@miller-tech/uap 1.136.2 → 1.137.2

package/dist/benchmarks/paired/report.d.ts

@@ -19,6 +19,8 @@ export interface ConditionSummary {
     meanCostUsd: number | null;
     meanLatencyMs: number;
     errorRate: number;
+    /** Mean composite quality (0-100) over runs that carried a quality score, or null. */
+    meanQuality: number | null;
 }
 export interface Comparison {
     label: string;
@@ -36,6 +38,14 @@ export interface Comparison {
     metrics: Partial<Record<ContinuousMetric, PairedDeltaResult>>;
     /** Per-metric win/tie/loss (lower-is-better for tokens/cost/turns/latency). */
     metricVerdicts: Partial<Record<ContinuousMetric, Verdict>>;
+    /**
+     * Paired composite-quality delta (treatment - baseline over cells where BOTH
+     * arms carried a quality score), higher-is-better. Present only when quality
+     * scores exist. This is the LLM-Self-Tuning signal the tuner optimizes.
+     */
+    quality?: PairedDeltaResult;
+    /** Win/tie/loss on composite quality (higher-is-better). */
+    qualityVerdict?: Verdict;
 }
 export interface AnalysisReport {
     meta: {
@@ -66,6 +76,9 @@ export interface AnalyzeOptions extends PairedOptions {
     /** Per-metric ROPE half-widths (same units as the metric). The open-challenge
      *  "deltas <4 TPS are ties" norm is a per-metric margin. */
     metricMargins?: Partial<Record<ContinuousMetric, number>>;
+    /** ROPE half-width (composite-quality points, 0..100) below which a quality
+     *  delta is a tie even if statistically significant. Default 0. */
+    qualityMargin?: number;
 }
 export declare function analyze(output: RunnerOutput, opts?: AnalyzeOptions): AnalysisReport;
 export declare function renderMarkdown(r: AnalysisReport): string;

package/dist/benchmarks/paired/report.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/paired/report.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAGL,aAAa,EAGb,iBAAiB,EACjB,aAAa,EAEb,OAAO,EACR,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC3C,OAAO,EAAsB,gBAAgB,EAA2B,MAAM,YAAY,CAAC;AAE3F,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,CAAC,EAAE,MAAM,CAAC;IACV,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;CACnB;AAED,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,WAAW,EAAE;QACX,YAAY,EAAE,MAAM,CAAC;QACrB,aAAa,EAAE,MAAM,CAAC;QACtB,KAAK,EAAE,iBAAiB,CAAC;QACzB,OAAO,EAAE,aAAa,CAAC;QACvB,yEAAyE;QACzE,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC;IACF,iFAAiF;IACjF,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC9D,+EAA+E;IAC/E,cAAc,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,OAAO,CAAC,CAAC,CAAC;CAC5D;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE;QACJ,KAAK,EAAE,MAAM,CAAC;QACd,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,EAAE,CAAC;QACrB,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACpB,CAAC;IACF,YAAY,EAAE,gBAAgB,EAAE,CAAC;IACjC,WAAW,EAAE,UAAU,EAAE,CAAC;IAC1B,uDAAuD;IACvD,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,EAAE,CAAC;CAC7E;AA2DD,MAAM,WAAW,cAAe,SAAQ,aAAa;IACnD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;2EAEuE;IACvE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;gEAC4D;IAC5D,aAAa,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;CAC3D;AAED,wBAAgB,OAAO,CAAC,MAAM,EAAE,YAAY,EAAE,IAAI,GAAE,cAAmB,GAAG,cAAc,CA0FvF;AA4BD,wBAAgB,cAAc,CAAC,CAAC,EAAE,cAAc,GAAG,MAAM,CA0ExD"}
+{"version":3,"file":"report.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/paired/report.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAGL,aAAa,EAGb,iBAAiB,EACjB,aAAa,EAEb,OAAO,EACR,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAC3C,OAAO,EAAsB,gBAAgB,EAA2B,MAAM,YAAY,CAAC;AAE3F,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,CAAC,EAAE,MAAM,CAAC;IACV,WAAW,EAAE,MAAM,CAAC;IACpB,UAAU,EAAE,MAAM,GAAG,IAAI,CAAC;IAC1B,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;IAC3B,aAAa,EAAE,MAAM,CAAC;IACtB,SAAS,EAAE,MAAM,CAAC;IAClB,sFAAsF;IACtF,WAAW,EAAE,MAAM,GAAG,IAAI,CAAC;CAC5B;AAED,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,WAAW,EAAE;QACX,YAAY,EAAE,MAAM,CAAC;QACrB,aAAa,EAAE,MAAM,CAAC;QACtB,KAAK,EAAE,iBAAiB,CAAC;QACzB,OAAO,EAAE,aAAa,CAAC;QACvB,yEAAyE;QACzE,OAAO,EAAE,OAAO,CAAC;KAClB,CAAC;IACF,iFAAiF;IACjF,OAAO,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,iBAAiB,CAAC,CAAC,CAAC;IAC9D,+EAA+E;IAC/E,cAAc,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,OAAO,CAAC,CAAC,CAAC;IAC3D;;;;OAIG;IACH,OAAO,CAAC,EAAE,iBAAiB,CAAC;IAC5B,4DAA4D;IAC5D,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B;AAED,MAAM,WAAW,cAAc;IAC7B,IAAI,EAAE;QACJ,KAAK,EAAE,MAAM,CAAC;QACd,OAAO,EAAE,MAAM,CAAC;QAChB,MAAM,EAAE,MAAM,CAAC;QACf,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,EAAE,CAAC;QACrB,QAAQ,EAAE,MAAM,CAAC;QACjB,SAAS,EAAE,MAAM,CAAC;QAClB,UAAU,EAAE,MAAM,CAAC;KACpB,CAAC;IACF,YAAY,EAAE,gBAAgB,EAAE,CAAC;IACjC,WAAW,EAAE,UAAU,EAAE,CAAC;IAC1B,uDAAuD;IACvD,MAAM,EAAE;QAAE,KAAK,EAAE,MAAM,CAAC;QAAC,WAAW,EAAE,MAAM,CAAC;QAAC,UAAU,EAAE,MAAM,GAAG,IAAI,CAAA;KAAE,EAAE,CAAC;CAC7E;AA6DD,MAAM,WAAW,cAAe,SAAQ,aAAa;IACnD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB;;2EAEuE;IACvE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB;gEAC4D;IAC5D,aAAa,CAAC,EAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAAC,CAAC;IAC1D;uEACmE;IACnE,aAAa,CAAC,EAAE,MAAM,CAAC;CACxB;AAED,wBAAgB,OAAO,CAAC,MAAM,EAAE,YAAY,EAAE,IAAI,GAAE,cAAmB,GAAG,cAAc,CAyGvF;AA4BD,wBAAgB,cAAc,CAAC,CAAC,EAAE,cAAc,GAAG,MAAM,CAuFxD"}

package/dist/benchmarks/paired/report.js

@@ -27,6 +27,7 @@ function summarize(label, recs) {
         const vals = recs.map((r) => sel(r.metrics)).filter((v) => v != null);
         return vals.length ? mean(vals) : null;
     };
+    const quals = recs.map((r) => r.qualityScore?.composite).filter((v) => v != null);
     return {
         label,
         n: recs.length,
@@ -36,6 +37,7 @@ function summarize(label, recs) {
         meanCostUsd: numeric((m) => m.costUsd),
         meanLatencyMs: numeric((m) => m.latencyMs) ?? NaN,
         errorRate: recs.length ? mean(recs.map((r) => (r.metrics.error ? 1 : 0))) : NaN,
+        meanQuality: quals.length ? mean(quals) : null,
     };
 }
 /** Build the paired delta arrays between a treatment and baseline cell map. */
@@ -92,6 +94,19 @@ export function analyze(output, opts = {}) {
             if (deltas.length > 0)
                 metrics[metric] = pairedDelta(deltas, opts);
         }
+        // Composite-quality paired delta (higher-is-better), over cells where BOTH
+        // arms carried a quality score. This is the LLM-Self-Tuning quality signal.
+        const qualityDeltas = [];
+        for (let i = 0; i < base.length; i++) {
+            const bq = base[i].qualityScore?.composite;
+            const tq = treat[i].qualityScore?.composite;
+            if (bq != null && tq != null)
+                qualityDeltas.push(tq - bq);
+        }
+        const quality = qualityDeltas.length > 0 ? pairedDelta(qualityDeltas, opts) : undefined;
+        const qualityVerdict = quality
+            ? verdict(quality, { margin: opts.qualityMargin ?? 0, higherIsBetter: true })
+            : undefined;
         // Win/tie/loss under the ROPE norm: correctness is higher-is-better; the
         // continuous metrics (tokens/cost/turns/latency) are lower-is-better.
         const correctnessVerdict = verdict(delta, { margin: opts.ropeMargin ?? 0, higherIsBetter: true });
@@ -117,6 +132,8 @@ export function analyze(output, opts = {}) {
             },
             metrics,
             metricVerdicts,
+            quality,
+            qualityVerdict,
         });
     }
     return {
@@ -172,11 +189,14 @@ export function renderMarkdown(r) {
     L.push('');
     L.push(`## Per-condition summary`);
     L.push('');
-    L.push(`| Condition | n | Success | Tokens | Turns | Cost $ | Latency ms | Err |`);
-    L.push(`|---|--:|--:|--:|--:|--:|--:|--:|`);
+    const anyQuality = r.perCondition.some((c) => c.meanQuality != null);
+    L.push(`| Condition | n | Success | Tokens | Turns | Cost $ | Latency ms | Err |` +
+        (anyQuality ? ` Quality |` : ``));
+    L.push(`|---|--:|--:|--:|--:|--:|--:|--:|` + (anyQuality ? `--:|` : ``));
     for (const c of r.perCondition) {
         L.push(`| \`${c.label}\` | ${c.n} | ${pct(c.successRate)} | ${fixed(c.meanTokens)} | ` +
-            `${fixed(c.meanTurns, 1)} | ${fixed(c.meanCostUsd, 4)} | ${fixed(c.meanLatencyMs)} | ${pct(c.errorRate)} |`);
+            `${fixed(c.meanTurns, 1)} | ${fixed(c.meanCostUsd, 4)} | ${fixed(c.meanLatencyMs)} | ${pct(c.errorRate)} |` +
+            (anyQuality ? ` ${fixed(c.meanQuality, 1)} |` : ``));
     }
     L.push('');
     L.push(`## Paired comparisons vs \`${r.meta.baseline}\``);
@@ -194,6 +214,12 @@ export function renderMarkdown(r) {
         L.push(`**Gate value (McNemar 2×2):** fixed ${m.onlyTreatment}, regressed ${m.onlyBaseline}, ` +
             `net ${m.netGain >= 0 ? '+' : ''}${m.netGain} (p=${m.pValue.toFixed(3)}); ` +
             `both✓ ${m.bothCorrect}, both✗ ${m.bothWrong}`);
+        if (cmp.quality) {
+            const q = cmp.quality;
+            L.push('');
+            L.push(`**Quality (composite 0-100):** ${vbadge(cmp.qualityVerdict ?? 'tie')}  ` +
+                `Δ ${q.meanDelta.toFixed(1)}, 95% CI ${ciStr(q.ci, 1)}, p=${q.pValue.toFixed(3)} ${sig(q)}`);
+        }
         L.push('');
         L.push(`| Metric | Δ mean | 95% CI | p | verdict |`);
         L.push(`|---|--:|--:|--:|:-:|`);

package/dist/benchmarks/paired/report.js.map

@@ -1 +1 @@
-{"version":3,"file":"report.js","sourceRoot":"","sources":["../../../src/benchmarks/paired/report.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAEL,OAAO,EAEP,IAAI,EACJ,WAAW,EAGX,OAAO,GAER,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,kBAAkB,EAA6C,MAAM,YAAY,CAAC;AAgD3F,SAAS,OAAO,CAAC,CAAY;IAC3B,OAAO,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;AACjC,CAAC;AAED,SAAS,WAAW,CAAC,OAAoB;IACvC,MAAM,CAAC,GAAG,IAAI,GAAG,EAAuB,CAAC;IACzC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;QACrC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACZ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;IAC1B,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,SAAS,CAAC,KAAa,EAAE,IAAiB;IACjD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7D,MAAM,OAAO,GAAG,CAAC,GAAuC,EAAiB,EAAE;QACzE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;QACnF,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACzC,CAAC,CAAC;IACF,OAAO;QACL,KAAK;QACL,CAAC,EAAE,IAAI,CAAC,MAAM;QACd,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG;QAC9C,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;QACpC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;QAClC,WAAW,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;QACtC,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,GAAG;QACjD,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG;KAChF,CAAC;AACJ,CAAC;AAED,+EAA+E;AAC/E,SAAS,SAAS,CAChB,QAAgC,EAChC,SAAiC;IAEjC,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,MAAM,IAAI,GAAgB,EAAE,CAAC;IAC7B,MAAM,KAAK,GAAgB,EAAE,CAAC;IAC9B,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,QAAQ,EAAE,CAAC;QAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC3B,IAAI,CAAC,EAAE,CAAC;YACN,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACb,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACb,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAChB,CAAC;IACH,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC;AAC/B,CAAC;AAED,SAAS,SAAS,CAAC,IAAiB;IAClC,MAAM,CAAC,GAAG,IAAI,GAAG,EAAqB,CAAC;IACvC,KAAK,MAAM,CAAC,IAAI,IAAI;QAAE,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC3C,OAAO,CAAC,CAAC;AACX,CAAC;AAaD,MAAM,UAAU,OAAO,CAAC,MAAoB,EAAE,OAAuB,EAAE;IACrE,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,IAAI,UAAU,CAAC;IACvD,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAE5C,MAAM,UAAU,GAAG,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACvC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,EAAE,CAAC;QAChC,MAAM,IAAI,KAAK,CACb,gCAAgC,aAAa,wBAAwB,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC9F,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAE,CAAC,CAAC,CAAC;IAEtF,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAE,CAAC,CAAC;IACvD,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC;IAEpE,MAAM,WAAW,GAAiB,EAAE,CAAC;IACrC,KAAK,MAAM,KAAK,IAAI,UAAU,EAAE,CAAC;QAC/B,IAAI,KAAK,KAAK,aAAa;YAAE,SAAS;QACtC,MAAM,QAAQ,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAE,CAAC,CAAC;QAChD,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,SAAS,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;QAErD,sEAAsE;QACtE,MAAM,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QACrG,MAAM,KAAK,GAAG,WAAW,CAAC,iBAAiB,EAAE,IAAI,CAAC,CAAC;QACnD,MAAM,EAAE,GAAG,OAAO,CAChB,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,EACnC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CACnC,CAAC;QAEF,wEAAwE;QACxE,MAAM,OAAO,GAAyD,EAAE,CAAC;QACzE,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACrC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;gBACnC,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;gBACpC,IAAI,EAAE,IAAI,IAAI,IAAI,EAAE,IAAI,IAAI;oBAAE,MAAM,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC;YACrD,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;gBAAE,OAAO,CAAC,MAAM,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;QACrE,CAAC;QAED,yEAAyE;QACzE,sEAAsE;QACtE,MAAM,kBAAkB,GAAG,OAAO,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC;QAClG,MAAM,cAAc,GAA+C,EAAE,CAAC;QACtE,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;YAC3B,IAAI,EAAE,EAAE,CAAC;gBACP,cAAc,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,EAAE,EAAE;oBACnC,MAAM,EAAE,IAAI,CAAC,aAAa,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;oBACzC,cAAc,EAAE,KAAK;iBACtB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,WAAW,CAAC,IAAI,CAAC;YACf,KAAK;YACL,QAAQ,EAAE,aAAa;YACvB,WAAW,EAAE;gBACX,YAAY,EAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;gBAC3D,aAAa,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;gBAC7D,KAAK;gBACL,OAAO,EAAE,EAAE;gBACX,OAAO,EAAE,kBAAkB;aAC5B;YACD,OAAO;YACP,cAAc;SACf,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,IAAI,EAAE;YACJ,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS;YACT,UAAU;YACV,QAAQ,EAAE,aAAa;YACvB,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B;QACD,YAAY;QACZ,WAAW;QACX,MAAM,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC/B,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,UAAU,EAAE,CAAC,CAAC,UAAU;SACzB,CAAC,CAAC;KACJ,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,CAAU;IACrB,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACnB,CAAC;AAED,8EAA8E;AAC9E,oBAAoB;AACpB,8EAA8E;AAE9E,SAAS,GAAG,CAAC,CAAS;IACpB,OAAO,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC;AACjE,CAAC;AACD,SAAS,KAAK,CAAC,CAAgB,EAAE,CAAC,GAAG,CAAC;IACpC,OAAO,CAAC,IAAI,IAAI,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACjE,CAAC;AACD,SAAS,KAAK,CAAC,EAAM,EAAE,CAAC,GAAG,CAAC;IAC1B,OAAO,IAAI,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;AAC5D,CAAC;AACD,SAAS,GAAG,CAAC,CAAoB;IAC/B,OAAO,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;AACnC,CAAC;AAED,+EAA+E;AAC/E,SAAS,MAAM,CAAC,CAAU;IACxB,OAAO,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;AACrE,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,CAAiB;IAC9C,MAAM,CAAC,GAAa,EAAE,CAAC;IACvB,CAAC,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACxC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CACJ,cAAc,CAAC,CAAC,IAAI,CAAC,KAAK,qBAAqB,CAAC,CAAC,IAAI,CAAC,OAAO,OAAO;QAClE,cAAc,CAAC,CAAC,IAAI,CAAC,SAAS,oBAAoB,CAAC,CAAC,IAAI,CAAC,MAAM,OAAO;QACtE,mBAAmB,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,CACzC,CAAC;IACF,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,MAAM,CAAC,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,CAAC;IACvD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEX,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACnC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CAAC,0EAA0E,CAAC,CAAC;IACnF,CAAC,CAAC,IAAI,CAAC,mCAAmC,CAAC,CAAC;IAC5C,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,CAAC;QAC/B,CAAC,CAAC,IAAI,CACJ,OAAO,CAAC,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK;YAC7E,GAAG,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,CAC9G,CAAC;IACJ,CAAC;IACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEX,CAAC,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;IAC1D,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,KAAK,MAAM,GAAG,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC;QAChC,CAAC,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,KAAK,IAAI,CAAC,CAAC;QAC/B,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,MAAM,EAAE,GAAG,GAAG,CAAC,WAAW,CAAC;QAC3B,CAAC,CAAC,IAAI,CACJ,gBAAgB,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI;YACpC,oBAAoB,GAAG,CAAC,EAAE,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,aAAa,CAAC,IAAI;YACvE,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI;YAC/E,KAAK,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CACtD,CAAC;QACF,MAAM,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC;QACrB,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,CAAC,CAAC,IAAI,CACJ,uCAAuC,CAAC,CAAC,aAAa,eAAe,CAAC,CAAC,YAAY,IAAI;YACrF,OAAO,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,OAAO,OAAO,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YAC3E,SAAS,CAAC,CAAC,WAAW,WAAW,CAAC,CAAC,SAAS,EAAE,CACjD,CAAC;QACF,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,CAAC,CAAC,IAAI,CAAC,4CAA4C,CAAC,CAAC;QACrD,CAAC,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC;QAChC,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,EAAE,GAAG,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,MAAM,CAAC,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACvC,MAAM,EAAE,GAAG,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC;YAC/C,CAAC,CAAC,IAAI,CACJ,KAAK,MAAM,MAAM,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK;gBAChE,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,MAAM,CAAC,EAAE,CAAC,IAAI,CAC9C,CAAC;QACJ,CAAC;QACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACb,CAAC;IAED,CAAC,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAClC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CAAC,uCAAuC,CAAC,CAAC;IAChD,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IACxB,KAAK,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,EAAE,CAAC;QAC5E,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,QAAQ,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAChF,CAAC;IACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CACJ,4EAA4E;QAC1E,6FAA6F;QAC7F,yEAAyE,CAC5E,CAAC;IACF,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACtB,CAAC"}
+{"version":3,"file":"report.js","sourceRoot":"","sources":["../../../src/benchmarks/paired/report.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAEH,OAAO,EAEL,OAAO,EAEP,IAAI,EACJ,WAAW,EAGX,OAAO,GAER,MAAM,YAAY,CAAC;AAEpB,OAAO,EAAE,kBAAkB,EAA6C,MAAM,YAAY,CAAC;AA0D3F,SAAS,OAAO,CAAC,CAAY;IAC3B,OAAO,GAAG,CAAC,CAAC,MAAM,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC;AACjC,CAAC;AAED,SAAS,WAAW,CAAC,OAAoB;IACvC,MAAM,CAAC,GAAG,IAAI,GAAG,EAAuB,CAAC;IACzC,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;QACrC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACZ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,SAAS,EAAE,GAAG,CAAC,CAAC;IAC1B,CAAC;IACD,OAAO,CAAC,CAAC;AACX,CAAC;AAED,SAAS,SAAS,CAAC,KAAa,EAAE,IAAiB;IACjD,MAAM,OAAO,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7D,MAAM,OAAO,GAAG,CAAC,GAAuC,EAAiB,EAAE;QACzE,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;QACnF,OAAO,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC;IACzC,CAAC,CAAC;IACF,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,EAAE,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,CAAC,IAAI,IAAI,CAAC,CAAC;IAC/F,OAAO;QACL,KAAK;QACL,CAAC,EAAE,IAAI,CAAC,MAAM;QACd,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG;QAC9C,UAAU,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC;QACpC,SAAS,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC;QAClC,WAAW,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC;QACtC,aAAa,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI,GAAG;QACjD,SAAS,EAAE,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG;QAC/E,WAAW,EAAE,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI;KAC/C,CAAC;AACJ,CAAC;AAED,+EAA+E;AAC/E,SAAS,SAAS,CAChB,QAAgC,EAChC,SAAiC;IAEjC,MAAM,IAAI,GAAa,EAAE,CAAC;IAC1B,MAAM,IAAI,GAAgB,EAAE,CAAC;IAC7B,MAAM,KAAK,GAAgB,EAAE,CAAC;IAC9B,KAAK,MAAM,CAAC,CAAC,EAAE,CAAC,CAAC,IAAI,QAAQ,EAAE,CAAC;QAC9B,MAAM,CAAC,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QAC3B,IAAI,CAAC,EAAE,CAAC;YACN,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACb,IAAI,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACb,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QAChB,CAAC;IACH,CAAC;IACD,OAAO,EAAE,IAAI,EAAE,IAAI,EAAE,KAAK,EAAE,CAAC;AAC/B,CAAC;AAED,SAAS,SAAS,CAAC,IAAiB;IAClC,MAAM,CAAC,GAAG,IAAI,GAAG,EAAqB,CAAC;IACvC,KAAK,MAAM,CAAC,IAAI,IAAI;QAAE,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC3C,OAAO,CAAC,CAAC;AACX,CAAC;AAgBD,MAAM,UAAU,OAAO,CAAC,MAAoB,EAAE,OAAuB,EAAE;IACrE,MAAM,aAAa,GAAG,IAAI,CAAC,aAAa,IAAI,UAAU,CAAC;IACvD,MAAM,OAAO,GAAG,WAAW,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC;IAE5C,MAAM,UAAU,GAAG,CAAC,GAAG,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC;IACvC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAC,EAAE,CAAC;QAChC,MAAM,IAAI,KAAK,CACb,gCAAgC,aAAa,wBAAwB,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,CAC9F,CAAC;IACJ,CAAC;IAED,MAAM,YAAY,GAAG,UAAU,CAAC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,SAAS,CAAC,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,KAAK,CAAE,CAAC,CAAC,CAAC;IAEtF,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,aAAa,CAAE,CAAC,CAAC;IACvD,MAAM,SAAS,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,CAAC;IAEpE,MAAM,WAAW,GAAiB,EAAE,CAAC;IACrC,KAAK,MAAM,KAAK,IAAI,UAAU,EAAE,CAAC;QAC/B,IAAI,KAAK,KAAK,aAAa;YAAE,SAAS;QACtC,MAAM,QAAQ,GAAG,SAAS,CAAC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAE,CAAC,CAAC;QAChD,MAAM,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,SAAS,CAAC,OAAO,EAAE,QAAQ,CAAC,CAAC;QAErD,sEAAsE;QACtE,MAAM,iBAAiB,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;QACrG,MAAM,KAAK,GAAG,WAAW,CAAC,iBAAiB,EAAE,IAAI,CAAC,CAAC;QACnD,MAAM,EAAE,GAAG,OAAO,CAChB,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,EACnC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CACnC,CAAC;QAEF,wEAAwE;QACxE,MAAM,OAAO,GAAyD,EAAE,CAAC;QACzE,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,MAAM,GAAa,EAAE,CAAC;YAC5B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;gBACrC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;gBACnC,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;gBACpC,IAAI,EAAE,IAAI,IAAI,IAAI,EAAE,IAAI,IAAI;oBAAE,MAAM,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC;YACrD,CAAC;YACD,IAAI,MAAM,CAAC,MAAM,GAAG,CAAC;gBAAE,OAAO,CAAC,MAAM,CAAC,GAAG,WAAW,CAAC,MAAM,EAAE,IAAI,CAAC,CAAC;QACrE,CAAC;QAED,2EAA2E;QAC3E,4EAA4E;QAC5E,MAAM,aAAa,GAAa,EAAE,CAAC;QACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YACrC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,YAAY,EAAE,SAAS,CAAC;YAC3C,MAAM,EAAE,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,YAAY,EAAE,SAAS,CAAC;YAC5C,IAAI,EAAE,IAAI,IAAI,IAAI,EAAE,IAAI,IAAI;gBAAE,aAAa,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC;QAC5D,CAAC;QACD,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,aAAa,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACxF,MAAM,cAAc,GAAG,OAAO;YAC5B,CAAC,CAAC,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,aAAa,IAAI,CAAC,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC;YAC7E,CAAC,CAAC,SAAS,CAAC;QAEd,yEAAyE;QACzE,sEAAsE;QACtE,MAAM,kBAAkB,GAAG,OAAO,CAAC,KAAK,EAAE,EAAE,MAAM,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,EAAE,cAAc,EAAE,IAAI,EAAE,CAAC,CAAC;QAClG,MAAM,cAAc,GAA+C,EAAE,CAAC;QACtE,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,EAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;YAC3B,IAAI,EAAE,EAAE,CAAC;gBACP,cAAc,CAAC,MAAM,CAAC,GAAG,OAAO,CAAC,EAAE,EAAE;oBACnC,MAAM,EAAE,IAAI,CAAC,aAAa,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC;oBACzC,cAAc,EAAE,KAAK;iBACtB,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,WAAW,CAAC,IAAI,CAAC;YACf,KAAK;YACL,QAAQ,EAAE,aAAa;YACvB,WAAW,EAAE;gBACX,YAAY,EAAE,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;gBAC3D,aAAa,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC;gBAC7D,KAAK;gBACL,OAAO,EAAE,EAAE;gBACX,OAAO,EAAE,kBAAkB;aAC5B;YACD,OAAO;YACP,cAAc;YACd,OAAO;YACP,cAAc;SACf,CAAC,CAAC;IACL,CAAC;IAED,OAAO;QACL,IAAI,EAAE;YACJ,KAAK,EAAE,MAAM,CAAC,KAAK;YACnB,OAAO,EAAE,MAAM,CAAC,OAAO;YACvB,MAAM,EAAE,MAAM,CAAC,MAAM;YACrB,SAAS;YACT,UAAU;YACV,QAAQ,EAAE,aAAa;YACvB,SAAS,EAAE,MAAM,CAAC,SAAS;YAC3B,UAAU,EAAE,MAAM,CAAC,UAAU;SAC9B;QACD,YAAY;QACZ,WAAW;QACX,MAAM,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;YAC/B,KAAK,EAAE,CAAC,CAAC,KAAK;YACd,WAAW,EAAE,CAAC,CAAC,WAAW;YAC1B,UAAU,EAAE,CAAC,CAAC,UAAU;SACzB,CAAC,CAAC;KACJ,CAAC;AACJ,CAAC;AAED,SAAS,GAAG,CAAC,CAAU;IACrB,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AACnB,CAAC;AAED,8EAA8E;AAC9E,oBAAoB;AACpB,8EAA8E;AAE9E,SAAS,GAAG,CAAC,CAAS;IACpB,OAAO,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,KAAK,CAAC;AACjE,CAAC;AACD,SAAS,KAAK,CAAC,CAAgB,EAAE,CAAC,GAAG,CAAC;IACpC,OAAO,CAAC,IAAI,IAAI,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;AACjE,CAAC;AACD,SAAS,KAAK,CAAC,EAAM,EAAE,CAAC,GAAG,CAAC;IAC1B,OAAO,IAAI,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,EAAE,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC;AAC5D,CAAC;AACD,SAAS,GAAG,CAAC,CAAoB;IAC/B,OAAO,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC;AACnC,CAAC;AAED,+EAA+E;AAC/E,SAAS,MAAM,CAAC,CAAU;IACxB,OAAO,CAAC,KAAK,KAAK,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,OAAO,CAAC;AACrE,CAAC;AAED,MAAM,UAAU,cAAc,CAAC,CAAiB;IAC9C,MAAM,CAAC,GAAa,EAAE,CAAC;IACvB,CAAC,CAAC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACxC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CACJ,cAAc,CAAC,CAAC,IAAI,CAAC,KAAK,qBAAqB,CAAC,CAAC,IAAI,CAAC,OAAO,OAAO;QAClE,cAAc,CAAC,CAAC,IAAI,CAAC,SAAS,oBAAoB,CAAC,CAAC,IAAI,CAAC,MAAM,OAAO;QACtE,mBAAmB,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,CACzC,CAAC;IACF,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,SAAS,MAAM,CAAC,CAAC,IAAI,CAAC,UAAU,GAAG,CAAC,CAAC;IACvD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEX,CAAC,CAAC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACnC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,MAAM,UAAU,GAAG,CAAC,CAAC,YAAY,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,IAAI,IAAI,CAAC,CAAC;IACrE,CAAC,CAAC,IAAI,CACJ,0EAA0E;QACxE,CAAC,UAAU,CAAC,CAAC,CAAC,YAAY,CAAC,CAAC,CAAC,EAAE,CAAC,CACnC,CAAC;IACF,CAAC,CAAC,IAAI,CAAC,mCAAmC,GAAG,CAAC,UAAU,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IACzE,KAAK,MAAM,CAAC,IAAI,CAAC,CAAC,YAAY,EAAE,CAAC;QAC/B,CAAC,CAAC,IAAI,CACJ,OAAO,CAAC,CAAC,KAAK,QAAQ,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,KAAK;YAC7E,GAAG,KAAK,CAAC,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,aAAa,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,IAAI;YAC3G,CAAC,UAAU,CAAC,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,CAAC,WAAW,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CACtD,CAAC;IACJ,CAAC;IACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IAEX,CAAC,CAAC,IAAI,CAAC,8BAA8B,CAAC,CAAC,IAAI,CAAC,QAAQ,IAAI,CAAC,CAAC;IAC1D,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,KAAK,MAAM,GAAG,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC;QAChC,CAAC,CAAC,IAAI,CAAC,SAAS,GAAG,CAAC,KAAK,IAAI,CAAC,CAAC;QAC/B,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,MAAM,EAAE,GAAG,GAAG,CAAC,WAAW,CAAC;QAC3B,CAAC,CAAC,IAAI,CACJ,gBAAgB,MAAM,CAAC,EAAE,CAAC,OAAO,CAAC,IAAI;YACpC,oBAAoB,GAAG,CAAC,EAAE,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,aAAa,CAAC,IAAI;YACvE,MAAM,CAAC,EAAE,CAAC,KAAK,CAAC,SAAS,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,cAAc,KAAK,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CAAC,IAAI;YAC/E,KAAK,EAAE,CAAC,KAAK,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK,GAAG,CAAC,EAAE,CAAC,KAAK,CAAC,EAAE,CACtD,CAAC;QACF,MAAM,CAAC,GAAG,EAAE,CAAC,OAAO,CAAC;QACrB,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,CAAC,CAAC,IAAI,CACJ,uCAAuC,CAAC,CAAC,aAAa,eAAe,CAAC,CAAC,YAAY,IAAI;YACrF,OAAO,CAAC,CAAC,OAAO,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,OAAO,OAAO,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,KAAK;YAC3E,SAAS,CAAC,CAAC,WAAW,WAAW,CAAC,CAAC,SAAS,EAAE,CACjD,CAAC;QACF,IAAI,GAAG,CAAC,OAAO,EAAE,CAAC;YAChB,MAAM,CAAC,GAAG,GAAG,CAAC,OAAO,CAAC;YACtB,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;YACX,CAAC,CAAC,IAAI,CACJ,kCAAkC,MAAM,CAAC,GAAG,CAAC,cAAc,IAAI,KAAK,CAAC,IAAI;gBACvE,KAAK,CAAC,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,GAAG,CAAC,CAAC,CAAC,EAAE,CAC9F,CAAC;QACJ,CAAC;QACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;QACX,CAAC,CAAC,IAAI,CAAC,4CAA4C,CAAC,CAAC;QACrD,CAAC,CAAC,IAAI,CAAC,uBAAuB,CAAC,CAAC;QAChC,KAAK,MAAM,MAAM,IAAI,kBAAkB,EAAE,CAAC;YACxC,MAAM,EAAE,GAAG,GAAG,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;YAC/B,IAAI,CAAC,EAAE;gBAAE,SAAS;YAClB,MAAM,CAAC,GAAG,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACvC,MAAM,EAAE,GAAG,GAAG,CAAC,cAAc,CAAC,MAAM,CAAC,IAAI,KAAK,CAAC;YAC/C,CAAC,CAAC,IAAI,CACJ,KAAK,MAAM,MAAM,EAAE,CAAC,SAAS,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,KAAK;gBAChE,GAAG,EAAE,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,MAAM,CAAC,EAAE,CAAC,IAAI,CAC9C,CAAC;QACJ,CAAC;QACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACb,CAAC;IAED,CAAC,CAAC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAClC,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CAAC,uCAAuC,CAAC,CAAC;IAChD,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,CAAC;IACxB,KAAK,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,EAAE,CAAC;QAC5E,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,QAAQ,GAAG,CAAC,CAAC,CAAC,WAAW,CAAC,MAAM,KAAK,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;IAChF,CAAC;IACD,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,CAAC,CAAC,IAAI,CACJ,4EAA4E;QAC1E,6FAA6F;QAC7F,yEAAyE,CAC5E,CAAC;IACF,CAAC,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC;IACX,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACtB,CAAC"}

package/dist/benchmarks/paired/types.d.ts

@@ -145,6 +145,59 @@ export type MetricVector = z.infer<typeof MetricVectorSchema>;
 /** The continuous (numeric) metric keys eligible for paired-delta analysis. */
 export declare const CONTINUOUS_METRICS: readonly ["tokens", "costUsd", "turns", "toolCalls", "latencyMs"];
 export type ContinuousMetric = (typeof CONTINUOUS_METRICS)[number];
+/**
+ * The quality dimensions scored for a single run. Correctness/quality/planning
+ * are judged (an LLM grades spec vs output); efficiency/toolReliability are
+ * derived deterministically from the metric vector. Together they replace the
+ * pass/fail binary with a signal the tuner can optimize toward Opus-level output.
+ * See docs/design/LLM_SELF_TUNING_ANALYSIS.md §3.3.1.
+ */
+export declare const QUALITY_DIMENSIONS: readonly ["correctness", "quality", "efficiency", "toolReliability", "planning"];
+export type QualityDimension = (typeof QUALITY_DIMENSIONS)[number];
+/**
+ * Default composite weights (sum to 1). Correctness dominates; quality is the
+ * next lever; efficiency + tool reliability are secondary. Callers may override
+ * per model/task via `compositeQuality`.
+ */
+export declare const DEFAULT_QUALITY_WEIGHTS: Readonly<Record<QualityDimension, number>>;
+export declare const QualityScoreSchema: z.ZodObject<{
+    correctness: z.ZodNumber;
+    quality: z.ZodNumber;
+    efficiency: z.ZodNumber;
+    toolReliability: z.ZodNumber;
+    planning: z.ZodNumber;
+    /** Weighted composite (0-100); recomputed from the dimensions on write. */
+    composite: z.ZodNumber;
+    /** How this score was produced: 'judge' (LLM), 'heuristic' (metrics-only), or 'hybrid'. */
+    source: z.ZodDefault<z.ZodEnum<["judge", "heuristic", "hybrid"]>>;
+    /** Optional one-line judge rationale (audit trail). */
+    rationale: z.ZodOptional<z.ZodString>;
+}, "strip", z.ZodTypeAny, {
+    planning: number;
+    source: "hybrid" | "judge" | "heuristic";
+    quality: number;
+    correctness: number;
+    efficiency: number;
+    toolReliability: number;
+    composite: number;
+    rationale?: string | undefined;
+}, {
+    planning: number;
+    quality: number;
+    correctness: number;
+    efficiency: number;
+    toolReliability: number;
+    composite: number;
+    source?: "hybrid" | "judge" | "heuristic" | undefined;
+    rationale?: string | undefined;
+}>;
+export type QualityScore = z.infer<typeof QualityScoreSchema>;
+/**
+ * Compute the weighted composite (0-100) from the five dimensions. Weights are
+ * renormalized over the dimensions actually supplied so a partial score (e.g.
+ * planning omitted) is not silently deflated.
+ */
+export declare function compositeQuality(dims: Record<QualityDimension, number>, weights?: Partial<Record<QualityDimension, number>>): number;
 /** One execution of one task under one condition at one seed/epoch. */
 export interface RunRecord {
     taskId: string;
@@ -156,6 +209,12 @@ export interface RunRecord {
     adapter: string;
     /** Model identifier the adapter ran against. */
     model: string;
+    /**
+     * Optional multi-dimensional quality score (LLM Self-Tuning P0). Present only
+     * when the run was scored; the paired report surfaces its composite delta
+     * alongside correctness so the tuner has a signal beyond pass/fail.
+     */
+    qualityScore?: QualityScore;
 }
 /**
  * Result returned by an AgentAdapter for a single run, before the verify

package/dist/benchmarks/paired/types.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/paired/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAMxB;;;;GAIG;AACH,eAAO,MAAM,cAAc;;;IAGzB,4EAA4E;;;IAG5E,iFAAiF;;IAEjF;;;OAGG;;IAEH;;;;OAIG;;IAEH;;;;;;OAMG;;IAEH,kFAAkF;;IAElF,0EAA0E;;IAE1E,qEAAqE;;;;;;;;;;;;;;;;;;;;;;;;;;EAErE,CAAC;AAEH,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AAMtD;;;GAGG;AACH,eAAO,MAAM,cAAc,2EAOjB,CAAC;AAEX,MAAM,MAAM,YAAY,GAAG,CAAC,OAAO,cAAc,CAAC,CAAC,MAAM,CAAC,CAAC;AAE3D;;;GAGG;AACH,MAAM,WAAW,SAAS;IACxB,6EAA6E;IAC7E,KAAK,EAAE,MAAM,CAAC;IACd,uCAAuC;IACvC,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,yEAAyE;IACzE,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,wBAAgB,qBAAqB,IAAI,SAAS,CAEjD;AAED,wBAAgB,iBAAiB,IAAI,SAAS,CAE7C;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,IAAI,SAAS,CAE7C;AAED,2EAA2E;AAC3E,wBAAgB,UAAU,CAAC,CAAC,EAAE,SAAS,GAAG,OAAO,CAEhD;AAMD,eAAO,MAAM,kBAAkB;IAC7B,iDAAiD;;IAEjD,0EAA0E;;IAE1E,uDAAuD;;IAEvD,kDAAkD;;IAElD,iDAAiD;;IAEjD,2DAA2D;;IAE3D,mFAAmF;;IAEnF,+DAA+D;;;;;;;;;;;;;;;;;;;;EAE/D,CAAC;AAEH,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D,+EAA+E;AAC/E,eAAO,MAAM,kBAAkB,mEAMrB,CAAC;AACX,MAAM,MAAM,gBAAgB,GAAG,CAAC,OAAO,kBAAkB,CAAC,CAAC,MAAM,CAAC,CAAC;AAMnE,uEAAuE;AACvE,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,+EAA+E;IAC/E,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,YAAY,CAAC;IACtB,0EAA0E;IAC1E,OAAO,EAAE,MAAM,CAAC;IAChB,gDAAgD;IAChD,KAAK,EAAE,MAAM,CAAC;CACf;AAMD;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,OAAO,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,kFAAkF;IAClF,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,QAAQ,CAAC;IACf,SAAS,EAAE,SAAS,CAAC;IACrB,8EAA8E;IAC9E,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,YAAY;IAC3B,yCAAyC;IACzC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,4EAA4E;IAC5E,8EAA8E;IAC9E,GAAG,CAAC,GAAG,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;CACpD;AAMD,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,QAAQ,EAAE,CAAC;IAClB,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB,OAAO,EAAE,YAAY,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,0EAA0E;IAC1E,MAAM,EAAE,MAAM,CAAC;IACf,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,+CAA+C;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,kCAAkC;IAClC,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;CACnE"}
+{"version":3,"file":"types.d.ts","sourceRoot":"","sources":["../../../src/benchmarks/paired/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAMxB;;;;GAIG;AACH,eAAO,MAAM,cAAc;;;IAGzB,4EAA4E;;;IAG5E,iFAAiF;;IAEjF;;;OAGG;;IAEH;;;;OAIG;;IAEH;;;;;;OAMG;;IAEH,kFAAkF;;IAElF,0EAA0E;;IAE1E,qEAAqE;;;;;;;;;;;;;;;;;;;;;;;;;;EAErE,CAAC;AAEH,MAAM,MAAM,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,cAAc,CAAC,CAAC;AAMtD;;;GAGG;AACH,eAAO,MAAM,cAAc,2EAOjB,CAAC;AAEX,MAAM,MAAM,YAAY,GAAG,CAAC,OAAO,cAAc,CAAC,CAAC,MAAM,CAAC,CAAC;AAE3D;;;GAGG;AACH,MAAM,WAAW,SAAS;IACxB,6EAA6E;IAC7E,KAAK,EAAE,MAAM,CAAC;IACd,uCAAuC;IACvC,UAAU,EAAE,WAAW,CAAC,YAAY,CAAC,CAAC;IACtC,yEAAyE;IACzE,IAAI,CAAC,EAAE,OAAO,CAAC;CAChB;AAED,wBAAgB,qBAAqB,IAAI,SAAS,CAEjD;AAED,wBAAgB,iBAAiB,IAAI,SAAS,CAE7C;AAED;;;;;GAKG;AACH,wBAAgB,iBAAiB,IAAI,SAAS,CAE7C;AAED,2EAA2E;AAC3E,wBAAgB,UAAU,CAAC,CAAC,EAAE,SAAS,GAAG,OAAO,CAEhD;AAMD,eAAO,MAAM,kBAAkB;IAC7B,iDAAiD;;IAEjD,0EAA0E;;IAE1E,uDAAuD;;IAEvD,kDAAkD;;IAElD,iDAAiD;;IAEjD,2DAA2D;;IAE3D,mFAAmF;;IAEnF,+DAA+D;;;;;;;;;;;;;;;;;;;;EAE/D,CAAC;AAEH,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D,+EAA+E;AAC/E,eAAO,MAAM,kBAAkB,mEAMrB,CAAC;AACX,MAAM,MAAM,gBAAgB,GAAG,CAAC,OAAO,kBAAkB,CAAC,CAAC,MAAM,CAAC,CAAC;AAMnE;;;;;;GAMG;AACH,eAAO,MAAM,kBAAkB,kFAMrB,CAAC;AACX,MAAM,MAAM,gBAAgB,GAAG,CAAC,OAAO,kBAAkB,CAAC,CAAC,MAAM,CAAC,CAAC;AAEnE;;;;GAIG;AACH,eAAO,MAAM,uBAAuB,EAAE,QAAQ,CAAC,MAAM,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAM9E,CAAC;AAEF,eAAO,MAAM,kBAAkB;;;;;;IAM7B,2EAA2E;;IAE3E,2FAA2F;;IAE3F,uDAAuD;;;;;;;;;;;;;;;;;;;;EAEvD,CAAC;AAEH,MAAM,MAAM,YAAY,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,kBAAkB,CAAC,CAAC;AAE9D;;;;GAIG;AACH,wBAAgB,gBAAgB,CAC9B,IAAI,EAAE,MAAM,CAAC,gBAAgB,EAAE,MAAM,CAAC,EACtC,OAAO,GAAE,OAAO,CAAC,MAAM,CAAC,gBAAgB,EAAE,MAAM,CAAC,CAA2B,GAC3E,MAAM,CAWR;AAMD,uEAAuE;AACvE,MAAM,WAAW,SAAS;IACxB,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,+EAA+E;IAC/E,IAAI,EAAE,MAAM,CAAC;IACb,OAAO,EAAE,YAAY,CAAC;IACtB,0EAA0E;IAC1E,OAAO,EAAE,MAAM,CAAC;IAChB,gDAAgD;IAChD,KAAK,EAAE,MAAM,CAAC;IACd;;;;OAIG;IACH,YAAY,CAAC,EAAE,YAAY,CAAC;CAC7B;AAMD;;;;GAIG;AACH,MAAM,WAAW,cAAc;IAC7B,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IACtB,OAAO,EAAE,MAAM,GAAG,IAAI,CAAC;IACvB,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,SAAS,EAAE,MAAM,GAAG,IAAI,CAAC;IACzB,UAAU,EAAE,OAAO,GAAG,IAAI,CAAC;IAC3B,KAAK,EAAE,MAAM,GAAG,IAAI,CAAC;IACrB,kFAAkF;IAClF,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,eAAe;IAC9B,IAAI,EAAE,QAAQ,CAAC;IACf,SAAS,EAAE,SAAS,CAAC;IACrB,8EAA8E;IAC9E,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,KAAK,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,YAAY;IAC3B,yCAAyC;IACzC,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,4EAA4E;IAC5E,8EAA8E;IAC9E,GAAG,CAAC,GAAG,EAAE,eAAe,GAAG,OAAO,CAAC,cAAc,CAAC,CAAC;CACpD;AAMD,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,QAAQ,EAAE,CAAC;IAClB,UAAU,EAAE,SAAS,EAAE,CAAC;IACxB,OAAO,EAAE,YAAY,CAAC;IACtB,KAAK,EAAE,MAAM,CAAC;IACd,0EAA0E;IAC1E,MAAM,EAAE,MAAM,CAAC;IACf,2BAA2B;IAC3B,WAAW,EAAE,MAAM,CAAC;IACpB,+CAA+C;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,kCAAkC;IAClC,UAAU,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,KAAK,IAAI,CAAC;CACnE"}

package/dist/benchmarks/paired/types.js

@@ -117,4 +117,64 @@ export const CONTINUOUS_METRICS = [
     'toolCalls',
     'latencyMs',
 ];
+// ============================================================================
+// Quality score — the multi-dimensional quality signal (LLM Self-Tuning P0)
+// ============================================================================
+/**
+ * The quality dimensions scored for a single run. Correctness/quality/planning
+ * are judged (an LLM grades spec vs output); efficiency/toolReliability are
+ * derived deterministically from the metric vector. Together they replace the
+ * pass/fail binary with a signal the tuner can optimize toward Opus-level output.
+ * See docs/design/LLM_SELF_TUNING_ANALYSIS.md §3.3.1.
+ */
+export const QUALITY_DIMENSIONS = [
+    'correctness', // behavioral correctness vs the task spec (0-100)
+    'quality', // output quality relative to a reference / best-practice (0-100)
+    'efficiency', // tokens-per-correct-answer, normalized to 0-100 (higher = leaner)
+    'toolReliability', // fraction of tool calls that were well-formed/productive (0-100)
+    'planning', // multi-turn planning coherence (0-100)
+];
+/**
+ * Default composite weights (sum to 1). Correctness dominates; quality is the
+ * next lever; efficiency + tool reliability are secondary. Callers may override
+ * per model/task via `compositeQuality`.
+ */
+export const DEFAULT_QUALITY_WEIGHTS = {
+    correctness: 0.4,
+    quality: 0.3,
+    efficiency: 0.15,
+    toolReliability: 0.15,
+    planning: 0.0,
+};
+export const QualityScoreSchema = z.object({
+    correctness: z.number().min(0).max(100),
+    quality: z.number().min(0).max(100),
+    efficiency: z.number().min(0).max(100),
+    toolReliability: z.number().min(0).max(100),
+    planning: z.number().min(0).max(100),
+    /** Weighted composite (0-100); recomputed from the dimensions on write. */
+    composite: z.number().min(0).max(100),
+    /** How this score was produced: 'judge' (LLM), 'heuristic' (metrics-only), or 'hybrid'. */
+    source: z.enum(['judge', 'heuristic', 'hybrid']).default('heuristic'),
+    /** Optional one-line judge rationale (audit trail). */
+    rationale: z.string().optional(),
+});
+/**
+ * Compute the weighted composite (0-100) from the five dimensions. Weights are
+ * renormalized over the dimensions actually supplied so a partial score (e.g.
+ * planning omitted) is not silently deflated.
+ */
+export function compositeQuality(dims, weights = DEFAULT_QUALITY_WEIGHTS) {
+    let wsum = 0;
+    let acc = 0;
+    for (const d of QUALITY_DIMENSIONS) {
+        const w = weights[d] ?? 0;
+        if (w <= 0)
+            continue;
+        const v = Number.isFinite(dims[d]) ? Math.max(0, Math.min(100, dims[d])) : 0;
+        acc += w * v;
+        wsum += w;
+    }
+    return wsum > 0 ? acc / wsum : 0;
+}
 //# sourceMappingURL=types.js.map

package/dist/benchmarks/paired/types.js.map

@@ -1 +1 @@
-{"version":3,"file":"types.js","sourceRoot":"","sources":["../../../src/benchmarks/paired/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E;;;;GAIG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IACrC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,4EAA4E;IAC5E,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE;IACvB,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;IAChE,iFAAiF;IACjF,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACrC;;;OAGG;IACH,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC;IACnC;;;;OAIG;IACH,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB;;;;;;OAMG;IACH,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC9B,kFAAkF;IAClF,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,0EAA0E;IAC1E,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC;IACzC,qEAAqE;IACrE,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC;CACzC,CAAC,CAAC;AAIH,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG;IAC5B,OAAO,EAAE,iDAAiD;IAC1D,UAAU,EAAE,qBAAqB;IACjC,QAAQ,EAAE,uCAAuC;IACjD,SAAS,EAAE,iBAAiB;IAC5B,QAAQ,EAAE,iBAAiB;IAC3B,UAAU,EAAE,iBAAiB;CACrB,CAAC;AAiBX,MAAM,UAAU,qBAAqB;IACnC,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,EAAE,EAAE,CAAC;AACtD,CAAC;AAED,MAAM,UAAU,iBAAiB;IAC/B,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,CAAC,cAAc,CAAC,EAAE,CAAC;AACpE,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,iBAAiB;IAC/B,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,CAAC,cAAc,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;AAChF,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,UAAU,CAAC,CAAY;IACrC,OAAO,CAAC,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,CAAC;AACjC,CAAC;AAED,+EAA+E;AAC/E,0CAA0C;AAC1C,+EAA+E;AAE/E,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,iDAAiD;IACjD,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE;IACpB,0EAA0E;IAC1E,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,uDAAuD;IACvD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC9B,kDAAkD;IAClD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC5B,iDAAiD;IACjD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,2DAA2D;IAC3D,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB,mFAAmF;IACnF,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;IAClC,+DAA+D;IAC/D,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC7B,CAAC,CAAC;AAIH,+EAA+E;AAC/E,MAAM,CAAC,MAAM,kBAAkB,GAAG;IAChC,QAAQ;IACR,SAAS;IACT,OAAO;IACP,WAAW;IACX,WAAW;CACH,CAAC"}
+{"version":3,"file":"types.js","sourceRoot":"","sources":["../../../src/benchmarks/paired/types.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,+EAA+E;AAC/E,aAAa;AACb,+EAA+E;AAE/E;;;;GAIG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,CAAC,MAAM,CAAC;IACrC,EAAE,EAAE,CAAC,CAAC,MAAM,EAAE;IACd,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE;IAChB,4EAA4E;IAC5E,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE;IACvB,UAAU,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,QAAQ,EAAE,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,QAAQ,CAAC;IAChE,iFAAiF;IACjF,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC;IACrC;;;OAGG;IACH,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,MAAM,CAAC;IACnC;;;;OAIG;IACH,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB;;;;;;OAMG;IACH,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC9B,kFAAkF;IAClF,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC/B,0EAA0E;IAC1E,gBAAgB,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC;IACzC,qEAAqE;IACrE,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC;CACzC,CAAC,CAAC;AAIH,+EAA+E;AAC/E,8BAA8B;AAC9B,+EAA+E;AAE/E;;;GAGG;AACH,MAAM,CAAC,MAAM,cAAc,GAAG;IAC5B,OAAO,EAAE,iDAAiD;IAC1D,UAAU,EAAE,qBAAqB;IACjC,QAAQ,EAAE,uCAAuC;IACjD,SAAS,EAAE,iBAAiB;IAC5B,QAAQ,EAAE,iBAAiB;IAC3B,UAAU,EAAE,iBAAiB;CACrB,CAAC;AAiBX,MAAM,UAAU,qBAAqB;IACnC,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,EAAE,EAAE,CAAC;AACtD,CAAC;AAED,MAAM,UAAU,iBAAiB;IAC/B,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,CAAC,cAAc,CAAC,EAAE,CAAC;AACpE,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,iBAAiB;IAC/B,OAAO,EAAE,KAAK,EAAE,UAAU,EAAE,UAAU,EAAE,IAAI,GAAG,CAAC,cAAc,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC;AAChF,CAAC;AAED,2EAA2E;AAC3E,MAAM,UAAU,UAAU,CAAC,CAAY;IACrC,OAAO,CAAC,CAAC,UAAU,CAAC,IAAI,KAAK,CAAC,CAAC;AACjC,CAAC;AAED,+EAA+E;AAC/E,0CAA0C;AAC1C,+EAA+E;AAE/E,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,iDAAiD;IACjD,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE;IACpB,0EAA0E;IAC1E,MAAM,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC7B,uDAAuD;IACvD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC9B,kDAAkD;IAClD,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAC5B,iDAAiD;IACjD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;IAChC,2DAA2D;IAC3D,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE;IACrB,mFAAmF;IACnF,UAAU,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,QAAQ,EAAE;IAClC,+DAA+D;IAC/D,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CAC7B,CAAC,CAAC;AAIH,+EAA+E;AAC/E,MAAM,CAAC,MAAM,kBAAkB,GAAG;IAChC,QAAQ;IACR,SAAS;IACT,OAAO;IACP,WAAW;IACX,WAAW;CACH,CAAC;AAGX,+EAA+E;AAC/E,4EAA4E;AAC5E,+EAA+E;AAE/E;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,kBAAkB,GAAG;IAChC,aAAa,EAAE,kDAAkD;IACjE,SAAS,EAAE,iEAAiE;IAC5E,YAAY,EAAE,mEAAmE;IACjF,iBAAiB,EAAE,kEAAkE;IACrF,UAAU,EAAE,wCAAwC;CAC5C,CAAC;AAGX;;;;GAIG;AACH,MAAM,CAAC,MAAM,uBAAuB,GAA+C;IACjF,WAAW,EAAE,GAAG;IAChB,OAAO,EAAE,GAAG;IACZ,UAAU,EAAE,IAAI;IAChB,eAAe,EAAE,IAAI;IACrB,QAAQ,EAAE,GAAG;CACd,CAAC;AAEF,MAAM,CAAC,MAAM,kBAAkB,GAAG,CAAC,CAAC,MAAM,CAAC;IACzC,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACvC,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACnC,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACtC,eAAe,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IAC3C,QAAQ,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACpC,2EAA2E;IAC3E,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC;IACrC,2FAA2F;IAC3F,MAAM,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,OAAO,EAAE,WAAW,EAAE,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC;IACrE,uDAAuD;IACvD,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE;CACjC,CAAC,CAAC;AAIH;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAC9B,IAAsC,EACtC,UAAqD,uBAAuB;IAE5E,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,MAAM,CAAC,IAAI,kBAAkB,EAAE,CAAC;QACnC,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC;QAC1B,IAAI,CAAC,IAAI,CAAC;YAAE,SAAS;QACrB,MAAM,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC7E,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC;QACb,IAAI,IAAI,CAAC,CAAC;IACZ,CAAC;IACD,OAAO,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;AACnC,CAAC"}

package/dist/bin/cli.js

@@ -43,6 +43,7 @@ const lazy = {
     handsfree: () => import('../cli/handsfree.js').then((m) => m.handsfreeCommand),
     proxy: () => import('../cli/proxy.js').then((m) => m.proxyCommand),
     benchPaired: () => import('../cli/bench.js').then((m) => m.benchPairedCommand),
+    tune: () => import('../cli/self-tuning.js').then((m) => m.tuneCommand),
     sandbox: () => import('../cli/sandbox.js').then((m) => m.sandboxCommand),
     design: () => import('../cli/design.js').then((m) => m.designCommand),
     challenge: () => import('../cli/challenge.js').then((m) => m.challengeCommand),
@@ -659,6 +660,35 @@ selfHarness
     const cmd = await lazy.selfHarness();
     await cmd('prune', options);
 });
+// LLM Self-Tuning — raise a small model toward Opus by tuning UAP flags with a
+// benchmark-validated, LLM-guided (or GP-BO) closed loop.
+function addTuneOptions(cmd) {
+    return cmd
+        .option('--model <id>', 'Executor model family to tune (default qwen36-a3b)')
+        .option('--suite <dir>', 'Real-gate task suite (default benchmarks/suites/real-gate)')
+        .option('--adapter <name>', 'Agent adapter: mock | opencode | claude | mini | raw | deliver (default mock)')
+        .option('--judge <id>', 'Judge/tuner model id (else recipes.judge.model, else GP-only)')
+        .option('--epochs <n>', 'Paired seeds per (task, arm) (default 5)')
+        .option('--concurrency <n>', 'Max concurrent cells (default 4)')
+        .option('--max-iterations <n>', 'Max tuning-loop iterations (default 6)')
+        .option('--phase <name>', 'Force one search phase: coarse | medium | fine | combinatorial')
+        .option('--iterations <n>', 'Bootstrap/permutation iterations for the paired stats (default 10000)')
+        .option('--seed <n>', 'RNG/statistics seed (default 1)')
+        .option('--apply', 'Commit accepted configs to .uap.json / proxy.env + save the profile (default: dry-run)')
+        .option('--json', 'Emit JSON instead of a human-readable report');
+}
+addTuneOptions(program
+    .command('tune')
+    .description('Self-tune UAP flags to raise a small model toward Opus (propose -> validate -> decide -> learn)')).action(async (options) => {
+    const cmd = await lazy.tune();
+    await cmd(options);
+});
+addTuneOptions(selfHarness
+    .command('tune')
+    .description('Alias of `uap tune`: the LLM/GP flag self-tuning loop')).action(async (options) => {
+    const cmd = await lazy.tune();
+    await cmd(options);
+});
 // Open-collider divergent-ideation commands
 const ideate = program
     .command('ideate')