@tangle-network/agent-eval 0.95.0 → 0.95.1

package/dist/rl/verifiable-reward.d.ts

@@ -1,124 +0,0 @@
-/**
- * Verifiable reward channel.
- *
- * For RL on coding / math / theorem-proving / structured-output tasks, the
- * reward signal is *decidable* — a test passes or fails, a proof checks or
- * doesn't, an output validates against a schema or doesn't. These rewards
- * are dramatically more useful for RL training than LLM-judge scores
- * because they don't drift, can't be Goodhart-gamed by the policy in the
- * same way, and don't require a separate calibration loop.
- *
- * The `MultiLayerVerifier` already produces this signal — it just doesn't
- * surface it in a shape that's clean enough for RL training. This module
- * wraps the verifier output so consumers can:
- *
- *   1. Extract a clean `VerifiableReward` from a `VerificationReport`
- *   2. Distinguish *deterministic* rewards (compile, test, schema) from
- *      *probabilistic* rewards (judge) so they can be weighted differently
- *      in the RL training step
- *   3. Filter `RunRecord[]` to only those with a verifiable reward,
- *      producing the clean training set that DeepSeek-R1-style GRPO and
- *      AlphaProof-style search both depend on
- *
- * Why this matters: every credible 2025-2026 frontier RL result on coding
- * agents leans on verifiable reward (DeepSeek-R1 GRPO on test pass-rate,
- * o-series RL on math/code, AlphaProof on Lean kernel checking). Mixing
- * judge scores into the reward signal poisons the gradient. This module
- * is the seam.
- */
-import type { VerificationReport } from '../multi-layer-verifier';
-import type { RunRecord } from '../run-record';
-export type VerifiableRewardSource = 'compile' | 'test' | 'schema' | 'sandbox' | 'judge' | 'composite';
-export interface VerifiableReward {
-    /** Scalar in [0, 1]. The RL training signal. */
-    value: number;
-    /** What produced the reward — different sources have different determinism. */
-    source: VerifiableRewardSource;
-    /**
-     * Determinism class. `'deterministic'` rewards are repeatable byte-for-byte
-     * given the same inputs (compile, test, schema validation, sandbox exit code).
-     * `'probabilistic'` rewards depend on a stochastic component (LLM judge).
-     * Mixing these in the same training batch without separation is a known
-     * footgun in production RLHF pipelines.
-     */
-    determinism: 'deterministic' | 'probabilistic';
-    /**
-     * Confidence in the reward value. For deterministic sources this is 1.0
-     * (the bit either flipped or didn't). For judge sources this is the
-     * judge-reported confidence or — when missing — a calibrated prior.
-     */
-    confidence: number;
-    /** The layer / judge id that produced the signal, for provenance. */
-    origin: string;
-    /**
-     * Per-source contribution to `value`, keyed by layer/judge id. Single-source
-     * rewards carry one entry (`{ [origin]: value }`); composite rewards carry
-     * every contributing layer's score — the anti-scalar-collapse surface RL
-     * consumers weight per-source instead of trusting one blended number.
-     */
-    components: Record<string, number>;
-    /**
-     * @deprecated Read `components` for per-source reward values. Kept for
-     * published-API compatibility: single-source rewards carry the layer's
-     * diagnostics here (e.g. `{ tests_passed: 7 }`); composite rewards carry
-     * the same per-layer scores `components` now holds.
-     */
-    breakdown?: Record<string, number>;
-}
-export interface VerifiableRewardExtractionOptions {
-    /**
-     * Which layers count as deterministic-reward sources. The verifier doesn't
-     * tag layers as "this is verifiable"; the caller declares it via this list
-     * (or via the layer name → source mapping). Default treats common names
-     * (`install`, `typecheck`, `build`, `lint`, `test`, `compile`, `schema`,
-     * `sandbox`) as deterministic.
-     */
-    deterministicLayers?: string[];
-    /**
-     * Map layer name → reward source. Defaults to a sensible string-match.
-     */
-    sourceFor?: (layerName: string) => VerifiableRewardSource;
-    /**
-     * Whether to fall back to a probabilistic (judge) reward when no
-     * deterministic layer produced a numeric score. Default `true`. Set to
-     * `false` for "deterministic-only" training pipelines that should
-     * discard runs without a verifiable signal.
-     */
-    fallbackToJudge?: boolean;
-    /**
-     * Default confidence for probabilistic (judge) rewards when the judge
-     * doesn't report one. Default `0.7`.
-     */
-    judgeConfidenceFloor?: number;
-}
-/**
- * Extract a `VerifiableReward` from a `VerificationReport`.
- *
- * Strategy: prefer the deterministic layers (in order: test → compile →
- * schema → sandbox), fall back to the judge layer if `fallbackToJudge` is
- * true, return `null` if no signal qualifies. When multiple deterministic
- * layers contribute, return a `'composite'` source with a weighted blend.
- */
-export declare function extractVerifiableReward(report: VerificationReport, opts?: VerifiableRewardExtractionOptions): VerifiableReward | null;
-/**
- * Extract verifiable rewards from `RunRecord[]` produced via the
- * `verificationReportToRunRecord` adapter (which encodes per-layer scores
- * in `outcome.raw['layer.<name>']`). For records that don't carry layer
- * scores, returns `null` for that record.
- *
- * This is the canonical bridge from "campaign-shaped artifacts" to
- * "RL-training-ready reward signals": every record that has a clean
- * verifiable reward becomes a training datum, every record that doesn't
- * gets filtered out (or kept with `'probabilistic'` determinism for
- * separate downstream handling).
- */
-export declare function extractVerifiableRewardsFromRecords(runs: RunRecord[], opts?: VerifiableRewardExtractionOptions): Array<{
-    runId: string;
-    reward: VerifiableReward | null;
-}>;
-/** Filter `RunRecord[]` to those with deterministic verifiable rewards. */
-export declare function filterDeterministicallyRewarded(runs: RunRecord[], opts?: VerifiableRewardExtractionOptions): Array<{
-    run: RunRecord;
-    reward: VerifiableReward;
-}>;
-//# sourceMappingURL=verifiable-reward.d.ts.map

package/dist/rl/verifiable-reward.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"verifiable-reward.d.ts","sourceRoot":"","sources":["../../src/rl/verifiable-reward.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;GA2BG;AAEH,OAAO,KAAK,EAAe,kBAAkB,EAAE,MAAM,yBAAyB,CAAA;AAC9E,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,eAAe,CAAA;AAE9C,MAAM,MAAM,sBAAsB,GAC9B,SAAS,GACT,MAAM,GACN,QAAQ,GACR,SAAS,GACT,OAAO,GACP,WAAW,CAAA;AAEf,MAAM,WAAW,gBAAgB;IAC/B,gDAAgD;IAChD,KAAK,EAAE,MAAM,CAAA;IACb,+EAA+E;IAC/E,MAAM,EAAE,sBAAsB,CAAA;IAC9B;;;;;;OAMG;IACH,WAAW,EAAE,eAAe,GAAG,eAAe,CAAA;IAC9C;;;;OAIG;IACH,UAAU,EAAE,MAAM,CAAA;IAClB,qEAAqE;IACrE,MAAM,EAAE,MAAM,CAAA;IACd;;;;;OAKG;IACH,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAClC;;;;;OAKG;IACH,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,iCAAiC;IAChD;;;;;;OAMG;IACH,mBAAmB,CAAC,EAAE,MAAM,EAAE,CAAA;IAC9B;;OAEG;IACH,SAAS,CAAC,EAAE,CAAC,SAAS,EAAE,MAAM,KAAK,sBAAsB,CAAA;IACzD;;;;;OAKG;IACH,eAAe,CAAC,EAAE,OAAO,CAAA;IACzB;;;OAGG;IACH,oBAAoB,CAAC,EAAE,MAAM,CAAA;CAC9B;AA+BD;;;;;;;GAOG;AACH,wBAAgB,uBAAuB,CACrC,MAAM,EAAE,kBAAkB,EAC1B,IAAI,GAAE,iCAAsC,GAC3C,gBAAgB,GAAG,IAAI,CAmEzB;AAED;;;;;;;;;;;GAWG;AACH,wBAAgB,mCAAmC,CACjD,IAAI,EAAE,SAAS,EAAE,EACjB,IAAI,GAAE,iCAAsC,GAC3C,KAAK,CAAC;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,gBAAgB,GAAG,IAAI,CAAA;CAAE,CAAC,CA0E3D;AAED,2EAA2E;AAC3E,wBAAgB,+BAA+B,CAC7C,IAAI,EAAE,SAAS,EAAE,EACjB,IAAI,GAAE,iCAAsC,GAC3C,KAAK,CAAC;IAAE,GAAG,EAAE,SAAS,CAAC;IAAC,MAAM,EAAE,gBAAgB,CAAA;CAAE,CAAC,CAUrD"}

package/dist/run-critic.d.ts

@@ -1,23 +0,0 @@
-import { type RunScore, type RunScoreWeights } from './run-score';
-import type { Artifact, BudgetLedgerEntry, Run, Span, TraceEvent, TraceStore } from './trace';
-export interface RunTrace {
-    run: Run;
-    spans: Span[];
-    events: TraceEvent[];
-    artifacts: Artifact[];
-    budget: BudgetLedgerEntry[];
-}
-export interface RunCriticOptions {
-    weights?: Partial<RunScoreWeights>;
-    driftPatterns?: RegExp[];
-}
-export declare class RunCritic {
-    private readonly weights?;
-    private readonly driftPatterns;
-    constructor(options?: RunCriticOptions);
-    score(store: TraceStore, runId: string): Promise<RunScore>;
-    scoreTrace(trace: RunTrace): RunScore;
-    rank(score: RunScore): number;
-    private isDrift;
-}
-//# sourceMappingURL=run-critic.d.ts.map

package/dist/run-critic.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-critic.d.ts","sourceRoot":"","sources":["../src/run-critic.ts"],"names":[],"mappings":"AACA,OAAO,EAA8B,KAAK,QAAQ,EAAE,KAAK,eAAe,EAAE,MAAM,aAAa,CAAA;AAC7F,OAAO,KAAK,EAAE,QAAQ,EAAE,iBAAiB,EAAE,GAAG,EAAE,IAAI,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,SAAS,CAAA;AAE7F,MAAM,WAAW,QAAQ;IACvB,GAAG,EAAE,GAAG,CAAA;IACR,KAAK,EAAE,IAAI,EAAE,CAAA;IACb,MAAM,EAAE,UAAU,EAAE,CAAA;IACpB,SAAS,EAAE,QAAQ,EAAE,CAAA;IACrB,MAAM,EAAE,iBAAiB,EAAE,CAAA;CAC5B;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,CAAC,EAAE,OAAO,CAAC,eAAe,CAAC,CAAA;IAClC,aAAa,CAAC,EAAE,MAAM,EAAE,CAAA;CACzB;AAWD,qBAAa,SAAS;IACpB,OAAO,CAAC,QAAQ,CAAC,OAAO,CAAC,CAA0B;IACnD,OAAO,CAAC,QAAQ,CAAC,aAAa,CAAU;gBAE5B,OAAO,GAAE,gBAAqB;IAKpC,KAAK,CAAC,KAAK,EAAE,UAAU,EAAE,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,QAAQ,CAAC;IAYhE,UAAU,CAAC,KAAK,EAAE,QAAQ,GAAG,QAAQ;IAmHrC,IAAI,CAAC,KAAK,EAAE,QAAQ,GAAG,MAAM;IAI7B,OAAO,CAAC,OAAO;CAGhB"}

package/dist/run-evidence.d.ts

@@ -1,32 +0,0 @@
-import type { ControlEvalResult, ControlRunResult } from './control-runtime';
-import { type RunRecord, type RunSplitTag, type RunTokenUsage } from './run-record';
-export interface RunEvidenceMetadata {
-    experimentId: string;
-    candidateId: string;
-    seed: number;
-    model: string;
-    promptHash: string;
-    configHash: string;
-    commitSha: string;
-    splitTag: RunSplitTag;
-    tokenUsage: RunTokenUsage;
-    queueMs?: number;
-    judgeMetadata?: RunRecord['judgeMetadata'];
-    raw?: Record<string, number>;
-}
-export interface ControlRunToRunRecordOptions extends RunEvidenceMetadata {
-    runId?: string;
-    score?: number;
-    failureMode?: string;
-}
-/**
- * Project a completed control-loop run into the strict RunRecord shape used by
- * release gates, optimizer tables, and research reports.
- *
- * The control loop owns live execution evidence. The caller still supplies the
- * experiment-cell metadata because prompt/config hashes, split assignment,
- * model snapshot, and commit SHA are product/harness concerns.
- */
-export declare function controlRunToRunRecord<TState, TAction, TActionResult, TEval extends ControlEvalResult = ControlEvalResult>(run: ControlRunResult<TState, TAction, TActionResult, TEval>, options: ControlRunToRunRecordOptions): RunRecord;
-export declare function scoreFromEvals(evals: readonly ControlEvalResult[]): number | undefined;
-//# sourceMappingURL=run-evidence.d.ts.map

package/dist/run-evidence.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-evidence.d.ts","sourceRoot":"","sources":["../src/run-evidence.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAA;AAC5E,OAAO,EACL,KAAK,SAAS,EACd,KAAK,WAAW,EAChB,KAAK,aAAa,EAEnB,MAAM,cAAc,CAAA;AAGrB,MAAM,WAAW,mBAAmB;IAClC,YAAY,EAAE,MAAM,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,IAAI,EAAE,MAAM,CAAA;IACZ,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,EAAE,MAAM,CAAA;IAClB,UAAU,EAAE,MAAM,CAAA;IAClB,SAAS,EAAE,MAAM,CAAA;IACjB,QAAQ,EAAE,WAAW,CAAA;IACrB,UAAU,EAAE,aAAa,CAAA;IACzB,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB,aAAa,CAAC,EAAE,SAAS,CAAC,eAAe,CAAC,CAAA;IAC1C,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAC7B;AAED,MAAM,WAAW,4BAA6B,SAAQ,mBAAmB;IACvE,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,WAAW,CAAC,EAAE,MAAM,CAAA;CACrB;AAED;;;;;;;GAOG;AACH,wBAAgB,qBAAqB,CACnC,MAAM,EACN,OAAO,EACP,aAAa,EACb,KAAK,SAAS,iBAAiB,GAAG,iBAAiB,EAEnD,GAAG,EAAE,gBAAgB,CAAC,MAAM,EAAE,OAAO,EAAE,aAAa,EAAE,KAAK,CAAC,EAC5D,OAAO,EAAE,4BAA4B,GACpC,SAAS,CA8BX;AAED,wBAAgB,cAAc,CAAC,KAAK,EAAE,SAAS,iBAAiB,EAAE,GAAG,MAAM,GAAG,SAAS,CAMtF"}

package/dist/run-record.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-record.d.ts","sourceRoot":"","sources":["../src/run-record.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;GAuBG;AAEH,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAA;AAE5D,OAAO,EAAE,eAAe,EAAE,MAAM,UAAU,CAAA;AAC1C,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAA;AAElD;sEACsE;AACtE,MAAM,MAAM,WAAW,GAAG,QAAQ,GAAG,KAAK,GAAG,SAAS,CAAA;AAEtD,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,MAAM,CAAA;IACb,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,CAAC,EAAE,MAAM,CAAA;CAChB;AAED,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAA;IACb,aAAa,EAAE,MAAM,CAAA;IACrB;mEAC+D;IAC/D,UAAU,EAAE,MAAM,CAAA;IAClB;kEAC8D;IAC9D,QAAQ,EAAE,OAAO,CAAA;CAClB;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,WAAW,iBAAiB;IAChC,kGAAkG;IAClG,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC,CAAA;IAChD,2EAA2E;IAC3E,UAAU,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAClC;mEAC+D;IAC/D,SAAS,EAAE,MAAM,CAAA;IACjB;;wDAEoD;IACpD,YAAY,CAAC,EAAE,MAAM,EAAE,CAAA;IACvB;iDAC6C;IAC7C,KAAK,CAAC,EAAE,MAAM,CAAA;CACf;AAED,MAAM,WAAW,UAAU;IACzB;6DACyD;IACzD,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB;kEAC8D;IAC9D,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB;;4BAEwB;IACxB,GAAG,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC3B;;;;+BAI2B;IAC3B,WAAW,CAAC,EAAE,iBAAiB,CAAA;IAC/B;;;;;kEAK8D;IAC9D,QAAQ,CAAC,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,KAAK,EAAE,OAAO,CAAC;QAAC,MAAM,CAAC,EAAE,MAAM,CAAA;KAAE,CAAA;CAC9D;AAED;;;;;;;;;;;;;;;GAeG;AACH,MAAM,WAAW,SAAS;IACxB,wBAAwB;IACxB,KAAK,EAAE,MAAM,CAAA;IACb;uDACmD;IACnD,YAAY,EAAE,MAAM,CAAA;IACpB;uEACmE;IACnE,WAAW,EAAE,MAAM,CAAA;IACnB;6DACyD;IACzD,IAAI,EAAE,MAAM,CAAA;IACZ,8CAA8C;IAC9C,KAAK,EAAE,MAAM,CAAA;IACb,sDAAsD;IACtD,UAAU,EAAE,MAAM,CAAA;IAClB,sCAAsC;IACtC,UAAU,EAAE,MAAM,CAAA;IAClB,wCAAwC;IACxC,SAAS,EAAE,MAAM,CAAA;IACjB,sDAAsD;IACtD,MAAM,EAAE,MAAM,CAAA;IACd,4DAA4D;IAC5D,OAAO,CAAC,EAAE,MAAM,CAAA;IAChB;0EACsE;IACtE,OAAO,EAAE,MAAM,CAAA;IACf,6BAA6B;IAC7B,UAAU,EAAE,aAAa,CAAA;IACzB,gDAAgD;IAChD,aAAa,CAAC,EAAE,gBAAgB,CAAA;IAChC,kCAAkC;IAClC,OAAO,EAAE,UAAU,CAAA;IACnB;;;;;yCAKqC;IACrC,YAAY,CAAC,EAAE,YAAY,CAAA;IAC3B;;8EAE0E;IAC1E,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,2CAA2C;IAC3C,QAAQ,EAAE,WAAW,CAAA;IACrB;;;;;;;OAOG;IACH,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB;;;;;;OAMG;IACH,YAAY,CAAC,EAAE,gBAAgB,CAAA;CAChC;AAsBD,qBAAa,wBAAyB,SAAQ,eAAe;IAC3D,QAAQ,CAAC,IAAI,EAAE,MAAM,CAAA;gBACT,OAAO,EAAE,MAAM,EAAE,IAAI,SAAK;CAIvC;AAED;;;;GAIG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,OAAO,GAAG,SAAS,CA4I3D;AAED,4DAA4D;AAC5D,wBAAgB,WAAW,CAAC,KAAK,EAAE,OAAO,GAAG,KAAK,IAAI,SAAS,CAO9D;AAED,8DAA8D;AAC9D,wBAAgB,kBAAkB,CAChC,KAAK,EAAE,OAAO,GACb;IAAE,EAAE,EAAE,IAAI,CAAC;IAAC,KAAK,EAAE,SAAS,CAAA;CAAE,GAAG;IAAE,EAAE,EAAE,KAAK,CAAC;IAAC,KAAK,EAAE,wBAAwB,CAAA;CAAE,CAOjF;AAED,8EAA8E;AAC9E,wBAAgB,kBAAkB,CAAC,MAAM,EAAE,SAAS,GAAG,SAAS,CAG/D"}

package/dist/run-record.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=run-record.test.d.ts.map

package/dist/run-record.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-record.test.d.ts","sourceRoot":"","sources":["../src/run-record.test.ts"],"names":[],"mappings":""}

package/dist/run-score.d.ts

@@ -1,31 +0,0 @@
-export interface RunScore {
-    success: number;
-    goalProgress: number;
-    repoGroundedness: number;
-    driftPenalty: number;
-    toolUseQuality: number;
-    patchQuality: number;
-    testReality: number;
-    finalGate: number;
-    reviewerBlockers: number;
-    costUsd: number;
-    wallSeconds: number;
-    notes?: string[];
-}
-export interface RunScoreWeights {
-    success: number;
-    goalProgress: number;
-    repoGroundedness: number;
-    driftPenalty: number;
-    toolUseQuality: number;
-    patchQuality: number;
-    testReality: number;
-    finalGate: number;
-    reviewerBlockers: number;
-    costUsd: number;
-    wallSeconds: number;
-}
-export declare const DEFAULT_RUN_SCORE_WEIGHTS: RunScoreWeights;
-export declare function aggregateRunScore(score: RunScore, weights?: Partial<RunScoreWeights>): number;
-export declare function clamp01(value: number): number;
-//# sourceMappingURL=run-score.d.ts.map

package/dist/run-score.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"run-score.d.ts","sourceRoot":"","sources":["../src/run-score.ts"],"names":[],"mappings":"AAAA,MAAM,WAAW,QAAQ;IACvB,OAAO,EAAE,MAAM,CAAA;IACf,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,YAAY,EAAE,MAAM,CAAA;IACpB,cAAc,EAAE,MAAM,CAAA;IACtB,YAAY,EAAE,MAAM,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,gBAAgB,EAAE,MAAM,CAAA;IACxB,OAAO,EAAE,MAAM,CAAA;IACf,WAAW,EAAE,MAAM,CAAA;IACnB,KAAK,CAAC,EAAE,MAAM,EAAE,CAAA;CACjB;AAED,MAAM,WAAW,eAAe;IAC9B,OAAO,EAAE,MAAM,CAAA;IACf,YAAY,EAAE,MAAM,CAAA;IACpB,gBAAgB,EAAE,MAAM,CAAA;IACxB,YAAY,EAAE,MAAM,CAAA;IACpB,cAAc,EAAE,MAAM,CAAA;IACtB,YAAY,EAAE,MAAM,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,SAAS,EAAE,MAAM,CAAA;IACjB,gBAAgB,EAAE,MAAM,CAAA;IACxB,OAAO,EAAE,MAAM,CAAA;IACf,WAAW,EAAE,MAAM,CAAA;CACpB;AAED,eAAO,MAAM,yBAAyB,EAAE,eAYvC,CAAA;AAED,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,QAAQ,EAAE,OAAO,GAAE,OAAO,CAAC,eAAe,CAAM,GAAG,MAAM,CAejG;AAED,wBAAgB,OAAO,CAAC,KAAK,EAAE,MAAM,GAAG,MAAM,CAG7C"}

package/dist/runtime-trajectory.d.ts

@@ -1,47 +0,0 @@
-import type { RunSplitTag } from './run-record';
-export interface RuntimeTrajectoryHookEvent {
-    id: string;
-    runId: string;
-    scenarioId?: string;
-    target: string;
-    phase: string;
-    timestamp: number;
-    stepIndex?: number;
-    parentId?: string;
-    payload?: unknown;
-    metadata?: Record<string, unknown>;
-}
-export interface RuntimeTrajectoryRecord {
-    id?: string;
-    scenarioId?: string;
-    splitTag?: RunSplitTag;
-    runtimeEvents?: unknown;
-    [key: string]: unknown;
-}
-export interface RuntimeTrajectoryRunRecord {
-    runId: string;
-    scenarioId?: string;
-    splitTag: RunSplitTag;
-}
-export interface RuntimeTrajectoryEvidenceSummary {
-    recordCount: number;
-    recordWithRuntimeEventsCount: number;
-    runtimeRunCount: number;
-    lifecycleEventCount: number;
-    defaultedSplitCount: number;
-}
-export interface RuntimeTrajectoryEvidenceProjection {
-    runs: RuntimeTrajectoryRunRecord[];
-    events: RuntimeTrajectoryHookEvent[];
-    summary: RuntimeTrajectoryEvidenceSummary;
-    diagnostics: string[];
-}
-export interface ProjectRuntimeTrajectoryEvidenceOptions<TRecord extends RuntimeTrajectoryRecord = RuntimeTrajectoryRecord> {
-    records: TRecord[];
-    defaultSplitTag?: RunSplitTag;
-    recordIdOf?: (record: TRecord, index: number) => string | undefined;
-    scenarioIdOf?: (record: TRecord, index: number) => string | undefined;
-}
-export declare function projectRuntimeTrajectoryEvidence<TRecord extends RuntimeTrajectoryRecord>(options: ProjectRuntimeTrajectoryEvidenceOptions<TRecord>): RuntimeTrajectoryEvidenceProjection;
-export declare function parseRuntimeTrajectoryHookEvent(input: unknown): RuntimeTrajectoryHookEvent | null;
-//# sourceMappingURL=runtime-trajectory.d.ts.map

package/dist/runtime-trajectory.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"runtime-trajectory.d.ts","sourceRoot":"","sources":["../src/runtime-trajectory.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,cAAc,CAAA;AAE/C,MAAM,WAAW,0BAA0B;IACzC,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,MAAM,EAAE,MAAM,CAAA;IACd,KAAK,EAAE,MAAM,CAAA;IACb,SAAS,EAAE,MAAM,CAAA;IACjB,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAA;IACjB,OAAO,CAAC,EAAE,OAAO,CAAA;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;CACnC;AAED,MAAM,WAAW,uBAAuB;IACtC,EAAE,CAAC,EAAE,MAAM,CAAA;IACX,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,QAAQ,CAAC,EAAE,WAAW,CAAA;IACtB,aAAa,CAAC,EAAE,OAAO,CAAA;IACvB,CAAC,GAAG,EAAE,MAAM,GAAG,OAAO,CAAA;CACvB;AAED,MAAM,WAAW,0BAA0B;IACzC,KAAK,EAAE,MAAM,CAAA;IACb,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,QAAQ,EAAE,WAAW,CAAA;CACtB;AAED,MAAM,WAAW,gCAAgC;IAC/C,WAAW,EAAE,MAAM,CAAA;IACnB,4BAA4B,EAAE,MAAM,CAAA;IACpC,eAAe,EAAE,MAAM,CAAA;IACvB,mBAAmB,EAAE,MAAM,CAAA;IAC3B,mBAAmB,EAAE,MAAM,CAAA;CAC5B;AAED,MAAM,WAAW,mCAAmC;IAClD,IAAI,EAAE,0BAA0B,EAAE,CAAA;IAClC,MAAM,EAAE,0BAA0B,EAAE,CAAA;IACpC,OAAO,EAAE,gCAAgC,CAAA;IACzC,WAAW,EAAE,MAAM,EAAE,CAAA;CACtB;AAED,MAAM,WAAW,uCAAuC,CACtD,OAAO,SAAS,uBAAuB,GAAG,uBAAuB;IAEjE,OAAO,EAAE,OAAO,EAAE,CAAA;IAClB,eAAe,CAAC,EAAE,WAAW,CAAA;IAC7B,UAAU,CAAC,EAAE,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,KAAK,MAAM,GAAG,SAAS,CAAA;IACnE,YAAY,CAAC,EAAE,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,MAAM,KAAK,MAAM,GAAG,SAAS,CAAA;CACtE;AAID,wBAAgB,gCAAgC,CAAC,OAAO,SAAS,uBAAuB,EACtF,OAAO,EAAE,uCAAuC,CAAC,OAAO,CAAC,GACxD,mCAAmC,CA8DrC;AAED,wBAAgB,+BAA+B,CAAC,KAAK,EAAE,OAAO,GAAG,0BAA0B,GAAG,IAAI,CAoBjG"}

package/dist/sandbox-harness.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-harness.d.ts","sourceRoot":"","sources":["../src/sandbox-harness.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAGnD,MAAM,WAAW,aAAa;IAC5B,wEAAwE;IACxE,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB,uCAAuC;IACvC,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,iFAAiF;IACjF,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB,wEAAwE;IACxE,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ,0DAA0D;IAC1D,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;;;;OAIG;IACH,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,mCAAmC;IACnC,KAAK,CAAC,EAAE,MAAM,CAAA;IACd,iDAAiD;IACjD,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IAC5B,8EAA8E;IAC9E,UAAU,CAAC,EAAE,gBAAgB,CAAA;CAC9B;AAED,MAAM,WAAW,gBAAgB;IAC/B,EAAE,EAAE,MAAM,CAAA;IACV,KAAK,CACH,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,MAAM,EACd,QAAQ,EAAE,MAAM,GACf;QAAE,UAAU,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAA;KAAE,GAAG,SAAS,CAAA;CAC3D;AAED,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,OAAO,GAAG,KAAK,GAAG,MAAM,CAAA;IAC/B,QAAQ,EAAE,MAAM,CAAA;IAChB,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,MAAM,EAAE,MAAM,CAAA;IACd,UAAU,CAAC,EAAE,MAAM,CAAA;IACnB,WAAW,CAAC,EAAE,MAAM,CAAA;IACpB;;;;;OAKG;IACH,eAAe,CAAC,EAAE,OAAO,CAAA;IACzB;;;;OAIG;IACH,eAAe,CAAC,EAAE,OAAO,CAAA;CAC1B;AAED,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,MAAM,CAAA;IACV,IAAI,CACF,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,EAC7B,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,aAAa,GACpB,OAAO,CAAC,aAAa,CAAC,CAAA;CAC1B;AAID,iEAAiE;AACjE,eAAO,MAAM,gBAAgB,EAAE,gBAkB9B,CAAA;AAED,mEAAmE;AACnE,eAAO,MAAM,gBAAgB,EAAE,gBAQ9B,CAAA;AAED,8DAA8D;AAC9D,eAAO,MAAM,cAAc,EAAE,gBAO5B,CAAA;AAED,2DAA2D;AAC3D,wBAAgB,cAAc,CAAC,GAAG,OAAO,EAAE,gBAAgB,EAAE,GAAG,gBAAgB,CAW/E;AAID,MAAM,WAAW,8BAA8B;IAC7C;;;;;;;OAOG;IACH,GAAG,CAAC,EAAE,MAAM,CAAA;IACZ;;;OAGG;IACH,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAC7B;AAED,qBAAa,uBAAwB,YAAW,aAAa;IAC3D,EAAE,SAAe;IACjB,OAAO,CAAC,UAAU,CAAC,CAAQ;IAC3B,OAAO,CAAC,UAAU,CAAC,CAAwB;gBAE/B,OAAO,GAAE,8BAAmC;IAKlD,IAAI,CACR,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,EAC7B,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,aAAa,GACpB,OAAO,CAAC,aAAa,CAAC;CAmI1B;AAED,qBAAa,mBAAoB,YAAW,aAAa;IACvD,EAAE,SAAW;IAEP,IAAI,CACR,KAAK,EAAE,aAAa,CAAC,OAAO,CAAC,EAC7B,OAAO,EAAE,MAAM,EACf,MAAM,EAAE,aAAa,GACpB,OAAO,CAAC,aAAa,CAAC;CAS1B;AASD,MAAM,WAAW,oBAAoB;IACnC,MAAM,EAAE,OAAO,CAAA;IACf,KAAK,CAAC,EAAE,aAAa,CAAA;IACrB,GAAG,CAAC,EAAE,aAAa,CAAA;IACnB,IAAI,CAAC,EAAE,aAAa,CAAA;IACpB,WAAW,EAAE,MAAM,CAAA;IACnB,uEAAuE;IACvE,KAAK,EAAE,MAAM,CAAA;CACd;AAED,qBAAa,cAAc;IACzB,OAAO,CAAC,MAAM,CAAe;gBACjB,MAAM,GAAE,aAA6C;IAI3D,GAAG,CAAC,MAAM,EAAE,aAAa,EAAE,OAAO,EAAE,YAAY,GAAG,OAAO,CAAC,oBAAoB,CAAC;CAoEvF"}

package/dist/sandbox-harness.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=sandbox-harness.test.d.ts.map

package/dist/sandbox-harness.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-harness.test.d.ts","sourceRoot":"","sources":["../src/sandbox-harness.test.ts"],"names":[],"mappings":""}

package/dist/sandbox-pool.d.ts

@@ -1,74 +0,0 @@
-/**
- * SandboxPool — bounded checkout/release pool for mutation slots.
- *
- * The composite-mutator's `code` channel needs an isolated workspace per
- * mutation attempt: a git worktree, a sandbox container, a tmpdir clone —
- * whatever the consumer's runtime is. Without a pool, every consumer
- * re-implements the same machinery (mint N slots, check one out per
- * mutation, reset before reuse, drain at the end, track utilisation for
- * the cost ledger). This primitive ships that machinery so consumers
- * supply only a `SlotFactory`.
- *
- * Generic over a slot resource `T` so the same pool serves git worktrees
- * (T = path), Tangle sandboxes (T = SandboxBox), or anything else with
- * the create/reset/destroy lifecycle.
- *
- * Concurrency: FIFO via the shared `Mutex` primitive. Each `checkout()`
- * either takes an idle slot or queues until one is released. Lifecycle
- * is single-process — multi-process pools need external coordination
- * (file locks, etc.) and are deliberately out of scope.
- */
-export interface PoolSlot<T> {
-    /** Stable id assigned at slot creation. Use for telemetry / lineage. */
-    readonly id: string;
-    /** Consumer-defined resource. */
-    readonly resource: T;
-}
-export interface SlotFactory<T> {
-    /** Build a new slot. Called lazily as the pool grows up to `size`. */
-    create(slotId: string): Promise<T>;
-    /**
-     * Reset a slot to a clean state before reuse. Called BEFORE every
-     * checkout returns it (including the first — so the factory's
-     * `create` can leave the slot dirty and let `reset` normalise).
-     * Optional; default is a no-op.
-     */
-    reset?(slot: PoolSlot<T>): Promise<void>;
-    /** Tear the slot down. Called by `drain()`. */
-    destroy(slot: PoolSlot<T>): Promise<void>;
-}
-export interface SandboxPool<T> {
-    /**
-     * Take a slot. If all slots are busy, the promise resolves when one
-     * is released. Always pair with the returned `release` (or wrap with
-     * `withSlot`).
-     */
-    checkout(): Promise<{
-        slot: PoolSlot<T>;
-        release: () => void;
-    }>;
-    /**
-     * Run `fn` with a checked-out slot, releasing on completion or throw.
-     * The convenience wrapper most callers should use.
-     */
-    withSlot<R>(fn: (slot: PoolSlot<T>) => Promise<R>): Promise<R>;
-    /** Destroy every slot. Idempotent. */
-    drain(): Promise<void>;
-    /** How many slots have been minted (≤ `size`). */
-    poolSize(): number;
-    /** How many checkouts are currently outstanding. */
-    activeCheckouts(): number;
-    /** Snapshot of busy/total durations for the cost ledger. */
-    utilization(): {
-        busyMs: number;
-        totalMs: number;
-        checkouts: number;
-    };
-}
-export interface CreateSandboxPoolOpts<T> {
-    /** Maximum concurrent slots. Slots are minted on first need, not eagerly. */
-    size: number;
-    factory: SlotFactory<T>;
-}
-export declare function createSandboxPool<T>(opts: CreateSandboxPoolOpts<T>): SandboxPool<T>;
-//# sourceMappingURL=sandbox-pool.d.ts.map

package/dist/sandbox-pool.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-pool.d.ts","sourceRoot":"","sources":["../src/sandbox-pool.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;GAmBG;AAIH,MAAM,WAAW,QAAQ,CAAC,CAAC;IACzB,wEAAwE;IACxE,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAA;IACnB,iCAAiC;IACjC,QAAQ,CAAC,QAAQ,EAAE,CAAC,CAAA;CACrB;AAED,MAAM,WAAW,WAAW,CAAC,CAAC;IAC5B,sEAAsE;IACtE,MAAM,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,CAAA;IAClC;;;;;OAKG;IACH,KAAK,CAAC,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;IACxC,+CAA+C;IAC/C,OAAO,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,CAAA;CAC1C;AAED,MAAM,WAAW,WAAW,CAAC,CAAC;IAC5B;;;;OAIG;IACH,QAAQ,IAAI,OAAO,CAAC;QAAE,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,CAAC;QAAC,OAAO,EAAE,MAAM,IAAI,CAAA;KAAE,CAAC,CAAA;IAC/D;;;OAGG;IACH,QAAQ,CAAC,CAAC,EAAE,EAAE,EAAE,CAAC,IAAI,EAAE,QAAQ,CAAC,CAAC,CAAC,KAAK,OAAO,CAAC,CAAC,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAA;IAC9D,sCAAsC;IACtC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC,CAAA;IACtB,kDAAkD;IAClD,QAAQ,IAAI,MAAM,CAAA;IAClB,oDAAoD;IACpD,eAAe,IAAI,MAAM,CAAA;IACzB,4DAA4D;IAC5D,WAAW,IAAI;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,OAAO,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAA;KAAE,CAAA;CACtE;AAED,MAAM,WAAW,qBAAqB,CAAC,CAAC;IACtC,6EAA6E;IAC7E,IAAI,EAAE,MAAM,CAAA;IACZ,OAAO,EAAE,WAAW,CAAC,CAAC,CAAC,CAAA;CACxB;AAOD,wBAAgB,iBAAiB,CAAC,CAAC,EAAE,IAAI,EAAE,qBAAqB,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAyMnF"}

package/dist/sandbox-pool.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=sandbox-pool.test.d.ts.map

package/dist/sandbox-pool.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-pool.test.d.ts","sourceRoot":"","sources":["../src/sandbox-pool.test.ts"],"names":[],"mappings":""}

package/dist/scorecard.d.ts

@@ -1,133 +0,0 @@
-/**
- * @stable
- *
- * Eval scorecard — the persistent (persona × profile) score timeline.
- *
- * Every benchmark run folds into per-cell entries; a cell is
- * `(scenarioId, profileHash)` and its timeline carries one entry per commit.
- * The scorecard answers the question a single run cannot: did THIS change
- * regress persona P on profile F, even while the aggregate improved?
- *
- * Storage is an append-only JSONL log — one line per (cell, commit). Appends
- * never read-modify-write, so concurrent campaign runs cannot clobber each
- * other; `loadScorecard` folds the log into the queryable `Scorecard`, and a
- * malformed line never breaks the read. `diffScorecard` compares the latest
- * entry of each cell against its predecessor with Cohen's d + Welch's t-test.
- */
-import type { AgentProfile } from './agent-profile';
-import type { RunRecord } from './run-record';
-/** One commit's measurement of one (scenario, profile) cell. */
-export interface ScorecardEntry {
-    commitSha: string;
-    /** ISO timestamp the entry was recorded. */
-    timestamp: string;
-    /** Per-seed (or per-rep) scores for this cell at this commit. */
-    scores: number[];
-    /** Median of `scores` — the cell's headline score for the commit. */
-    composite: number;
-    /** Per-dimension means, when the runs carried a judge breakdown. */
-    perDimension?: Record<string, number>;
-    /** RunRecord ids folded into this entry — provenance. */
-    runIds: string[];
-}
-/** A (scenario, profile) cell and its commit-ordered score timeline. */
-export interface ScorecardCell {
-    scenarioId: string;
-    profileHash: string;
-    /** Model id — denormalised from the profile for readable filtering. */
-    model: string;
-    timeline: ScorecardEntry[];
-}
-/** The folded scorecard: every cell, plus the profile definitions by hash. */
-export interface Scorecard {
-    cells: ScorecardCell[];
-    /** Profile definitions seen — keeps the scorecard self-describing. */
-    profiles: Record<string, AgentProfile>;
-}
-/** One append-only log line — a single cell's entry for a single commit. */
-export interface ScorecardLogLine {
-    scenarioId: string;
-    profileHash: string;
-    model: string;
-    profile: AgentProfile;
-    entry: ScorecardEntry;
-}
-export interface RecordRunsOptions {
-    /** The profile that produced these runs — keys the cell. */
-    profile: AgentProfile;
-    commitSha: string;
-    /** Defaults to `new Date().toISOString()`. */
-    timestamp?: string;
-}
-/**
- * Fold a benchmark's `RunRecord[]` into per-cell scorecard log lines — one
- * line per scenario the runs cover. All runs are attributed to the single
- * `profile` in `opts` (the harness ran them under it); the cell key is
- * `(scenarioId, agentProfileHash(profile))`.
- */
-export declare function recordRuns(runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
-/** Append cell entries to the JSONL scorecard log. Creates the file/dir. */
-export declare function appendScorecard(logPath: string, lines: ScorecardLogLine[]): void;
-/** Record runs and append them to the log in one call. Returns the lines. */
-export declare function recordRunsToScorecard(logPath: string, runs: RunRecord[], opts: RecordRunsOptions): ScorecardLogLine[];
-/**
- * Fold the JSONL log into a queryable `Scorecard`. A missing file yields an
- * empty scorecard; a malformed line is skipped — a corrupt append never
- * breaks the read. Each cell's timeline is sorted chronologically.
- */
-export declare function loadScorecard(logPath: string): Scorecard;
-export type CellVerdict = 'improved' | 'regressed' | 'flat' | 'new';
-export interface ScorecardCellDiff {
-    scenarioId: string;
-    profileHash: string;
-    model: string;
-    verdict: CellVerdict;
-    /** Composite of the latest entry. */
-    current: number;
-    /** Composite of the comparison entry — `null` when `verdict === 'new'`. */
-    baseline: number | null;
-    /** `current − baseline` — `null` when new. */
-    delta: number | null;
-    /** Cohen's d of current vs baseline samples — `null` when new or n < 2. */
-    cohensD: number | null;
-    /** Welch's t-test p-value — `null` when new or n < 2. */
-    pValue: number | null;
-    currentCommit: string;
-    baselineCommit: string | null;
-}
-export interface ScorecardDiff {
-    cells: ScorecardCellDiff[];
-    summary: {
-        improved: number;
-        regressed: number;
-        flat: number;
-        new: number;
-    };
-}
-export interface DiffScorecardOptions {
-    /** Compare each cell against this commit instead of its immediate predecessor. */
-    baselineCommit?: string;
-    /** |Cohen's d| at/above which a move counts as real. Default 0.5. */
-    minEffect?: number;
-    /** p-value at/below which a move is significant. Default 0.05. */
-    maxP?: number;
-    /**
-     * |delta| at/above which a move counts when statistics are unavailable
-     * (a cell with fewer than 2 samples on either side). Default 0.05.
-     */
-    minDelta?: number;
-}
-/**
- * Compare the latest entry of every cell against its predecessor (or against
- * `baselineCommit`) and classify the move. A move is `improved`/`regressed`
- * only when it clears both the effect-size and significance gates; otherwise
- * `flat`. Cells with no prior entry are `new`.
- */
-export declare function diffScorecard(scorecard: Scorecard, opts?: DiffScorecardOptions): ScorecardDiff;
-/**
- * Render a scorecard diff as a human-readable report — the block a feature
- * PR prints. Regressions are listed first; flat cells are summarised, not
- * enumerated.
- */
-export declare function formatScorecardDiff(diff: ScorecardDiff): string;
-//# sourceMappingURL=scorecard.d.ts.map

package/dist/scorecard.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"scorecard.d.ts","sourceRoot":"","sources":["../src/scorecard.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAIH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,iBAAiB,CAAA;AAGnD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,cAAc,CAAA;AAK7C,gEAAgE;AAChE,MAAM,WAAW,cAAc;IAC7B,SAAS,EAAE,MAAM,CAAA;IACjB,4CAA4C;IAC5C,SAAS,EAAE,MAAM,CAAA;IACjB,iEAAiE;IACjE,MAAM,EAAE,MAAM,EAAE,CAAA;IAChB,qEAAqE;IACrE,SAAS,EAAE,MAAM,CAAA;IACjB,oEAAoE;IACpE,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;IACrC,yDAAyD;IACzD,MAAM,EAAE,MAAM,EAAE,CAAA;CACjB;AAED,wEAAwE;AACxE,MAAM,WAAW,aAAa;IAC5B,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,uEAAuE;IACvE,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,cAAc,EAAE,CAAA;CAC3B;AAED,8EAA8E;AAC9E,MAAM,WAAW,SAAS;IACxB,KAAK,EAAE,aAAa,EAAE,CAAA;IACtB,sEAAsE;IACtE,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAA;CACvC;AAED,4EAA4E;AAC5E,MAAM,WAAW,gBAAgB;IAC/B,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,YAAY,CAAA;IACrB,KAAK,EAAE,cAAc,CAAA;CACtB;AAID,MAAM,WAAW,iBAAiB;IAChC,4DAA4D;IAC5D,OAAO,EAAE,YAAY,CAAA;IACrB,SAAS,EAAE,MAAM,CAAA;IACjB,8CAA8C;IAC9C,SAAS,CAAC,EAAE,MAAM,CAAA;CACnB;AAkCD;;;;;GAKG;AACH,wBAAgB,UAAU,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,IAAI,EAAE,iBAAiB,GAAG,gBAAgB,EAAE,CAuCzF;AAID,4EAA4E;AAC5E,wBAAgB,eAAe,CAAC,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,gBAAgB,EAAE,GAAG,IAAI,CAIhF;AAED,6EAA6E;AAC7E,wBAAgB,qBAAqB,CACnC,OAAO,EAAE,MAAM,EACf,IAAI,EAAE,SAAS,EAAE,EACjB,IAAI,EAAE,iBAAiB,GACtB,gBAAgB,EAAE,CAIpB;AAED;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,OAAO,EAAE,MAAM,GAAG,SAAS,CAkCxD;AAID,MAAM,MAAM,WAAW,GAAG,UAAU,GAAG,WAAW,GAAG,MAAM,GAAG,KAAK,CAAA;AAEnE,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,MAAM,CAAA;IAClB,WAAW,EAAE,MAAM,CAAA;IACnB,KAAK,EAAE,MAAM,CAAA;IACb,OAAO,EAAE,WAAW,CAAA;IACpB,qCAAqC;IACrC,OAAO,EAAE,MAAM,CAAA;IACf,2EAA2E;IAC3E,QAAQ,EAAE,MAAM,GAAG,IAAI,CAAA;IACvB,8CAA8C;IAC9C,KAAK,EAAE,MAAM,GAAG,IAAI,CAAA;IACpB,2EAA2E;IAC3E,OAAO,EAAE,MAAM,GAAG,IAAI,CAAA;IACtB,yDAAyD;IACzD,MAAM,EAAE,MAAM,GAAG,IAAI,CAAA;IACrB,aAAa,EAAE,MAAM,CAAA;IACrB,cAAc,EAAE,MAAM,GAAG,IAAI,CAAA;CAC9B;AAED,MAAM,WAAW,aAAa;IAC5B,KAAK,EAAE,iBAAiB,EAAE,CAAA;IAC1B,OAAO,EAAE;QAAE,QAAQ,EAAE,MAAM,CAAC;QAAC,SAAS,EAAE,MAAM,CAAC;QAAC,IAAI,EAAE,MAAM,CAAC;QAAC,GAAG,EAAE,MAAM,CAAA;KAAE,CAAA;CAC5E;AAED,MAAM,WAAW,oBAAoB;IACnC,kFAAkF;IAClF,cAAc,CAAC,EAAE,MAAM,CAAA;IACvB,qEAAqE;IACrE,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB,kEAAkE;IAClE,IAAI,CAAC,EAAE,MAAM,CAAA;IACb;;;OAGG;IACH,QAAQ,CAAC,EAAE,MAAM,CAAA;CAClB;AAED;;;;;GAKG;AACH,wBAAgB,aAAa,CAC3B,SAAS,EAAE,SAAS,EACpB,IAAI,GAAE,oBAAyB,GAC9B,aAAa,CAsEf;AAED;;;;GAIG;AACH,wBAAgB,mBAAmB,CAAC,IAAI,EAAE,aAAa,GAAG,MAAM,CAgC/D"}

package/dist/scorecard.test.d.ts

@@ -1,2 +0,0 @@
-export {};
-//# sourceMappingURL=scorecard.test.d.ts.map

package/dist/scorecard.test.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"scorecard.test.d.ts","sourceRoot":"","sources":["../src/scorecard.test.ts"],"names":[],"mappings":""}

package/dist/self-play.d.ts

@@ -1,69 +0,0 @@
-/**
- * Self-play scenario evolution — agents generate adversarial scenarios
- * against each other; survivors become part of the eval corpus.
- *
- * Framework-agnostic about how scenarios are generated. Caller supplies:
- *   - `propose`: asks a "proposer" agent for candidate scenarios
- *   - `scoreAgainst`: runs a target agent against a scenario and returns
- *     its score
- *
- * A scenario *survives* if it reveals a meaningful score difference
- * between two target agents (or between a target agent and itself on
- * different runs). Survivors are promoted to a Dataset; the caller
- * decides what to do with them (hold-out, training, regression set).
- *
- * Guard rails: minimum absolute score delta to consider a scenario
- * informative; floor on absolute target score so degenerate break-all
- * scenarios (noise, gibberish) don't flood the corpus.
- */
-import { Dataset } from './dataset';
-export interface CandidateScenario {
-    id: string;
-    payload: unknown;
-    /** Free-form tags (domain, generation, parent). */
-    tags?: Record<string, string>;
-}
-export interface ScoredTarget {
-    targetId: string;
-    score: number;
-}
-export interface EvolutionRound {
-    round: number;
-    proposed: CandidateScenario[];
-    survived: CandidateScenario[];
-    rejected: Array<{
-        candidate: CandidateScenario;
-        reason: string;
-    }>;
-    scoredBreakdown: Array<{
-        candidate: CandidateScenario;
-        scores: ScoredTarget[];
-        spread: number;
-    }>;
-}
-export interface SelfPlayOptions {
-    /** Minimum score spread across targets for a scenario to survive. Default 0.1. */
-    minSpread?: number;
-    /** Minimum floor score across targets — keeps degenerate break-all scenarios
-     *  out. Default 0.1 (if every target scores below this, discard). */
-    minAbsoluteFloor?: number;
-    /** Hard cap on survivors per round. Default 50. */
-    maxSurvivors?: number;
-    /** Rounds to run. Default 1. Each round's survivors can be fed back into
-     *  `propose` to compound. */
-    rounds?: number;
-    /** Seed for scenario id generation if proposer doesn't provide one. */
-    seed?: number;
-}
-export interface SelfPlayProposer {
-    propose(round: number, priorSurvivors: CandidateScenario[]): Promise<CandidateScenario[]>;
-}
-export interface SelfPlayScorer {
-    /** Score one candidate against every target; returns parallel array. */
-    scoreCandidate(candidate: CandidateScenario, targets: string[]): Promise<ScoredTarget[]>;
-}
-export declare function runSelfPlay(proposer: SelfPlayProposer, scorer: SelfPlayScorer, targets: string[], options?: SelfPlayOptions): Promise<{
-    rounds: EvolutionRound[];
-    dataset: Dataset;
-}>;
-//# sourceMappingURL=self-play.d.ts.map

package/dist/self-play.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"self-play.d.ts","sourceRoot":"","sources":["../src/self-play.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;GAiBG;AAEH,OAAO,EAAE,OAAO,EAAwB,MAAM,WAAW,CAAA;AAEzD,MAAM,WAAW,iBAAiB;IAChC,EAAE,EAAE,MAAM,CAAA;IACV,OAAO,EAAE,OAAO,CAAA;IAChB,mDAAmD;IACnD,IAAI,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAA;CAC9B;AAED,MAAM,WAAW,YAAY;IAC3B,QAAQ,EAAE,MAAM,CAAA;IAChB,KAAK,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,cAAc;IAC7B,KAAK,EAAE,MAAM,CAAA;IACb,QAAQ,EAAE,iBAAiB,EAAE,CAAA;IAC7B,QAAQ,EAAE,iBAAiB,EAAE,CAAA;IAC7B,QAAQ,EAAE,KAAK,CAAC;QAAE,SAAS,EAAE,iBAAiB,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;IACjE,eAAe,EAAE,KAAK,CAAC;QAAE,SAAS,EAAE,iBAAiB,CAAC;QAAC,MAAM,EAAE,YAAY,EAAE,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,CAAC,CAAA;CACjG;AAED,MAAM,WAAW,eAAe;IAC9B,kFAAkF;IAClF,SAAS,CAAC,EAAE,MAAM,CAAA;IAClB;yEACqE;IACrE,gBAAgB,CAAC,EAAE,MAAM,CAAA;IACzB,mDAAmD;IACnD,YAAY,CAAC,EAAE,MAAM,CAAA;IACrB;iCAC6B;IAC7B,MAAM,CAAC,EAAE,MAAM,CAAA;IACf,uEAAuE;IACvE,IAAI,CAAC,EAAE,MAAM,CAAA;CACd;AAED,MAAM,WAAW,gBAAgB;IAC/B,OAAO,CAAC,KAAK,EAAE,MAAM,EAAE,cAAc,EAAE,iBAAiB,EAAE,GAAG,OAAO,CAAC,iBAAiB,EAAE,CAAC,CAAA;CAC1F;AAED,MAAM,WAAW,cAAc;IAC7B,wEAAwE;IACxE,cAAc,CAAC,SAAS,EAAE,iBAAiB,EAAE,OAAO,EAAE,MAAM,EAAE,GAAG,OAAO,CAAC,YAAY,EAAE,CAAC,CAAA;CACzF;AAED,wBAAsB,WAAW,CAC/B,QAAQ,EAAE,gBAAgB,EAC1B,MAAM,EAAE,cAAc,EACtB,OAAO,EAAE,MAAM,EAAE,EACjB,OAAO,GAAE,eAAoB,GAC5B,OAAO,CAAC;IAAE,MAAM,EAAE,cAAc,EAAE,CAAC;IAAC,OAAO,EAAE,OAAO,CAAA;CAAE,CAAC,CA4EzD"}