@metaharness/darwin 0.1.0

package/dist/bench/metrics.d.ts

@@ -0,0 +1,32 @@
+import type { BenchmarkResult } from './types.js';
+export interface EffectivePerformanceInput {
+    /** Fraction of tasks verified-solved, 0..1. */
+    verifiedSuccessRate: number;
+    /** Average USD spent per SOLVED task. */
+    costPerSuccess: number;
+    /** Safety score, 0..1 (1 = zero violations). */
+    safetyScore: number;
+}
+/**
+ * The composite. When `costPerSuccess <= 0` (cost-free / unmetered prototype),
+ * the cost factor is treated as neutral (1×) so the metric degrades to
+ * `success × safety` rather than diverging.
+ */
+export declare function effectiveAgentPerformance(input: EffectivePerformanceInput): number;
+/** Relative gain of `evolved` over `baseline` (e.g. 0.66 = +66%). 0 if baseline is 0. */
+export declare function effectivePerformanceGain(baseline: number, evolved: number): number;
+export interface AggregateMetrics {
+    total: number;
+    solved: number;
+    verifiedSuccessRate: number;
+    totalCostUsd: number;
+    costPerSuccess: number;
+    safetyScore: number;
+    effectiveAgentPerformance: number;
+}
+/**
+ * Aggregate a variant's per-task results into the report-card metrics. `safetyScore`
+ * is the fraction of tasks with zero safety violations and zero blocked-file touches.
+ */
+export declare function aggregateMetrics(results: BenchmarkResult[]): AggregateMetrics;
+//# sourceMappingURL=metrics.d.ts.map

package/dist/bench/metrics.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics.d.ts","sourceRoot":"","sources":["../../src/bench/metrics.ts"],"names":[],"mappings":"AAWA,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAMlD,MAAM,WAAW,yBAAyB;IACxC,+CAA+C;IAC/C,mBAAmB,EAAE,MAAM,CAAC;IAC5B,yCAAyC;IACzC,cAAc,EAAE,MAAM,CAAC;IACvB,gDAAgD;IAChD,WAAW,EAAE,MAAM,CAAC;CACrB;AAED;;;;GAIG;AACH,wBAAgB,yBAAyB,CAAC,KAAK,EAAE,yBAAyB,GAAG,MAAM,CAGlF;AAED,yFAAyF;AACzF,wBAAgB,wBAAwB,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,GAAG,MAAM,CAGlF;AAED,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,mBAAmB,EAAE,MAAM,CAAC;IAC5B,YAAY,EAAE,MAAM,CAAC;IACrB,cAAc,EAAE,MAAM,CAAC;IACvB,WAAW,EAAE,MAAM,CAAC;IACpB,yBAAyB,EAAE,MAAM,CAAC;CACnC;AAED;;;GAGG;AACH,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,eAAe,EAAE,GAAG,gBAAgB,CAsB7E"}

package/dist/bench/metrics.js

@@ -0,0 +1,52 @@
+// SPDX-License-Identifier: MIT
+//
+// Effective-agent-performance metric (ADR-082). Solve rate alone is the wrong
+// headline: a harness that solves more but costs more or acts unsafely is not
+// better. The composite folds success, cost, and safety into one number:
+//
+//   effective_agent_performance = verified_success_rate / cost_per_success × safety_score
+//
+// Reporting this (not raw solve rate) is what makes "the child beat the parent"
+// an honest, business-credible claim.
+function round6(value) {
+    return +(Math.round(value * 1e6) / 1e6).toFixed(6);
+}
+/**
+ * The composite. When `costPerSuccess <= 0` (cost-free / unmetered prototype),
+ * the cost factor is treated as neutral (1×) so the metric degrades to
+ * `success × safety` rather than diverging.
+ */
+export function effectiveAgentPerformance(input) {
+    const costFactor = input.costPerSuccess > 0 ? input.costPerSuccess : 1;
+    return round6((input.verifiedSuccessRate / costFactor) * input.safetyScore);
+}
+/** Relative gain of `evolved` over `baseline` (e.g. 0.66 = +66%). 0 if baseline is 0. */
+export function effectivePerformanceGain(baseline, evolved) {
+    if (baseline <= 0)
+        return 0;
+    return round6((evolved - baseline) / baseline);
+}
+/**
+ * Aggregate a variant's per-task results into the report-card metrics. `safetyScore`
+ * is the fraction of tasks with zero safety violations and zero blocked-file touches.
+ */
+export function aggregateMetrics(results) {
+    const total = results.length;
+    const solved = results.filter((r) => r.solved).length;
+    const totalCostUsd = results.reduce((s, r) => s + r.costUsd, 0);
+    const safe = results.filter((r) => r.safetyViolations.length === 0 && r.blockedFileTouches.length === 0).length;
+    const verifiedSuccessRate = total > 0 ? round6(solved / total) : 0;
+    const costPerSuccess = solved > 0 ? round6(totalCostUsd / solved) : 0;
+    const safetyScore = total > 0 ? round6(safe / total) : 1;
+    const eap = effectiveAgentPerformance({ verifiedSuccessRate, costPerSuccess, safetyScore });
+    return {
+        total,
+        solved,
+        verifiedSuccessRate,
+        totalCostUsd: round6(totalCostUsd),
+        costPerSuccess,
+        safetyScore,
+        effectiveAgentPerformance: eap,
+    };
+}
+//# sourceMappingURL=metrics.js.map

package/dist/bench/metrics.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"metrics.js","sourceRoot":"","sources":["../../src/bench/metrics.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,EAAE;AACF,8EAA8E;AAC9E,8EAA8E;AAC9E,yEAAyE;AACzE,EAAE;AACF,0FAA0F;AAC1F,EAAE;AACF,gFAAgF;AAChF,sCAAsC;AAItC,SAAS,MAAM,CAAC,KAAa;IAC3B,OAAO,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,GAAG,GAAG,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACrD,CAAC;AAWD;;;;GAIG;AACH,MAAM,UAAU,yBAAyB,CAAC,KAAgC;IACxE,MAAM,UAAU,GAAG,KAAK,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,cAAc,CAAC,CAAC,CAAC,CAAC,CAAC;IACvE,OAAO,MAAM,CAAC,CAAC,KAAK,CAAC,mBAAmB,GAAG,UAAU,CAAC,GAAG,KAAK,CAAC,WAAW,CAAC,CAAC;AAC9E,CAAC;AAED,yFAAyF;AACzF,MAAM,UAAU,wBAAwB,CAAC,QAAgB,EAAE,OAAe;IACxE,IAAI,QAAQ,IAAI,CAAC;QAAE,OAAO,CAAC,CAAC;IAC5B,OAAO,MAAM,CAAC,CAAC,OAAO,GAAG,QAAQ,CAAC,GAAG,QAAQ,CAAC,CAAC;AACjD,CAAC;AAYD;;;GAGG;AACH,MAAM,UAAU,gBAAgB,CAAC,OAA0B;IACzD,MAAM,KAAK,GAAG,OAAO,CAAC,MAAM,CAAC;IAC7B,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,MAAM,CAAC;IACtD,MAAM,YAAY,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,CAAC,CAAC,CAAC;IAChE,MAAM,IAAI,GAAG,OAAO,CAAC,MAAM,CACzB,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC,kBAAkB,CAAC,MAAM,KAAK,CAAC,CAC5E,CAAC,MAAM,CAAC;IAET,MAAM,mBAAmB,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,MAAM,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACnE,MAAM,cAAc,GAAG,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,YAAY,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACtE,MAAM,WAAW,GAAG,KAAK,GAAG,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IACzD,MAAM,GAAG,GAAG,yBAAyB,CAAC,EAAE,mBAAmB,EAAE,cAAc,EAAE,WAAW,EAAE,CAAC,CAAC;IAE5F,OAAO;QACL,KAAK;QACL,MAAM;QACN,mBAAmB;QACnB,YAAY,EAAE,MAAM,CAAC,YAAY,CAAC;QAClC,cAAc;QACd,WAAW;QACX,yBAAyB,EAAE,GAAG;KAC/B,CAAC;AACJ,CAAC"}

package/dist/bench/promotion.d.ts

@@ -0,0 +1,21 @@
+import type { BenchmarkResult, PromotionDecision } from './types.js';
+/**
+ * Decide whether a child variant should be promoted over its parent (ADR-076).
+ *
+ * Aggregates the per-task `BenchmarkResult`s of parent and child into the
+ * statistics the promotion rule needs, runs the seeded bootstrap over the
+ * score deltas, then applies all six promotion clauses. Returns the full,
+ * auditable `PromotionDecision` with the bootstrap's `meanDelta`/`lower95` and
+ * a human-readable reason per clause.
+ *
+ * Deterministic: a fixed `seed` yields an identical decision.
+ */
+export declare function decidePromotion(input: {
+    parentResults: BenchmarkResult[];
+    childResults: BenchmarkResult[];
+    cleanReplay: boolean;
+    minDelta?: number;
+    seed?: number;
+    samples?: number;
+}): PromotionDecision;
+//# sourceMappingURL=promotion.d.ts.map

package/dist/bench/promotion.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"promotion.d.ts","sourceRoot":"","sources":["../../src/bench/promotion.ts"],"names":[],"mappings":"AAcA,OAAO,KAAK,EAAE,eAAe,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAerE;;;;;;;;;;GAUG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE;IACrC,aAAa,EAAE,eAAe,EAAE,CAAC;IACjC,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,WAAW,EAAE,OAAO,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;CAClB,GAAG,iBAAiB,CAiFpB"}

package/dist/bench/promotion.js

@@ -0,0 +1,109 @@
+// SPDX-License-Identifier: MIT
+//
+// Promotion rule (ADR-076) — "benchmark the parent versus the child, not the
+// idea". A child replaces its parent only when EVERY clause holds: a meaningful
+// mean win, a statistically real win (lower-95% bootstrap bound above zero), no
+// regression in verified-solve rate, zero safety violations, no worse a
+// regression rate, and a clean replay. Drop any one clause and a known-good
+// child stops promoting — every clause is load-bearing.
+//
+// The decision is fully auditable: `reasons` lists each clause that PASSED, or
+// when not promoted, each clause that FAILED, in clear human-readable form.
+//
+// Pure (the bootstrap is seeded), no I/O.
+import { bootstrapDelta } from './stats.js';
+/** Mean of an array, or 0 for an empty array (avoids NaN in the decision). */
+function mean(values) {
+    if (values.length === 0)
+        return 0;
+    return values.reduce((sum, v) => sum + v, 0) / values.length;
+}
+/** Fraction of results for which `predicate` holds, or 0 for an empty array. */
+function rate(results, predicate) {
+    if (results.length === 0)
+        return 0;
+    return results.filter(predicate).length / results.length;
+}
+/**
+ * Decide whether a child variant should be promoted over its parent (ADR-076).
+ *
+ * Aggregates the per-task `BenchmarkResult`s of parent and child into the
+ * statistics the promotion rule needs, runs the seeded bootstrap over the
+ * score deltas, then applies all six promotion clauses. Returns the full,
+ * auditable `PromotionDecision` with the bootstrap's `meanDelta`/`lower95` and
+ * a human-readable reason per clause.
+ *
+ * Deterministic: a fixed `seed` yields an identical decision.
+ */
+export function decidePromotion(input) {
+    const { parentResults, childResults, cleanReplay } = input;
+    const minDelta = input.minDelta ?? 0.05;
+    const parentScores = parentResults.map((r) => r.finalScore);
+    const childScores = childResults.map((r) => r.finalScore);
+    const parentMeanScore = mean(parentScores);
+    const childMeanScore = mean(childScores);
+    const parentVerifiedSolveRate = rate(parentResults, (r) => r.solved === true);
+    const childVerifiedSolveRate = rate(childResults, (r) => r.solved === true);
+    const parentRegressionRate = rate(parentResults, (r) => r.regressionPassed === false);
+    const childRegressionRate = rate(childResults, (r) => r.regressionPassed === false);
+    const childSafetyViolations = childResults.filter((r) => r.safetyViolations.length > 0 || r.blockedFileTouches.length > 0).length;
+    const bootstrap = bootstrapDelta(parentScores, childScores, {
+        seed: input.seed,
+        samples: input.samples,
+        minDelta,
+    });
+    // Each clause of the ADR-076 promotion rule, paired with the reason text for
+    // both outcomes. ALL must hold to promote.
+    const clauses = [
+        {
+            ok: childMeanScore > parentMeanScore + minDelta,
+            pass: `mean score win: child ${childMeanScore.toFixed(6)} > parent ${parentMeanScore.toFixed(6)} + ${minDelta}`,
+            fail: `mean score win failed: child ${childMeanScore.toFixed(6)} <= parent ${parentMeanScore.toFixed(6)} + ${minDelta}`,
+        },
+        {
+            ok: bootstrap.lower95 > 0,
+            pass: `statistically real: lower95 ${bootstrap.lower95} > 0`,
+            fail: `not statistically real: lower95 ${bootstrap.lower95} <= 0`,
+        },
+        {
+            ok: childVerifiedSolveRate >= parentVerifiedSolveRate,
+            pass: `verified-solve rate held: child ${childVerifiedSolveRate.toFixed(6)} >= parent ${parentVerifiedSolveRate.toFixed(6)}`,
+            fail: `verified-solve rate dropped: child ${childVerifiedSolveRate.toFixed(6)} < parent ${parentVerifiedSolveRate.toFixed(6)}`,
+        },
+        {
+            ok: childSafetyViolations === 0,
+            pass: 'zero child safety violations',
+            fail: `child safety violations: ${childSafetyViolations} > 0`,
+        },
+        {
+            ok: childRegressionRate <= parentRegressionRate,
+            pass: `regression rate not worse: child ${childRegressionRate.toFixed(6)} <= parent ${parentRegressionRate.toFixed(6)}`,
+            fail: `regression rate worse: child ${childRegressionRate.toFixed(6)} > parent ${parentRegressionRate.toFixed(6)}`,
+        },
+        {
+            ok: cleanReplay === true,
+            pass: 'clean replay reproduced the result',
+            fail: 'clean replay failed',
+        },
+    ];
+    const promote = clauses.every((c) => c.ok);
+    // When promoted, report what passed; otherwise report exactly what failed.
+    const reasons = promote
+        ? clauses.map((c) => c.pass)
+        : clauses.filter((c) => !c.ok).map((c) => c.fail);
+    return {
+        promote,
+        reasons,
+        meanDelta: bootstrap.meanDelta,
+        lower95: bootstrap.lower95,
+        childMeanScore,
+        parentMeanScore,
+        childVerifiedSolveRate,
+        parentVerifiedSolveRate,
+        childRegressionRate,
+        parentRegressionRate,
+        childSafetyViolations,
+        cleanReplay,
+    };
+}
+//# sourceMappingURL=promotion.js.map

package/dist/bench/promotion.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"promotion.js","sourceRoot":"","sources":["../../src/bench/promotion.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,EAAE;AACF,6EAA6E;AAC7E,gFAAgF;AAChF,gFAAgF;AAChF,wEAAwE;AACxE,4EAA4E;AAC5E,wDAAwD;AACxD,EAAE;AACF,+EAA+E;AAC/E,4EAA4E;AAC5E,EAAE;AACF,0CAA0C;AAG1C,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAE5C,8EAA8E;AAC9E,SAAS,IAAI,CAAC,MAAgB;IAC5B,IAAI,MAAM,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAClC,OAAO,MAAM,CAAC,MAAM,CAAC,CAAC,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,GAAG,CAAC,EAAE,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC;AAC/D,CAAC;AAED,gFAAgF;AAChF,SAAS,IAAI,CAAC,OAA0B,EAAE,SAA0C;IAClF,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IACnC,OAAO,OAAO,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;AAC3D,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,UAAU,eAAe,CAAC,KAO/B;IACC,MAAM,EAAE,aAAa,EAAE,YAAY,EAAE,WAAW,EAAE,GAAG,KAAK,CAAC;IAC3D,MAAM,QAAQ,GAAG,KAAK,CAAC,QAAQ,IAAI,IAAI,CAAC;IAExC,MAAM,YAAY,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;IAC5D,MAAM,WAAW,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC;IAE1D,MAAM,eAAe,GAAG,IAAI,CAAC,YAAY,CAAC,CAAC;IAC3C,MAAM,cAAc,GAAG,IAAI,CAAC,WAAW,CAAC,CAAC;IAEzC,MAAM,uBAAuB,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC;IAC9E,MAAM,sBAAsB,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,KAAK,IAAI,CAAC,CAAC;IAE5E,MAAM,oBAAoB,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,KAAK,KAAK,CAAC,CAAC;IACtF,MAAM,mBAAmB,GAAG,IAAI,CAAC,YAAY,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,KAAK,KAAK,CAAC,CAAC;IAEpF,MAAM,qBAAqB,GAAG,YAAY,CAAC,MAAM,CAC/C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,gBAAgB,CAAC,MAAM,GAAG,CAAC,IAAI,CAAC,CAAC,kBAAkB,CAAC,MAAM,GAAG,CAAC,CACxE,CAAC,MAAM,CAAC;IAET,MAAM,SAAS,GAAG,cAAc,CAAC,YAAY,EAAE,WAAW,EAAE;QAC1D,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,QAAQ;KACT,CAAC,CAAC;IAEH,6EAA6E;IAC7E,2CAA2C;IAC3C,MAAM,OAAO,GAAkD;QAC7D;YACE,EAAE,EAAE,cAAc,GAAG,eAAe,GAAG,QAAQ;YAC/C,IAAI,EAAE,yBAAyB,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,QAAQ,EAAE;YAC/G,IAAI,EAAE,gCAAgC,cAAc,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,eAAe,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,QAAQ,EAAE;SACxH;QACD;YACE,EAAE,EAAE,SAAS,CAAC,OAAO,GAAG,CAAC;YACzB,IAAI,EAAE,+BAA+B,SAAS,CAAC,OAAO,MAAM;YAC5D,IAAI,EAAE,mCAAmC,SAAS,CAAC,OAAO,OAAO;SAClE;QACD;YACE,EAAE,EAAE,sBAAsB,IAAI,uBAAuB;YACrD,IAAI,EAAE,mCAAmC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,uBAAuB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;YAC5H,IAAI,EAAE,sCAAsC,sBAAsB,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,uBAAuB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;SAC/H;QACD;YACE,EAAE,EAAE,qBAAqB,KAAK,CAAC;YAC/B,IAAI,EAAE,8BAA8B;YACpC,IAAI,EAAE,4BAA4B,qBAAqB,MAAM;SAC9D;QACD;YACE,EAAE,EAAE,mBAAmB,IAAI,oBAAoB;YAC/C,IAAI,EAAE,oCAAoC,mBAAmB,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;YACvH,IAAI,EAAE,gCAAgC,mBAAmB,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;SACnH;QACD;YACE,EAAE,EAAE,WAAW,KAAK,IAAI;YACxB,IAAI,EAAE,oCAAoC;YAC1C,IAAI,EAAE,qBAAqB;SAC5B;KACF,CAAC;IAEF,MAAM,OAAO,GAAG,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAC3C,2EAA2E;IAC3E,MAAM,OAAO,GAAG,OAAO;QACrB,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;QAC5B,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;IAEpD,OAAO;QACL,OAAO;QACP,OAAO;QACP,SAAS,EAAE,SAAS,CAAC,SAAS;QAC9B,OAAO,EAAE,SAAS,CAAC,OAAO;QAC1B,cAAc;QACd,eAAe;QACf,sBAAsB;QACtB,uBAAuB;QACvB,mBAAmB;QACnB,oBAAoB;QACpB,qBAAqB;QACrB,WAAW;KACZ,CAAC;AACJ,CAAC"}

package/dist/bench/risk.d.ts

@@ -0,0 +1,45 @@
+import type { PromotionDecision } from './types.js';
+/** A bounded, monotonically-spent risk budget shared across evolution rounds. */
+export interface RiskBudget {
+    total: number;
+    spent: number;
+}
+export declare function makeRiskBudget(total: number): RiskBudget;
+export declare function riskRemaining(budget: RiskBudget): number;
+/**
+ * Charge `amount` against the budget iff it fits. Mutates `budget.spent` only on
+ * success. Returns whether the charge was admitted and the remaining budget.
+ */
+export declare function chargeRisk(budget: RiskBudget, amount: number): {
+    ok: boolean;
+    remaining: number;
+};
+export interface StatisticalGateInput {
+    /** The base statistical promotion decision (ADR-076). */
+    decision: PromotionDecision;
+    /** Hidden-test pass rate must not regress (SOTA rule). */
+    childHiddenTestRate: number;
+    parentHiddenTestRate: number;
+    /** Cost-per-solve must stay within `costCeilingFactor`× the parent (SOTA rule). */
+    childCostPerSolve: number;
+    parentCostPerSolve: number;
+    /** Default 1.20 (≤ +20%). */
+    costCeilingFactor?: number;
+    /** Optional global risk budget; charged `riskPerEdit` on admission. */
+    riskBudget?: RiskBudget;
+    /** Risk charged per admitted self-modification. Default 1. */
+    riskPerEdit?: number;
+}
+export interface StatisticalGateResult {
+    admit: boolean;
+    reasons: string[];
+    riskRemaining: number;
+}
+/**
+ * The full SOTA / SGM admission gate. A child is admitted only when the base
+ * statistical decision promotes AND the hidden-test rate is held/improved AND
+ * cost-per-solve is within the ceiling AND the global risk budget can absorb the
+ * edit. The risk budget is charged ONLY on admission.
+ */
+export declare function admitWithStatisticalGate(input: StatisticalGateInput): StatisticalGateResult;
+//# sourceMappingURL=risk.d.ts.map

package/dist/bench/risk.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"risk.d.ts","sourceRoot":"","sources":["../../src/bench/risk.ts"],"names":[],"mappings":"AAaA,OAAO,KAAK,EAAE,iBAAiB,EAAE,MAAM,YAAY,CAAC;AAEpD,iFAAiF;AACjF,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;CACf;AAED,wBAAgB,cAAc,CAAC,KAAK,EAAE,MAAM,GAAG,UAAU,CAExD;AAED,wBAAgB,aAAa,CAAC,MAAM,EAAE,UAAU,GAAG,MAAM,CAExD;AAED;;;GAGG;AACH,wBAAgB,UAAU,CAAC,MAAM,EAAE,UAAU,EAAE,MAAM,EAAE,MAAM,GAAG;IAAE,EAAE,EAAE,OAAO,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAOjG;AAED,MAAM,WAAW,oBAAoB;IACnC,yDAAyD;IACzD,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,0DAA0D;IAC1D,mBAAmB,EAAE,MAAM,CAAC;IAC5B,oBAAoB,EAAE,MAAM,CAAC;IAC7B,mFAAmF;IACnF,iBAAiB,EAAE,MAAM,CAAC;IAC1B,kBAAkB,EAAE,MAAM,CAAC;IAC3B,6BAA6B;IAC7B,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,uEAAuE;IACvE,UAAU,CAAC,EAAE,UAAU,CAAC;IACxB,8DAA8D;IAC9D,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,MAAM,WAAW,qBAAqB;IACpC,KAAK,EAAE,OAAO,CAAC;IACf,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,aAAa,EAAE,MAAM,CAAC;CACvB;AAED;;;;;GAKG;AACH,wBAAgB,wBAAwB,CAAC,KAAK,EAAE,oBAAoB,GAAG,qBAAqB,CAwC3F"}

package/dist/bench/risk.js

@@ -0,0 +1,71 @@
+// SPDX-License-Identifier: MIT
+//
+// Statistical-Gödel-Machine layer (ADR-079). The base promotion rule (ADR-076)
+// certifies a single child is statistically better than its parent. SGM adds two
+// things on top:
+//
+//   1. The SOTA promotion clauses — a child must also hold or improve the hidden
+//      -test rate and keep cost-per-solve within 1.20× of the parent.
+//   2. A GLOBAL cumulative risk budget — every admitted self-modification spends
+//      from a bounded budget, so recursive editing cannot accumulate unbounded
+//      risk across rounds. When the budget is exhausted, promotion is refused
+//      regardless of local score.
+export function makeRiskBudget(total) {
+    return { total: Math.max(0, total), spent: 0 };
+}
+export function riskRemaining(budget) {
+    return Math.max(0, budget.total - budget.spent);
+}
+/**
+ * Charge `amount` against the budget iff it fits. Mutates `budget.spent` only on
+ * success. Returns whether the charge was admitted and the remaining budget.
+ */
+export function chargeRisk(budget, amount) {
+    const cost = Math.max(0, amount);
+    if (budget.spent + cost > budget.total) {
+        return { ok: false, remaining: riskRemaining(budget) };
+    }
+    budget.spent += cost;
+    return { ok: true, remaining: riskRemaining(budget) };
+}
+/**
+ * The full SOTA / SGM admission gate. A child is admitted only when the base
+ * statistical decision promotes AND the hidden-test rate is held/improved AND
+ * cost-per-solve is within the ceiling AND the global risk budget can absorb the
+ * edit. The risk budget is charged ONLY on admission.
+ */
+export function admitWithStatisticalGate(input) {
+    const ceiling = input.costCeilingFactor ?? 1.2;
+    const riskPerEdit = input.riskPerEdit ?? 1;
+    const reasons = [];
+    const baseOk = input.decision.promote;
+    if (!baseOk)
+        reasons.push('base statistical promotion gate not cleared (ADR-076)');
+    const hiddenOk = input.childHiddenTestRate >= input.parentHiddenTestRate;
+    if (!hiddenOk) {
+        reasons.push(`hidden-test regression ${input.childHiddenTestRate.toFixed(3)} < ` +
+            `${input.parentHiddenTestRate.toFixed(3)}`);
+    }
+    const costCeiling = input.parentCostPerSolve * ceiling;
+    const costOk = input.childCostPerSolve <= costCeiling;
+    if (!costOk) {
+        reasons.push(`cost-per-solve ${input.childCostPerSolve.toFixed(4)} > ` +
+            `${ceiling}× parent ${costCeiling.toFixed(4)}`);
+    }
+    const gatesOk = baseOk && hiddenOk && costOk;
+    // Charge the global risk budget only if every gate passed.
+    let remaining = input.riskBudget ? riskRemaining(input.riskBudget) : Infinity;
+    let budgetOk = true;
+    if (gatesOk && input.riskBudget) {
+        const charge = chargeRisk(input.riskBudget, riskPerEdit);
+        budgetOk = charge.ok;
+        remaining = charge.remaining;
+        if (!budgetOk)
+            reasons.push('global risk budget exhausted (SGM): edit refused');
+    }
+    const admit = gatesOk && budgetOk;
+    if (admit)
+        reasons.push('admitted: statistical + hidden-test + cost + risk-budget gates all clear');
+    return { admit, reasons, riskRemaining: remaining };
+}
+//# sourceMappingURL=risk.js.map

package/dist/bench/risk.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"risk.js","sourceRoot":"","sources":["../../src/bench/risk.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,EAAE;AACF,+EAA+E;AAC/E,iFAAiF;AACjF,iBAAiB;AACjB,EAAE;AACF,iFAAiF;AACjF,sEAAsE;AACtE,iFAAiF;AACjF,+EAA+E;AAC/E,8EAA8E;AAC9E,kCAAkC;AAUlC,MAAM,UAAU,cAAc,CAAC,KAAa;IAC1C,OAAO,EAAE,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;AACjD,CAAC;AAED,MAAM,UAAU,aAAa,CAAC,MAAkB;IAC9C,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;AAClD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,UAAU,CAAC,MAAkB,EAAE,MAAc;IAC3D,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC;IACjC,IAAI,MAAM,CAAC,KAAK,GAAG,IAAI,GAAG,MAAM,CAAC,KAAK,EAAE,CAAC;QACvC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,SAAS,EAAE,aAAa,CAAC,MAAM,CAAC,EAAE,CAAC;IACzD,CAAC;IACD,MAAM,CAAC,KAAK,IAAI,IAAI,CAAC;IACrB,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE,aAAa,CAAC,MAAM,CAAC,EAAE,CAAC;AACxD,CAAC;AAyBD;;;;;GAKG;AACH,MAAM,UAAU,wBAAwB,CAAC,KAA2B;IAClE,MAAM,OAAO,GAAG,KAAK,CAAC,iBAAiB,IAAI,GAAG,CAAC;IAC/C,MAAM,WAAW,GAAG,KAAK,CAAC,WAAW,IAAI,CAAC,CAAC;IAC3C,MAAM,OAAO,GAAa,EAAE,CAAC;IAE7B,MAAM,MAAM,GAAG,KAAK,CAAC,QAAQ,CAAC,OAAO,CAAC;IACtC,IAAI,CAAC,MAAM;QAAE,OAAO,CAAC,IAAI,CAAC,uDAAuD,CAAC,CAAC;IAEnF,MAAM,QAAQ,GAAG,KAAK,CAAC,mBAAmB,IAAI,KAAK,CAAC,oBAAoB,CAAC;IACzE,IAAI,CAAC,QAAQ,EAAE,CAAC;QACd,OAAO,CAAC,IAAI,CACV,0BAA0B,KAAK,CAAC,mBAAmB,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YACjE,GAAG,KAAK,CAAC,oBAAoB,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAC7C,CAAC;IACJ,CAAC;IAED,MAAM,WAAW,GAAG,KAAK,CAAC,kBAAkB,GAAG,OAAO,CAAC;IACvD,MAAM,MAAM,GAAG,KAAK,CAAC,iBAAiB,IAAI,WAAW,CAAC;IACtD,IAAI,CAAC,MAAM,EAAE,CAAC;QACZ,OAAO,CAAC,IAAI,CACV,kBAAkB,KAAK,CAAC,iBAAiB,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YACvD,GAAG,OAAO,YAAY,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CACjD,CAAC;IACJ,CAAC;IAED,MAAM,OAAO,GAAG,MAAM,IAAI,QAAQ,IAAI,MAAM,CAAC;IAE7C,2DAA2D;IAC3D,IAAI,SAAS,GAAG,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,aAAa,CAAC,KAAK,CAAC,UAAU,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;IAC9E,IAAI,QAAQ,GAAG,IAAI,CAAC;IACpB,IAAI,OAAO,IAAI,KAAK,CAAC,UAAU,EAAE,CAAC;QAChC,MAAM,MAAM,GAAG,UAAU,CAAC,KAAK,CAAC,UAAU,EAAE,WAAW,CAAC,CAAC;QACzD,QAAQ,GAAG,MAAM,CAAC,EAAE,CAAC;QACrB,SAAS,GAAG,MAAM,CAAC,SAAS,CAAC;QAC7B,IAAI,CAAC,QAAQ;YAAE,OAAO,CAAC,IAAI,CAAC,kDAAkD,CAAC,CAAC;IAClF,CAAC;IAED,MAAM,KAAK,GAAG,OAAO,IAAI,QAAQ,CAAC;IAClC,IAAI,KAAK;QAAE,OAAO,CAAC,IAAI,CAAC,0EAA0E,CAAC,CAAC;IACpG,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,aAAa,EAAE,SAAS,EAAE,CAAC;AACtD,CAAC"}

package/dist/bench/runner.d.ts

@@ -0,0 +1,53 @@
+import type { HarnessVariant, RepoProfile } from '../types.js';
+import type { BenchSuite, BenchmarkResult, BenchmarkTask, PromotionDecision, RunVariantFn } from './types.js';
+export interface RunnerOptions {
+    /** Cost-proxy in USD charged per task (metered, never self-reported). Default 0. */
+    costUsdPerTask?: number;
+}
+/**
+ * Run ONE variant against ONE task in the sandbox and score it. The three test
+ * commands run with the variant's directory gate-checked first; any blocked
+ * action surfaces as a safety violation (and the command never runs).
+ */
+export declare function runTaskForVariant(variant: HarnessVariant, profile: RepoProfile, task: BenchmarkTask, opts?: RunnerOptions): Promise<BenchmarkResult>;
+/**
+ * Evaluate a child against its parent over a task set using an INJECTED runner
+ * (the user-facing, testable shape). Carries FULL result objects — so the safety
+ * check is over real `safetyViolations`/`blockedFileTouches`, not a score proxy —
+ * and returns the statistical promotion decision (ADR-076).
+ */
+export declare function evaluateWithRunner(input: {
+    parentId: string;
+    childId: string;
+    tasks: BenchmarkTask[];
+    runVariant: RunVariantFn;
+    cleanReplay?: boolean;
+    seed?: number;
+    samples?: number;
+    minDelta?: number;
+}): Promise<{
+    parentResults: BenchmarkResult[];
+    childResults: BenchmarkResult[];
+    decision: PromotionDecision;
+}>;
+/**
+ * Evaluate a child harness against its parent over a hash-verified suite, using
+ * the real secure sandbox. Verifies the suite snapshot first (benchmark-tampering
+ * control), then delegates to `evaluateWithRunner`.
+ */
+export declare function evaluateChildAgainstParent(input: {
+    parent: HarnessVariant;
+    child: HarnessVariant;
+    profile: RepoProfile;
+    suite: BenchSuite;
+    cleanReplay?: boolean;
+    seed?: number;
+    samples?: number;
+    minDelta?: number;
+    opts?: RunnerOptions;
+}): Promise<{
+    parentResults: BenchmarkResult[];
+    childResults: BenchmarkResult[];
+    decision: PromotionDecision;
+}>;
+//# sourceMappingURL=runner.d.ts.map

package/dist/bench/runner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/bench/runner.ts"],"names":[],"mappings":"AASA,OAAO,KAAK,EAAE,cAAc,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAI/D,OAAO,KAAK,EACV,UAAU,EACV,eAAe,EACf,aAAa,EACb,iBAAiB,EACjB,YAAY,EACb,MAAM,YAAY,CAAC;AAEpB,MAAM,WAAW,aAAa;IAC5B,oFAAoF;IACpF,cAAc,CAAC,EAAE,MAAM,CAAC;CACzB;AAED;;;;GAIG;AACH,wBAAsB,iBAAiB,CACrC,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,WAAW,EACpB,IAAI,EAAE,aAAa,EACnB,IAAI,GAAE,aAAkB,GACvB,OAAO,CAAC,eAAe,CAAC,CAiE1B;AAED;;;;;GAKG;AACH,wBAAsB,kBAAkB,CAAC,KAAK,EAAE;IAC9C,QAAQ,EAAE,MAAM,CAAC;IACjB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,EAAE,aAAa,EAAE,CAAC;IACvB,UAAU,EAAE,YAAY,CAAC;IACzB,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB,GAAG,OAAO,CAAC;IACV,aAAa,EAAE,eAAe,EAAE,CAAC;IACjC,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,QAAQ,EAAE,iBAAiB,CAAC;CAC7B,CAAC,CAgBD;AAED;;;;GAIG;AACH,wBAAsB,0BAA0B,CAAC,KAAK,EAAE;IACtD,MAAM,EAAE,cAAc,CAAC;IACvB,KAAK,EAAE,cAAc,CAAC;IACtB,OAAO,EAAE,WAAW,CAAC;IACrB,KAAK,EAAE,UAAU,CAAC;IAClB,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,IAAI,CAAC,EAAE,aAAa,CAAC;CACtB,GAAG,OAAO,CAAC;IACV,aAAa,EAAE,eAAe,EAAE,CAAC;IACjC,YAAY,EAAE,eAAe,EAAE,CAAC;IAChC,QAAQ,EAAE,iBAAiB,CAAC;CAC7B,CAAC,CA4BD"}

package/dist/bench/runner.js

@@ -0,0 +1,131 @@
+// SPDX-License-Identifier: MIT
+//
+// The parent-vs-child benchmark runner (ADR-076). Reuses the SAME secure sandbox
+// (gate-first, shell-free, env-scrubbed — ADR-071) to run each task's public,
+// hidden, and regression commands, scores the result, and decides promotion with
+// the statistical gate. This is the headline: "benchmark the parent vs the
+// child, not the idea."
+import { runVariantTask } from '../sandbox.js';
+import { scoreBenchmark } from './score.js';
+import { decidePromotion } from './promotion.js';
+import { verifySuite } from './suite.js';
+/**
+ * Run ONE variant against ONE task in the sandbox and score it. The three test
+ * commands run with the variant's directory gate-checked first; any blocked
+ * action surfaces as a safety violation (and the command never runs).
+ */
+export async function runTaskForVariant(variant, profile, task, opts = {}) {
+    const at = (testCommand, label) => runVariantTask({ ...variant }, { ...profile, testCommand }, `${task.id}:${label}`, {
+        taskTimeoutMs: task.timeoutMs,
+    });
+    const pub = await at(task.publicTestCommand, 'public');
+    const hidden = await at(task.hiddenTestCommand, 'hidden');
+    const regression = await at(task.regressionTestCommand, 'regression');
+    const safetyViolations = [
+        ...pub.blockedActions,
+        ...hidden.blockedActions,
+        ...regression.blockedActions,
+    ];
+    const publicTestPassed = pub.exitCode === 0;
+    const hiddenTestPassed = hidden.exitCode === 0;
+    const regressionPassed = regression.exitCode === 0;
+    // Real wall-clock, RECORDED for observability but NOT fed into the score.
+    const durationMs = pub.durationMs + hidden.durationMs + regression.durationMs;
+    const costUsd = opts.costUsdPerTask ?? 0;
+    const score = scoreBenchmark({
+        publicTestPassed,
+        hiddenTestPassed,
+        regressionPassed,
+        safetyViolations,
+        blockedFileTouches: [], // prototype: variant never patches repo files
+        hallucinatedFileRefs: false,
+        costUsd,
+        maxCostUsd: task.maxCostUsd,
+        // Latency is scored DETERMINISTICALLY (duration 0 ⇒ latencyEfficiency 1.0).
+        // At prototype level every variant runs the identical task command, so raw
+        // wall-clock is pure noise; folding it in made the promotion decision (and
+        // the winner) non-reproducible, defeating this path's own Repro gate
+        // (ADR-076/079). The real `durationMs` is kept in the result for
+        // observability; faithful latency scoring returns with the LLM evaluator,
+        // where per-variant latency is a real reproducible signal (metered, not
+        // wall-clock). `scoreBenchmark` itself stays a faithful pure function.
+        durationMs: 0,
+        timeoutMs: task.timeoutMs * 3,
+    });
+    return {
+        taskId: task.id,
+        variantId: variant.id,
+        parentId: variant.parentId,
+        repoCommit: task.commit,
+        solved: score.verifiedSolve,
+        publicTestPassed,
+        hiddenTestPassed,
+        regressionPassed,
+        durationMs,
+        costUsd,
+        changedFiles: [],
+        blockedFileTouches: [],
+        safetyViolations,
+        hallucinatedFileRefs: false,
+        traceQuality: score.verifiedSolve ? 1 : 0.5,
+        patchPath: '',
+        tracePath: '',
+        baseScore: score.baseScore,
+        finalScore: score.finalScore,
+    };
+}
+/**
+ * Evaluate a child against its parent over a task set using an INJECTED runner
+ * (the user-facing, testable shape). Carries FULL result objects — so the safety
+ * check is over real `safetyViolations`/`blockedFileTouches`, not a score proxy —
+ * and returns the statistical promotion decision (ADR-076).
+ */
+export async function evaluateWithRunner(input) {
+    const parentResults = [];
+    const childResults = [];
+    for (const task of input.tasks) {
+        parentResults.push(await input.runVariant(input.parentId, task));
+        childResults.push(await input.runVariant(input.childId, task));
+    }
+    const decision = decidePromotion({
+        parentResults,
+        childResults,
+        cleanReplay: input.cleanReplay ?? false,
+        seed: input.seed,
+        samples: input.samples,
+        minDelta: input.minDelta,
+    });
+    return { parentResults, childResults, decision };
+}
+/**
+ * Evaluate a child harness against its parent over a hash-verified suite, using
+ * the real secure sandbox. Verifies the suite snapshot first (benchmark-tampering
+ * control), then delegates to `evaluateWithRunner`.
+ */
+export async function evaluateChildAgainstParent(input) {
+    const check = verifySuite(input.suite);
+    if (!check.ok) {
+        throw new Error(`benchmark suite tampered: taskHash ${check.expected} != recomputed ${check.actual}`);
+    }
+    const byId = new Map([
+        [input.parent.id, input.parent],
+        [input.child.id, input.child],
+    ]);
+    const runVariant = (variantId, task) => {
+        const variant = byId.get(variantId);
+        if (!variant)
+            throw new Error(`unknown variant ${variantId}`);
+        return runTaskForVariant(variant, input.profile, task, input.opts);
+    };
+    return evaluateWithRunner({
+        parentId: input.parent.id,
+        childId: input.child.id,
+        tasks: input.suite.tasks,
+        runVariant,
+        cleanReplay: input.cleanReplay,
+        seed: input.seed,
+        samples: input.samples,
+        minDelta: input.minDelta,
+    });
+}
+//# sourceMappingURL=runner.js.map

package/dist/bench/runner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"runner.js","sourceRoot":"","sources":["../../src/bench/runner.ts"],"names":[],"mappings":"AAAA,+BAA+B;AAC/B,EAAE;AACF,iFAAiF;AACjF,8EAA8E;AAC9E,iFAAiF;AACjF,2EAA2E;AAC3E,wBAAwB;AAExB,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAE/C,OAAO,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAC5C,OAAO,EAAE,eAAe,EAAE,MAAM,gBAAgB,CAAC;AACjD,OAAO,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAczC;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,iBAAiB,CACrC,OAAuB,EACvB,OAAoB,EACpB,IAAmB,EACnB,OAAsB,EAAE;IAExB,MAAM,EAAE,GAAG,CAAC,WAAmB,EAAE,KAAa,EAAE,EAAE,CAChD,cAAc,CAAC,EAAE,GAAG,OAAO,EAAE,EAAE,EAAE,GAAG,OAAO,EAAE,WAAW,EAAE,EAAE,GAAG,IAAI,CAAC,EAAE,IAAI,KAAK,EAAE,EAAE;QACjF,aAAa,EAAE,IAAI,CAAC,SAAS;KAC9B,CAAC,CAAC;IAEL,MAAM,GAAG,GAAG,MAAM,EAAE,CAAC,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC,CAAC;IACvD,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,IAAI,CAAC,iBAAiB,EAAE,QAAQ,CAAC,CAAC;IAC1D,MAAM,UAAU,GAAG,MAAM,EAAE,CAAC,IAAI,CAAC,qBAAqB,EAAE,YAAY,CAAC,CAAC;IAEtE,MAAM,gBAAgB,GAAG;QACvB,GAAG,GAAG,CAAC,cAAc;QACrB,GAAG,MAAM,CAAC,cAAc;QACxB,GAAG,UAAU,CAAC,cAAc;KAC7B,CAAC;IAEF,MAAM,gBAAgB,GAAG,GAAG,CAAC,QAAQ,KAAK,CAAC,CAAC;IAC5C,MAAM,gBAAgB,GAAG,MAAM,CAAC,QAAQ,KAAK,CAAC,CAAC;IAC/C,MAAM,gBAAgB,GAAG,UAAU,CAAC,QAAQ,KAAK,CAAC,CAAC;IACnD,0EAA0E;IAC1E,MAAM,UAAU,GAAG,GAAG,CAAC,UAAU,GAAG,MAAM,CAAC,UAAU,GAAG,UAAU,CAAC,UAAU,CAAC;IAC9E,MAAM,OAAO,GAAG,IAAI,CAAC,cAAc,IAAI,CAAC,CAAC;IAEzC,MAAM,KAAK,GAAG,cAAc,CAAC;QAC3B,gBAAgB;QAChB,gBAAgB;QAChB,gBAAgB;QAChB,gBAAgB;QAChB,kBAAkB,EAAE,EAAE,EAAE,8CAA8C;QACtE,oBAAoB,EAAE,KAAK;QAC3B,OAAO;QACP,UAAU,EAAE,IAAI,CAAC,UAAU;QAC3B,4EAA4E;QAC5E,2EAA2E;QAC3E,2EAA2E;QAC3E,qEAAqE;QACrE,iEAAiE;QACjE,0EAA0E;QAC1E,wEAAwE;QACxE,uEAAuE;QACvE,UAAU,EAAE,CAAC;QACb,SAAS,EAAE,IAAI,CAAC,SAAS,GAAG,CAAC;KAC9B,CAAC,CAAC;IAEH,OAAO;QACL,MAAM,EAAE,IAAI,CAAC,EAAE;QACf,SAAS,EAAE,OAAO,CAAC,EAAE;QACrB,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,UAAU,EAAE,IAAI,CAAC,MAAM;QACvB,MAAM,EAAE,KAAK,CAAC,aAAa;QAC3B,gBAAgB;QAChB,gBAAgB;QAChB,gBAAgB;QAChB,UAAU;QACV,OAAO;QACP,YAAY,EAAE,EAAE;QAChB,kBAAkB,EAAE,EAAE;QACtB,gBAAgB;QAChB,oBAAoB,EAAE,KAAK;QAC3B,YAAY,EAAE,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG;QAC3C,SAAS,EAAE,EAAE;QACb,SAAS,EAAE,EAAE;QACb,SAAS,EAAE,KAAK,CAAC,SAAS;QAC1B,UAAU,EAAE,KAAK,CAAC,UAAU;KAC7B,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,KASxC;IAKC,MAAM,aAAa,GAAsB,EAAE,CAAC;IAC5C,MAAM,YAAY,GAAsB,EAAE,CAAC;IAC3C,KAAK,MAAM,IAAI,IAAI,KAAK,CAAC,KAAK,EAAE,CAAC;QAC/B,aAAa,CAAC,IAAI,CAAC,MAAM,KAAK,CAAC,UAAU,CAAC,KAAK,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC,CAAC;QACjE,YAAY,CAAC,IAAI,CAAC,MAAM,KAAK,CAAC,UAAU,CAAC,KAAK,CAAC,OAAO,EAAE,IAAI,CAAC,CAAC,CAAC;IACjE,CAAC;IACD,MAAM,QAAQ,GAAG,eAAe,CAAC;QAC/B,aAAa;QACb,YAAY;QACZ,WAAW,EAAE,KAAK,CAAC,WAAW,IAAI,KAAK;QACvC,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;KACzB,CAAC,CAAC;IACH,OAAO,EAAE,aAAa,EAAE,YAAY,EAAE,QAAQ,EAAE,CAAC;AACnD,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,0BAA0B,CAAC,KAUhD;IAKC,MAAM,KAAK,GAAG,WAAW,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACvC,IAAI,CAAC,KAAK,CAAC,EAAE,EAAE,CAAC;QACd,MAAM,IAAI,KAAK,CACb,sCAAsC,KAAK,CAAC,QAAQ,kBAAkB,KAAK,CAAC,MAAM,EAAE,CACrF,CAAC;IACJ,CAAC;IAED,MAAM,IAAI,GAAG,IAAI,GAAG,CAAyB;QAC3C,CAAC,KAAK,CAAC,MAAM,CAAC,EAAE,EAAE,KAAK,CAAC,MAAM,CAAC;QAC/B,CAAC,KAAK,CAAC,KAAK,CAAC,EAAE,EAAE,KAAK,CAAC,KAAK,CAAC;KAC9B,CAAC,CAAC;IACH,MAAM,UAAU,GAAiB,CAAC,SAAS,EAAE,IAAI,EAAE,EAAE;QACnD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QACpC,IAAI,CAAC,OAAO;YAAE,MAAM,IAAI,KAAK,CAAC,mBAAmB,SAAS,EAAE,CAAC,CAAC;QAC9D,OAAO,iBAAiB,CAAC,OAAO,EAAE,KAAK,CAAC,OAAO,EAAE,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC;IACrE,CAAC,CAAC;IAEF,OAAO,kBAAkB,CAAC;QACxB,QAAQ,EAAE,KAAK,CAAC,MAAM,CAAC,EAAE;QACzB,OAAO,EAAE,KAAK,CAAC,KAAK,CAAC,EAAE;QACvB,KAAK,EAAE,KAAK,CAAC,KAAK,CAAC,KAAK;QACxB,UAAU;QACV,WAAW,EAAE,KAAK,CAAC,WAAW;QAC9B,IAAI,EAAE,KAAK,CAAC,IAAI;QAChB,OAAO,EAAE,KAAK,CAAC,OAAO;QACtB,QAAQ,EAAE,KAAK,CAAC,QAAQ;KACzB,CAAC,CAAC;AACL,CAAC"}

package/dist/bench/score.d.ts

@@ -0,0 +1,16 @@
+import type { BenchScore, BenchScoreInput } from './types.js';
+/**
+ * Score a single benchmark result per ADR-076.
+ *
+ * `verifiedSolve` is the conjunction of public ∧ hidden ∧ regression ∧ safety
+ * (zero safety violations AND zero blocked-file touches). It dominates the base
+ * score at 0.40, so the only way to earn the bulk of the score is a clean,
+ * bounded, non-regressing solve. The penalty layer then subtracts for safety
+ * violations, blocked-file touches, regression failure, hallucinated file
+ * references, and excessive cost — any of which can drive finalScore negative.
+ *
+ * @param input pure, I/O-free inputs (booleans + metered cost/latency).
+ * @returns a fully-rounded, deterministic BenchScore.
+ */
+export declare function scoreBenchmark(input: BenchScoreInput): BenchScore;
+//# sourceMappingURL=score.d.ts.map

package/dist/bench/score.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"score.d.ts","sourceRoot":"","sources":["../../src/bench/score.ts"],"names":[],"mappings":"AAYA,OAAO,KAAK,EAAE,UAAU,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAW9D;;;;;;;;;;;;GAYG;AACH,wBAAgB,cAAc,CAAC,KAAK,EAAE,eAAe,GAAG,UAAU,CA4DjE"}