code-surface provenance now uses content-addressed hashes instead of 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.

package/README.md

@@ -95,31 +95,6 @@ const traceRuns = fromOtelSpans({ spans: yourOtelSpans })
 await analyzeRuns({ runs: [...runs, ...traceRuns], raterScores })
 ```
 
-### 4. Export to PrimeIntellect
-
-Use this when you want PrimeIntellect Verifiers to run or train against tasks produced by Tangle evals, runtime runs, or knowledge runs.
-
-```ts
-import {
-  primeIntellectRowsFromRunRecords,
-  writePrimeIntellectEnvironmentPackage,
-} from '@tangle-network/agent-eval/primeintellect'
-
-const rows = primeIntellectRowsFromRunRecords({
-  records, // RunRecord[] with scenarioId
-  scenarios, // prompts plus answer or requiredSubstrings
-})
-
-await writePrimeIntellectEnvironmentPackage('prime-envs/refund-policy', {
-  name: 'refund-policy',
-  rows,
-  runRecords: records,
-})
-```
-
-The generated package contains `load_environment()`, `data/dataset.jsonl`, `data/run_records.jsonl`, and a manifest.
-Run it with `uv run vf-eval refund-policy` or upload it with `prime env push`.
-
 ---
 
 ## Core concepts
@@ -149,7 +124,6 @@ Each example: `README.md` + a single `index.ts` runnable via `pnpm tsx`. Prints
 |---|---|
 | `…/contract` | **The headline, frozen surface — new code starts here.** `defineAgentEval`, `selfImprove`, `analyzeRuns`, `runEval`, `runCampaign`, `runImprovementLoop`, `diffRuns`; intake adapters (`fromFeedbackTable`, `fromOtelSpans`); proposers (`gepaProposer`, `evolutionaryProposer`); gates (`defaultProductionGate`, `heldOutGate`, `paretoSignificanceGate`, `neutralizationGate`, `composeGate`); the deployment-outcome store; storage; and the five core types `Scenario` / `Dispatch` / `JudgeConfig` / `SurfaceProposer` / `Gate`. |
 | `…/hosted` | `createHostedClient` / `hostedClientFromEnv` + the wire types to ship eval-run events + trace spans to a hosted orchestrator (ours or your own implementation of the spec) |
-| `…/primeintellect` | Export validated `RunRecord`s plus scenario prompts into a PrimeIntellect Verifiers package: Python `load_environment()`, `data/dataset.jsonl`, `data/run_records.jsonl`, and export manifest |
 | `…/adapters/otel` | `createOtelBridge` — forwards OpenTelemetry-shape spans into the hosted-tier ingest, no `@opentelemetry/*` dependency |
 | `…/adapters/langchain` | Wrap any LangChain `Runnable` as a `Dispatch` (or `JudgeConfig`), no `@langchain/core` peer dep |
 | `…/adapters/http` | `httpDispatch` + `runDispatchServer` — run a campaign's worker on another machine (multi-region, remote worker execution) |

package/dist/benchmarks/index.d.ts

@@ -356,6 +356,10 @@ interface Scenario {
     kind: string;
     tags?: string[];
 }
+/** Redacted identity of a complete scenario payload retained in campaign results. */
+interface CampaignScenarioIdentity extends Pick<Scenario, 'id' | 'kind'> {
+    scenarioDigest: `sha256:${string}`;
+}
 /** Context handed to every dispatch invocation. Scoped — every
  *  trace/span carries the cellId, every artifact write lands under the cell's
  *  artifact root, the cost meter accumulates per cell. */
@@ -564,7 +568,11 @@ interface CampaignAggregates {
 interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scenario> {
     /** sha256(scenarios, judges, dispatch source ref, optimizer config, seed). Stable identity for reruns. */
     manifestHash: string;
+    /** Canonical identity of the exact scenario payloads and replicate count. */
+    splitDigest: `sha256:${string}`;
     seed: number;
+    /** Replicates designed for every scenario in this campaign. */
+    reps: number;
     startedAt: string;
     endedAt: string;
     durationMs: number;
@@ -578,11 +586,9 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
     prUrl?: string;
     runDir: string;
     artifactsByPath: Record<string, string>;
-    /** Substrate strips the input scenarios to id+kind for the result manifest;
-     *  consumers needing full payload look it up via the original input. The
-     *  type parameter `TScenario` is propagated for downstream consumers that
-     *  want narrowed types when extending `CampaignResult`. */
-    scenarios: Array<Pick<TScenario, 'id' | 'kind'>>;
+    /** Redacted identities that let consumers verify the exact scenario payloads
+     *  without retaining customer task content in the result. */
+    scenarios: Array<CampaignScenarioIdentity & Pick<TScenario, 'id' | 'kind'>>;
 }
 
 /**

package/dist/benchmarks/index.js

@@ -16,10 +16,10 @@ import {
   routing_exports,
   runBenchmarkAdapter,
   summarizeBenchmarkCampaign
-} from "../chunk-D7AEXSM5.js";
-import "../chunk-XMBOU5W7.js";
-import "../chunk-PSXJ32SR.js";
-import "../chunk-PAHNGS65.js";
+} from "../chunk-PICTDURQ.js";
+import "../chunk-JN2FCO5W.js";
+import "../chunk-32BZXMSO.js";
+import "../chunk-XDIRG3TO.js";
 import "../chunk-3YYRZDON.js";
 import "../chunk-S5TT5R3L.js";
 import "../chunk-ARU2PZFM.js";

package/dist/campaign/index.d.ts

@@ -1455,6 +1455,10 @@ interface Scenario {
     kind: string;
     tags?: string[];
 }
+/** Redacted identity of a complete scenario payload retained in campaign results. */
+interface CampaignScenarioIdentity extends Pick<Scenario, 'id' | 'kind'> {
+    scenarioDigest: `sha256:${string}`;
+}
 /** Context handed to every dispatch invocation. Scoped — every
  *  trace/span carries the cellId, every artifact write lands under the cell's
  *  artifact root, the cost meter accumulates per cell. */
@@ -2037,7 +2041,11 @@ interface CampaignAggregates {
 interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scenario> {
     /** sha256(scenarios, judges, dispatch source ref, optimizer config, seed). Stable identity for reruns. */
     manifestHash: string;
+    /** Canonical identity of the exact scenario payloads and replicate count. */
+    splitDigest: `sha256:${string}`;
     seed: number;
+    /** Replicates designed for every scenario in this campaign. */
+    reps: number;
     startedAt: string;
     endedAt: string;
     durationMs: number;
@@ -2051,11 +2059,9 @@ interface CampaignResult<TArtifact = unknown, TScenario extends Scenario = Scena
     prUrl?: string;
     runDir: string;
     artifactsByPath: Record<string, string>;
-    /** Substrate strips the input scenarios to id+kind for the result manifest;
-     *  consumers needing full payload look it up via the original input. The
-     *  type parameter `TScenario` is propagated for downstream consumers that
-     *  want narrowed types when extending `CampaignResult`. */
-    scenarios: Array<Pick<TScenario, 'id' | 'kind'>>;
+    /** Redacted identities that let consumers verify the exact scenario payloads
+     *  without retaining customer task content in the result. */
+    scenarios: Array<CampaignScenarioIdentity & Pick<TScenario, 'id' | 'kind'>>;
 }
 
 /**
@@ -2384,6 +2390,17 @@ interface OpenAutoPrResult {
  */
 declare function openAutoPr<TArtifact, TScenario extends Scenario>(options: OpenAutoPrOptions<TArtifact, TScenario>): OpenAutoPrResult;
 
+/** Reject campaign designs whose denominator cannot be identified exactly. */
+declare function assertCampaignDesign<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): void;
+/** Redacted but independently verifiable identity of one complete scenario. */
+declare function campaignScenarioIdentity<TScenario extends Scenario>(scenario: TScenario): CampaignScenarioIdentity & Pick<TScenario, 'id' | 'kind'>;
+/** Canonical split identity reconstructed from redacted scenario identities. */
+declare function campaignSplitDigestFromIdentities(scenarios: readonly CampaignScenarioIdentity[], reps: number): `sha256:${string}`;
+/** Canonical identity of the exact scenario payloads and replicate count. */
+declare function campaignSplitDigest<TScenario extends Scenario>(scenarios: readonly TScenario[], reps: number): `sha256:${string}`;
+/** Refuse a campaign whose retained task identities contradict its split digest. */
+declare function assertCampaignSplitIdentity(scenarios: readonly CampaignScenarioIdentity[], reps: number, splitDigest: string): void;
+
 interface ContinuousAgreement {
     /** Cohen's κ_w with quadratic weights, computed on raw [0,1] scores. */
     weightedKappa: number;
@@ -2977,6 +2994,7 @@ interface CampaignRunPlanCell {
 }
 interface CampaignRunPlan {
     manifestHash: string;
+    splitDigest: `sha256:${string}`;
     totalCells: number;
     cellsCached: number;
     cellsToRun: number;
@@ -4086,17 +4104,6 @@ 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;
-
 /**
  * `runOptimization` — the improvement loop body. Runs N generations: the
  * `SurfaceProposer` proposes K candidate surfaces per generation, each
@@ -4221,8 +4228,8 @@ declare function runOptimization<TScenario extends Scenario, TArtifact>(opts: Ru
  *
  * Hard-refuses unsafe configurations:
  *   - `tracing: 'off'` when a proposer is wired (improvement is unattributable)
- *   - `autoOnPromote: 'config'` — DEFERRED to Pass B; v0.40 only ships
- *     `'pr'` and `'none'`.
+ *   - `autoOnPromote: 'config'` — live mutation is unsupported without
+ *     isolated deployment, rollback, and independent validation.
  */
 
 type RunImprovementLoopOptions<TScenario extends Scenario, TArtifact> = RunOptimizationOptions<TScenario, TArtifact> & {
@@ -4236,15 +4243,11 @@ type RunImprovementLoopOptions<TScenario extends Scenario, TArtifact> = RunOptim
     /** What to do when the gate ships:
      *   - `'pr'`: open a PR via `openAutoPr`
      *   - `'none'`: just report — caller decides what to do with the winner
-     *  v0.40 does NOT support `'config'` (live-runtime self-mutation) —
-     *  deferred to Pass B behind safety stack. */
+     *  Live-runtime self-mutation is intentionally unsupported. */
     autoOnPromote: 'pr' | 'none';
     /** GH owner / repo for the auto-PR. Required when autoOnPromote === 'pr'. */
     ghOwner?: string;
     ghRepo?: string;
-    /** Optional render override — substrate writes a diff-shaped surface; pass
-     *  a function to format the promoted surface differently. */
-    renderPromotedDiff?: (winnerSurface: MutableSurface, baselineSurface: MutableSurface) => string;
     /** Placebo control. When supplied AND the winner differs from baseline, the
      *  loop scores a THIRD holdout arm: the winner surface with its content
      *  footprint-matched-blanked by this function (typically via `neutralizeText`).
@@ -4257,6 +4260,8 @@ type RunImprovementLoopOptions<TScenario extends Scenario, TArtifact> = RunOptim
 interface RunImprovementLoopResult<TArtifact, TScenario extends Scenario> extends RunOptimizationResult<TArtifact, TScenario> {
     baselineOnHoldout: CampaignResult<TArtifact, TScenario>;
     winnerOnHoldout: CampaignResult<TArtifact, TScenario>;
+    neutralizedOnHoldout?: CampaignResult<TArtifact, TScenario>;
+    neutralizedSurface?: MutableSurface;
     gateResult: Awaited<ReturnType<Gate<TArtifact, TScenario>['decide']>>;
     /** Unified baseline→winner surface diff. Computed UNCONDITIONALLY (not only
      *  when `autoOnPromote === 'pr'`) so the diff that the gate decided on is
@@ -4269,10 +4274,6 @@ interface RunImprovementLoopResult<TArtifact, TScenario extends Scenario> extend
  * Gated-promotion shell over `runOptimization`: scores the winner against the baseline on a holdout set, runs the release gate, and optionally opens a PR.
  */
 declare function runImprovementLoop<TScenario extends Scenario, TArtifact>(opts: RunImprovementLoopOptions<TScenario, TArtifact>): Promise<RunImprovementLoopResult<TArtifact, TScenario>>;
-/**
- * Default surface diff renderer: produces a unified baseline/winner text diff for prompt surfaces or a worktree-ref summary for code surfaces.
- */
-declare function defaultRenderDiff(winnerSurface: MutableSurface, baselineSurface: MutableSurface): string;
 
 /**
  * `compareProposers` — a head-to-head lift benchmark across surface proposers
@@ -6653,9 +6654,8 @@ interface HostedClient {
  *      reads). The hosted `/v1/ingest/traces` endpoint receives the FULL loop,
  *      not just the `cost.*` spans `runCampaign` already emits per cell.
  *
- * The record is built from the substrate's own loop result + the per-call
- * `RunRecord`s the worker emitted — no new measurement, no recomputation that
- * could drift from what the gate actually saw.
+ * The record is built from the loop result and its settled cost receipts — no
+ * second usage collector can contradict what the measured cells recorded.
  */
 
 interface LoopProvenanceCandidate {
@@ -6665,6 +6665,8 @@ interface LoopProvenanceCandidate {
     surfaceHash: string;
     /** Full sha256 content hash — byte-identical-verifiable. */
     contentHash: string;
+    /** Exact scored rows that produced this candidate's search result. */
+    campaignDigest: `sha256:${string}`;
     /** Proposer label, when the proposer returned a `ProposedCandidate`. */
     label?: string;
     /** Proposer rationale — the "because Z". When the proposer returned a bare
@@ -6698,13 +6700,33 @@ interface LoopProvenanceBackend {
     totalOutputTokens: number;
     totalCostUsd: number;
 }
+interface LoopProvenanceEvidence {
+    search: {
+        splitDigest: `sha256:${string}`;
+        baselineCampaignDigest: `sha256:${string}`;
+    };
+    holdout: {
+        splitDigest: `sha256:${string}`;
+        baselineCampaignDigest: `sha256:${string}`;
+        winnerCampaignDigest: `sha256:${string}`;
+        neutralized?: {
+            contentHash: `sha256:${string}`;
+            campaignDigest: `sha256:${string}`;
+            composite: number;
+            lift: number;
+        };
+    };
+    costReceiptsDigest: `sha256:${string}`;
+}
 /**
  * The durable provenance record. Aligns to the hosted `EvalRunEvent` path but
  * ADDS the rationale + the explicit baseline→candidate diff (both omitted from
  * the bare hosted event) + backend provenance.
  */
 interface LoopProvenanceRecord {
-    schema: 'tangle.loop-provenance.v3';
+    schema: 'tangle.loop-provenance';
+    /** SHA-256 over the canonical record with this field omitted. */
+    recordDigest: `sha256:${string}`;
     runId: string;
     runDir: string;
     timestamp: string;
@@ -6718,6 +6740,8 @@ interface LoopProvenanceRecord {
     diff: string;
     /** Every candidate across every generation, with its rationale and structured cause. */
     candidates: LoopProvenanceCandidate[];
+    /** Exact campaign, split, surface, and receipt identities behind every summary. */
+    evidence: LoopProvenanceEvidence;
     /** Baseline composite on the search split that generated the candidates. */
     baselineSearchComposite: number;
     /** The gate verdict — decision + reasons + contributing gates + delta. */
@@ -6728,6 +6752,7 @@ interface LoopProvenanceRecord {
         contributingGates: Array<{
             name: string;
             passed: boolean;
+            detail: unknown;
         }>;
     };
     /** baseline-on-holdout composite mean. */
@@ -6749,9 +6774,8 @@ interface BuildLoopProvenanceArgs<TArtifact, TScenario extends Scenario> {
     winnerSurface: MutableSurface;
     winnerLabel?: string;
     winnerRationale?: string;
-    diff: string;
-    /** Baseline composite on the search split, distinct from holdout scoring. */
-    baselineSearchComposite: number;
+    /** Exact baseline campaign on the search split. */
+    baselineSearchCampaign: CampaignResult<TArtifact, TScenario>;
     /** Per-generation candidate records straight off the loop result. */
     generations: Array<{
         generationIndex: number;
@@ -6762,18 +6786,37 @@ interface BuildLoopProvenanceArgs<TArtifact, TScenario extends Scenario> {
         surfaces: Array<{
             surfaceHash: string;
             surface: MutableSurface;
+            campaign: CampaignResult<TArtifact, TScenario>;
         }>;
     }>;
     gate: GateResult;
     baselineOnHoldout: CampaignResult<TArtifact, TScenario>;
     winnerOnHoldout: CampaignResult<TArtifact, TScenario>;
-    /** Worker call records — the source for backend provenance. */
-    workerRecords: ReadonlyArray<RunRecord>;
+    neutralizedSurface?: MutableSurface;
+    neutralizedOnHoldout?: CampaignResult<TArtifact, TScenario>;
+    /** Settled run-wide receipts — agent calls are the source for backend provenance. */
+    costReceipts: ReadonlyArray<CostReceipt>;
+    totalCostUsd: number;
+    totalDurationMs: number;
+}
+interface LoopProvenanceArgsFromResult<TArtifact, TScenario extends Scenario> {
+    runId: string;
+    runDir: string;
+    timestamp: string;
+    baselineSurface: MutableSurface;
+    result: RunImprovementLoopResult<TArtifact, TScenario>;
+    costReceipts: ReadonlyArray<CostReceipt>;
     totalCostUsd: number;
     totalDurationMs: number;
 }
+/** One translation from a completed improvement loop into durable evidence. */
+declare function loopProvenanceArgsFromResult<TArtifact, TScenario extends Scenario>(input: LoopProvenanceArgsFromResult<TArtifact, TScenario>): BuildLoopProvenanceArgs<TArtifact, TScenario>;
 /** Build the durable provenance record from a completed loop result. */
 declare function buildLoopProvenanceRecord<TArtifact, TScenario extends Scenario>(args: BuildLoopProvenanceArgs<TArtifact, TScenario>): LoopProvenanceRecord;
+/** Digest the exact campaign fields that can affect a measured comparison. */
+declare function campaignMeasurementDigest<TArtifact, TScenario extends Scenario>(campaign: CampaignResult<TArtifact, TScenario>): `sha256:${string}`;
+/** Recompute and validate the self-addressed durable record. */
+declare function verifyLoopProvenanceRecord(record: LoopProvenanceRecord): LoopProvenanceRecord;
 /**
  * Build the loop's OTLP-ingestable spans from a provenance record. One root
  * span per loop (`tangle.runId`), one span per generation, one span per
@@ -7281,6 +7324,19 @@ interface SingleRunLock {
  */
 declare function acquireSingleRunLock(opts: SingleRunLockOptions): SingleRunLock;
 
+/** Validate the immutable identity shape; the owning executor verifies the Git objects and patch. */
+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;
+/** Canonical customer-visible description of the exact before/after surfaces. */
+declare function renderSurfaceDiff(winnerSurface: MutableSurface, baselineSurface: MutableSurface): string;
+
 /**
  * Transient-transport-failure classification for dispatch retry policies.
  *
@@ -7395,4 +7451,4 @@ declare function verifyCodeSurface(surface: CodeSurface, worktreeDir?: string):
  *  identity against the checkout at `worktreeRef`. */
 declare function resolveWorktreePath(surface: CodeSurface, worktreeDir?: string): string;
 
-export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type ApplySkillPatchResult, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CodeSurface, type CodeSurfaceVerification, type CompareProposersOptions, type CompositeProposerOptions, type CostLedgerHandle, 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_POLICY_EDIT_HISTORY_LIMITS, type DefaultProductionGateOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type EvolutionaryProposerOptions, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoEntryConfig, type FapoFailureCluster, type FapoOptimizationLevel, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateContext, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaProposerConstraints, type GepaProposerOptions, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPolicyEditTargetSurface, type JsonPrimitive, type JsonValue, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, type LlmJudgeDimension, type LlmJudgeOptions, type LlmPolicyEditProposerOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceRecord, type MemoryCurationProposerOptions, type MutableSurface, type Mutator, type NeutralizationGateOptions, type ObjectiveSource, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationProposer, type OptimizerConfig, type OptimizerEntryConfig, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, type ParetoParent, type ParetoSignificanceGateOptions, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditAuthorScenarioRow, type PolicyEditCandidateRecord, type PolicyEditCandidateSummary, type PolicyEditFindingInput, type PolicyEditFindingSource, type PolicyEditHistoryCandidateContext, type PolicyEditHistoryGenerationContext, type PolicyEditHistoryProjectionOptions, type PolicyEditObjective, type PolicyEditOutcomeContext, type PolicyEditProposerOptions, type PowerPreflight, type PowerPreflightOptions, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposeContext, type ProposePatchesArgs, type ProposedCandidate, type ProposerComparison, type ProposerEntry, type ProposerPairwise, type ProposerScore, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAccountingAudit, type SearchArtifactRef, type SearchAttemptAccounting, type SearchCandidateDecidedEvent, type SearchCandidateLineage, type SearchCandidateRegisteredEvent, type SearchCandidateSlot, type SearchCandidateSlotClosedEvent, type SearchCandidateSurface, type SearchCompletedEvent, type SearchCostAccounting, type SearchFailureReason, type SearchLedger, type SearchLedgerAppendResult, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerReplay, type SearchModelIdentity, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchPlan, type SearchPlannedEvent, type SearchPlannedOperation, type SearchPlannedTask, type SearchSourceRef, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTaskAttemptedEvent, type SearchTaskOutcome, type SearchTokenAccounting, type SelectPolicyEditAuthorRowsOptions, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SerializedJsonBudget, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, type SurfaceProposer, type SurfaceScore, type TraceAnalystProposerOptions, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, analyzeCrossSurfaceInteractions, applySkillPatch, assertCodeSurfaceIdentity, assertPolicyEditAuthorContextBudget, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, callbackGovernor, campaignBreakdown, campaignMeanComposite, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareProposers, composeGate, compositeProposer, countSentenceEdits, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, defaultRenderDiff, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, evolutionaryProposer, extractFapoAttributionSignals, extractH2Sections, failureModeRecallJudge, fapoEscalationEntry, fapoProposer, fsCampaignStorage, fsLineageStore, gepaParetoEntry, gepaProposer, gepaReflectionEntry, gitWorktreeAdapter, haloProposer, heldOutGate, heldoutSignificance, heuristicGovernor, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, lineageNodeId, llmJudge, llmPolicyEditProposer, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceSpans, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, pairHoldout, parameterSweepProposer, paretoPolicy, paretoSignificanceGate, parseSkillPatchResponse, patchEditCount, planCampaignRun, planEvalFixtureRun, policyEditProposer, powerPreflight, projectPolicyEditHistory, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runLineage, runLineageLoop, runOptimization, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, selectPolicyEditAuthorRows, sequentialDecide, sequentialPairedGate, skillOptEntry, skillOptProposer, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard, validatePolicyEditCandidateRecord, validateSearchLedgerEvent, verifyCodeSurface };
+export { type AcceptedEdit, type AceProposerOptions, type AnalystArtifact, type AnalystScenario, type AnalyzeCrossSurfaceInteractionsInput, type ApplySkillPatchResult, type AxisEvidence, type AxisVerdict, type BuildAnalystSurfaceDispatchOptions, type BuildEvidenceVectorOptions, type BuildLoopProvenanceArgs, type CampaignAggregates, type CampaignArtifactWriter, type CampaignBreakdown, type CampaignCellResult, type CampaignCostMeter, type CampaignResult, type CampaignRunPlan, type CampaignRunPlanCell, type CampaignScenarioIdentity, type CampaignStorage, type CampaignTokenUsage, type CampaignTraceWriter, type CodeSurface, type CodeSurfaceVerification, type CompareProposersOptions, type CompositeProposerOptions, type CostLedgerHandle, 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_POLICY_EDIT_HISTORY_LIMITS, type DefaultProductionGateOptions, type DimensionRegression, type DiscriminationScore, type DispatchContext, type DispatchFn, type EmitLoopProvenanceArgs, type EmitLoopProvenanceResult, type EvalFixture, type EvalFixtureFile, type EvalFixtureLoadOptions, type EvalFixtureRunPlan, type EvalFixtureScenario, type EvalFixtureValidationMode, type EvidenceVector, type EvolutionaryProposerOptions, type FailureModeRecallJudgeOptions, type FapoAttributionSignals, type FapoEntryConfig, type FapoFailureCluster, type FapoOptimizationLevel, type FapoProposerOptions, type FapoReviewInput, type FapoReviewIssue, type FapoReviewResult, type FapoScopeContract, FileSearchLedger, FsLabeledScenarioStore, type FsLabeledScenarioStoreOptions, type Gate, type GateContext, type GateDecision, type GateResult, type GenerationCandidate, type GenerationRecord, type GepaProposerConstraints, type GepaProposerOptions, type GitWorktreeAdapterOptions, type Governor, type GovernorContext, type GovernorOp, type HaloProposerOptions, type HeldOutGateOptions, type HeldoutSignificance, type HeldoutSignificanceOptions, type HeuristicGovernorOptions, type JsonPolicyEditTargetSurface, type JsonPrimitive, type JsonValue, type JudgeAggregate, type JudgeConfig, type JudgeDimension, type JudgeScore, type LabelTrust, type LabeledScenarioRecord, type LabeledScenarioSampleArgs, type LabeledScenarioSource, type LabeledScenarioStore, LabeledScenarioStoreError, type LabeledScenarioWrite, Lineage, type LineageEdge, type LineageGraph, type LineageNode, type LineageNodeInput, type LineageStore, type LlmJudgeDimension, type LlmJudgeOptions, type LlmPolicyEditProposerOptions, type LoadEvalFixtureScenariosOptions, type LoopProvenanceArgsFromResult, type LoopProvenanceBackend, type LoopProvenanceCandidate, type LoopProvenanceEvidence, type LoopProvenanceRecord, type MemoryCurationProposerOptions, type MutableSurface, type Mutator, type NeutralizationGateOptions, type ObjectiveSource, type OpenAutoPrOptions, type OpenAutoPrResult, type OpenSearchLedgerOptions, type OptimizationProposer, type OptimizerConfig, type OptimizerEntryConfig, POLICY_EDIT_CANDIDATE_RECORD_SCHEMA, type PairedHoldout, type ParameterCandidate, type ParameterChange, type ParameterSweepProposerOptions, type ParetoParent, type ParetoSignificanceGateOptions, type PlanCampaignRunOptions, type PlanEvalFixtureRunOptions, type PlaybackContext, type PlaybackDriver, type PlaybackStep, type PolicyEditAuthorScenarioRow, type PolicyEditCandidateRecord, type PolicyEditCandidateSummary, type PolicyEditFindingInput, type PolicyEditFindingSource, type PolicyEditHistoryCandidateContext, type PolicyEditHistoryGenerationContext, type PolicyEditHistoryProjectionOptions, type PolicyEditObjective, type PolicyEditOutcomeContext, type PolicyEditProposerOptions, type PowerPreflight, type PowerPreflightOptions, type ProfileDispatchFn, ProfileMatrixError, type ProfileSummary, type PromotionObjective, type PromotionPolicy, type ProposeContext, type ProposePatchesArgs, type ProposedCandidate, type ProposerComparison, type ProposerEntry, type ProposerPairwise, type ProposerScore, type RedactionStatus, type ReferenceEquivalenceJudgeOptions, type ReferenceEquivalenceScenario, type RejectedEdit, type RolloutArgumentDiff, type RolloutArgumentDiffOptions, type RolloutCall, type RunCampaignOptions, type RunEvalOptions, type RunImprovementLoopOptions, type RunImprovementLoopResult, type RunLineageLoopOptions, type RunLineageLoopResult, type RunLineageLoopSeed, type RunLineageOptions, type RunLineageResult, type RunLineageSeed, type RunLineageStepResult, type RunOptimizationOptions, type RunOptimizationResult, type RunProfileMatrixOptions, type RunProfileMatrixResult, type RunSkillOptOptions, type RunSkillOptResult, SEARCH_LEDGER_SCHEMA, type Scenario, type ScenarioAggregate, type ScenarioRollup, type ScenarioSignal, type ScoreboardRenderOptions, type ScoreboardRow, type ScoreboardSummary, type ScoredRollout, type ScoredSurfaceOutcome, type SearchAccountingAudit, type SearchArtifactRef, type SearchAttemptAccounting, type SearchCandidateDecidedEvent, type SearchCandidateLineage, type SearchCandidateRegisteredEvent, type SearchCandidateSlot, type SearchCandidateSlotClosedEvent, type SearchCandidateSurface, type SearchCompletedEvent, type SearchCostAccounting, type SearchFailureReason, type SearchLedger, type SearchLedgerAppendResult, SearchLedgerConflictError, type SearchLedgerEntry, SearchLedgerError, type SearchLedgerEvent, type SearchLedgerHash, SearchLedgerIntegrityError, type SearchLedgerReplay, type SearchModelIdentity, type SearchOperationKind, type SearchOperationRecordedEvent, type SearchPlan, type SearchPlannedEvent, type SearchPlannedOperation, type SearchPlannedTask, type SearchSourceRef, type SearchSurfaceEffect, type SearchSurfaceEvidence, type SearchSurfaceKind, type SearchTaskAttemptedEvent, type SearchTaskOutcome, type SearchTokenAccounting, type SelectPolicyEditAuthorRowsOptions, type SequentialDecideFn, type SequentialDecideOptions, type SequentialDecision, type SequentialObservation, type SequentialPairedGate, type SequentialPairedGateOptions, type SerializedJsonBudget, type SessionScript, type SingleRunLock, type SingleRunLockOptions, type SkillOptEpochRecord, type SkillOptEvidence, type SkillOptProposer, type SkillOptProposerOptions, type SkillPatch, type SkillPatchOp, SkillPatchParseError, type SkillPatchRejection, type SurfaceProposer, type SurfaceScore, type TraceAnalystProposerOptions, type TraceSpan, type TransientFailureOptions, type UngroundedLiteralReport, type UserStory, type UserStoryVerdict, type Worktree, type WorktreeAdapter, WorktreeAdapterError, aceProposer, acquireSingleRunLock, analyzeCrossSurfaceInteractions, applySkillPatch, assertCampaignDesign, assertCampaignSplitIdentity, assertCodeSurfaceIdentity, assertPolicyEditAuthorContextBudget, buildAnalystSurfaceDispatch, buildEvidenceVector, buildLoopProvenanceRecord, callbackGovernor, campaignBreakdown, campaignMeanComposite, campaignMeasurementDigest, campaignScenarioIdentity, campaignSplitDigest, campaignSplitDigestFromIdentities, classifyUngroundedLiterals, codeSurfaceIdentityMaterial, compareProposers, composeGate, compositeProposer, countSentenceEdits, createReferenceEquivalenceJudge, createRunCostLedger, defaultProductionGate, detectScale, dimensionRegressions, discoverEvalFixtures, emitLoopProvenance, evolutionaryProposer, extractFapoAttributionSignals, extractH2Sections, failureModeRecallJudge, fapoEscalationEntry, fapoProposer, fsCampaignStorage, fsLineageStore, gepaParetoEntry, gepaProposer, gepaReflectionEntry, gitWorktreeAdapter, haloProposer, heldOutGate, heldoutSignificance, heuristicGovernor, inMemoryCampaignStorage, isProposedCandidate, isTransientTransportFailure, labelTrustRank, lineageNodeId, llmJudge, llmPolicyEditProposer, loadEvalFixture, loadEvalFixtureScenarios, loopProvenanceArgsFromResult, loopProvenanceSpans, makePlaybackDispatch, memLineageStore, memoryCurationProposer, neutralizationGate, neutralizeText, openAutoPr, openSearchLedger, pairHoldout, parameterSweepProposer, paretoPolicy, paretoSignificanceGate, parseSkillPatchResponse, patchEditCount, planCampaignRun, planEvalFixtureRun, policyEditProposer, powerPreflight, projectPolicyEditHistory, provenanceRecordPath, provenanceSpansPath, renderScoreboardMarkdown, renderSurfaceDiff, resolveRunDir, resolveWorktreePath, rolloutArgumentDiff, runCampaign, runEval, runImprovementLoop, runLineage, runLineageLoop, runOptimization, runProfileMatrix, runSkillOpt, scoreDiscrimination, scoreUserStory, scoreboardSummary, selectDiscriminative, selectPolicyEditAuthorRows, sequentialDecide, sequentialPairedGate, skillOptEntry, skillOptProposer, surfaceContentHash, surfaceHash, tangleTracesRoot, traceAnalystProposer, userStoryScoreboard, validatePolicyEditCandidateRecord, validateSearchLedgerEvent, verifyCodeSurface, verifyLoopProvenanceRecord };

package/dist/campaign/index.js

@@ -66,19 +66,19 @@ import {
   userStoryScoreboard,
   validateSearchLedgerEvent,
   verifyCodeSurface
-} from "../chunk-XMBOU5W7.js";
+} from "../chunk-JN2FCO5W.js";
 import {
   assertCodeSurfaceIdentity,
   buildEvidenceVector,
   buildLoopProvenanceRecord,
   campaignBreakdown,
   campaignMeanComposite,
+  campaignMeasurementDigest,
   codeSurfaceIdentityMaterial,
   composeGate,
   countSentenceEdits,
   createReferenceEquivalenceJudge,
   defaultProductionGate,
-  defaultRenderDiff,
   detectScale,
   dimensionRegressions,
   emitLoopProvenance,
@@ -90,6 +90,7 @@ import {
   isProposedCandidate,
   labelTrustRank,
   llmJudge,
+  loopProvenanceArgsFromResult,
   loopProvenanceSpans,
   openAutoPr,
   pairHoldout,
@@ -98,16 +99,23 @@ import {
   powerPreflight,
   provenanceRecordPath,
   provenanceSpansPath,
+  renderSurfaceDiff,
   runEval,
   runImprovementLoop,
   runOptimization,
   surfaceContentHash,
-  surfaceHash
-} from "../chunk-PSXJ32SR.js";
+  surfaceHash,
+  verifyLoopProvenanceRecord
+} from "../chunk-32BZXMSO.js";
 import {
   SearchLedgerConflictError,
   SearchLedgerError,
   SearchLedgerIntegrityError,
+  assertCampaignDesign,
+  assertCampaignSplitIdentity,
+  campaignScenarioIdentity,
+  campaignSplitDigest,
+  campaignSplitDigestFromIdentities,
   createRunCostLedger,
   fsCampaignStorage,
   inMemoryCampaignStorage,
@@ -115,7 +123,7 @@ import {
   resolveRunDir,
   runCampaign,
   tangleTracesRoot
-} from "../chunk-PAHNGS65.js";
+} from "../chunk-XDIRG3TO.js";
 import "../chunk-3YYRZDON.js";
 import {
   POLICY_EDIT_CANDIDATE_RECORD_SCHEMA,
@@ -155,6 +163,8 @@ export {
   acquireSingleRunLock,
   analyzeCrossSurfaceInteractions,
   applySkillPatch,
+  assertCampaignDesign,
+  assertCampaignSplitIdentity,
   assertCodeSurfaceIdentity,
   assertPolicyEditAuthorContextBudget,
   buildAnalystSurfaceDispatch,
@@ -163,6 +173,10 @@ export {
   callbackGovernor,
   campaignBreakdown,
   campaignMeanComposite,
+  campaignMeasurementDigest,
+  campaignScenarioIdentity,
+  campaignSplitDigest,
+  campaignSplitDigestFromIdentities,
   classifyUngroundedLiterals,
   codeSurfaceIdentityMaterial,
   compareProposers,
@@ -172,7 +186,6 @@ export {
   createReferenceEquivalenceJudge,
   createRunCostLedger,
   defaultProductionGate,
-  defaultRenderDiff,
   detectScale,
   dimensionRegressions,
   discoverEvalFixtures,
@@ -202,6 +215,7 @@ export {
   llmPolicyEditProposer,
   loadEvalFixture,
   loadEvalFixtureScenarios,
+  loopProvenanceArgsFromResult,
   loopProvenanceSpans,
   makePlaybackDispatch,
   memLineageStore,
@@ -224,6 +238,7 @@ export {
   provenanceRecordPath,
   provenanceSpansPath,
   renderScoreboardMarkdown,
+  renderSurfaceDiff,
   resolveRunDir,
   resolveWorktreePath,
   rolloutArgumentDiff,
@@ -251,6 +266,7 @@ export {
   userStoryScoreboard,
   validatePolicyEditCandidateRecord,
   validateSearchLedgerEvent,
-  verifyCodeSurface
+  verifyCodeSurface,
+  verifyLoopProvenanceRecord
 };
 //# sourceMappingURL=index.js.map