@miller-tech/uap 1.136.1 → 1.137.1

package/dist/self-tuning/quality-scorer.js

@@ -0,0 +1,180 @@
+/**
+ * LLM Self-Tuning — the Quality Scorer (P0).
+ *
+ * The paired benchmark answers "did the code work" (pass/fail). Raising a small
+ * model toward Opus 4.8 is a QUALITY problem, not just a correctness one, so the
+ * tuner needs a richer signal. This module produces a `QualityScore`
+ * (correctness, quality, efficiency, toolReliability, planning + composite) per
+ * run so the tuner has something continuous to optimize.
+ *
+ * Two production paths, chosen by whether a judge model is available:
+ *   - HEURISTIC (no judge): every dimension derived deterministically from the
+ *     metric vector + ground-truth `correct`. Zero cost, fully reproducible,
+ *     network-free — the default and the test path.
+ *   - JUDGE (a stronger model configured): correctness/quality/planning graded
+ *     by the judge over the task spec + agent output; efficiency/toolReliability
+ *     stay deterministic (they are metric facts, not opinions). The two are
+ *     fused into a 'hybrid' score.
+ *
+ * The composite weighting (correctness 40 / quality 30 / efficiency 15 /
+ * toolReliability 15) matches docs/design/LLM_SELF_TUNING_ANALYSIS.md §3.3.1.
+ */
+import { QualityScoreSchema, compositeQuality, DEFAULT_QUALITY_WEIGHTS, } from '../benchmarks/paired/types.js';
+import { parseJsonLenient } from './judge.js';
+const DEFAULT_REF_TOKENS = 15000;
+const DEFAULT_REF_TURNS = 6;
+function clamp100(n) {
+    return Math.max(0, Math.min(100, n));
+}
+/**
+ * Deterministic efficiency (0-100). An INCORRECT run spent its tokens on a wrong
+ * answer, so efficiency is heavily penalized (tokens were wasted). A correct run
+ * scores by leanness against `refTokens`.
+ */
+export function efficiencyScore(metrics, refTokens = DEFAULT_REF_TOKENS) {
+    const tokens = metrics.tokens;
+    if (tokens == null)
+        return metrics.correct ? 60 : 20; // unknown spend: neutral-ish
+    const leanness = (refTokens / (refTokens + Math.max(0, tokens))) * 100; // 100→50→0
+    return clamp100(metrics.correct ? leanness : leanness * 0.3);
+}
+/**
+ * Deterministic tool reliability (0-100) from well-formedness + productivity.
+ * A malformed edit is the small-model failure mode this dimension surfaces.
+ */
+export function toolReliabilityScore(metrics) {
+    let base;
+    if (metrics.wellFormed === false)
+        base = 30;
+    else if (metrics.wellFormed === true)
+        base = 90;
+    else
+        base = 70; // unknown
+    // Productivity: a run that made tool calls but never got correct is churning.
+    const productivity = metrics.correct ? 1 : 0.7;
+    // Penalize a run that made no tool calls yet was expected to act.
+    if (metrics.toolCalls != null && metrics.toolCalls === 0 && !metrics.correct)
+        base *= 0.7;
+    return clamp100(base * productivity);
+}
+/** Deterministic planning heuristic (0-100): correct + few turns == coherent. */
+export function planningScore(metrics, refTurns = DEFAULT_REF_TURNS) {
+    const turns = metrics.turns;
+    const floor = metrics.correct ? 60 : 25;
+    if (turns == null)
+        return floor;
+    // Fewer turns than the reference reads as tighter planning; more reads looser.
+    const turnFactor = refTurns / (refTurns + Math.max(0, turns - refTurns));
+    return clamp100(floor + (metrics.correct ? 40 : 15) * turnFactor);
+}
+/**
+ * Purely deterministic score — no network. Correctness is the ground-truth
+ * boolean projected to 0/100; quality is proxied from correctness + wellFormed;
+ * the rest come from the heuristics above. This is the default and test path.
+ */
+export function heuristicQuality(input, opts = {}) {
+    const m = input.metrics;
+    const correctness = input.correct ? 100 : 0;
+    // Quality proxy: a correct, well-formed answer is decent; without a judge we
+    // cannot see nuance, so we stay conservative (never award judge-level highs).
+    const qualityProxy = input.correct ? (m.wellFormed === false ? 60 : 78) : m.wellFormed === true ? 30 : 20;
+    const dims = {
+        correctness,
+        quality: qualityProxy,
+        efficiency: efficiencyScore(m, opts.refTokens),
+        toolReliability: toolReliabilityScore(m),
+        planning: planningScore(m, opts.refTurns),
+    };
+    const composite = compositeQuality(dims, opts.weights ?? DEFAULT_QUALITY_WEIGHTS);
+    return QualityScoreSchema.parse({ ...dims, composite, source: 'heuristic' });
+}
+/** Build the judge prompt: grade correctness/quality/planning as 0-100 JSON. */
+export function buildJudgePrompt(input, maxOutputChars) {
+    const output = (input.output ?? '').slice(0, maxOutputChars);
+    return [
+        'You are a strict senior engineer grading an AI agent\'s work on a task.',
+        'Grade three dimensions on a 0-100 scale, where 100 is what a top-tier model',
+        '(Claude Opus) would produce and 50 is mediocre-but-acceptable.',
+        '',
+        'DIMENSIONS:',
+        '- correctness: does the work actually satisfy the task as specified?',
+        '- quality: is the solution clean, idiomatic, complete, and free of shortcuts?',
+        '- planning: is the approach coherent and well-sequenced (vs flailing)?',
+        '',
+        `GROUND TRUTH (from a hidden deterministic test): the task was ${input.correct ? 'RESOLVED' : 'NOT resolved'}.`,
+        'Weight correctness heavily toward that ground truth, but you may still judge',
+        'quality/planning on partial or over-engineered work.',
+        '',
+        'TASK INSTRUCTION:',
+        input.taskInstruction,
+        ...(input.reference ? ['', 'REFERENCE / RUBRIC:', input.reference] : []),
+        '',
+        'AGENT OUTPUT (diff / answer / log tail):',
+        output || '(no output captured)',
+        '',
+        'Respond with ONLY a JSON object, no prose:',
+        '{"correctness": <0-100>, "quality": <0-100>, "planning": <0-100>, "rationale": "<one sentence>"}',
+    ].join('\n');
+}
+/**
+ * Score a single run. Uses the judge for correctness/quality/planning when one
+ * is supplied and reachable; efficiency/toolReliability are always the
+ * deterministic metric facts. Any judge failure degrades gracefully to the
+ * heuristic score (never throws), so a scoring pass can never wedge a bench.
+ */
+export async function scoreQuality(input, opts = {}) {
+    const heuristic = heuristicQuality(input, opts);
+    if (!opts.judge)
+        return heuristic;
+    const maxOut = opts.maxOutputChars ?? 6000;
+    let verdict = null;
+    try {
+        const raw = await opts.judge.complete(buildJudgePrompt(input, maxOut), { json: true, temperature: 0.1 });
+        verdict = parseJsonLenient(raw);
+    }
+    catch {
+        verdict = null; // network/parse failure → heuristic fallback
+    }
+    if (!verdict || typeof verdict.correctness !== 'number')
+        return heuristic;
+    const num = (v, fallback) => typeof v === 'number' && Number.isFinite(v) ? clamp100(v) : fallback;
+    const dims = {
+        // Anchor judged correctness to ground truth: a judge that contradicts the
+        // hidden test is pulled toward it (70/30) so a hallucinated pass can't win.
+        correctness: clamp100(0.3 * num(verdict.correctness, input.correct ? 100 : 0) + 0.7 * (input.correct ? 100 : 0)),
+        quality: num(verdict.quality, heuristic.quality),
+        efficiency: heuristic.efficiency,
+        toolReliability: heuristic.toolReliability,
+        planning: num(verdict.planning, heuristic.planning),
+    };
+    const composite = compositeQuality(dims, opts.weights ?? DEFAULT_QUALITY_WEIGHTS);
+    return QualityScoreSchema.parse({
+        ...dims,
+        composite,
+        source: 'hybrid',
+        ...(verdict.rationale ? { rationale: String(verdict.rationale).slice(0, 300) } : {}),
+    });
+}
+/**
+ * Attach quality scores to a set of run records in place-returning form. The
+ * `inputFor` mapper supplies the per-record evidence (task instruction, output).
+ * Scoring is bounded-concurrent to avoid overwhelming the judge endpoint.
+ */
+export async function scoreRuns(records, inputFor, opts = {}) {
+    const concurrency = Math.max(1, opts.concurrency ?? (opts.judge ? 4 : 16));
+    const out = new Array(records.length);
+    let next = 0;
+    async function worker() {
+        for (;;) {
+            const i = next++;
+            if (i >= records.length)
+                return;
+            const r = records[i];
+            const score = await scoreQuality(inputFor(r), opts);
+            out[i] = { ...r, qualityScore: score };
+        }
+    }
+    await Promise.all(Array.from({ length: Math.min(concurrency, records.length) }, () => worker()));
+    return out;
+}
+//# sourceMappingURL=quality-scorer.js.map

package/dist/self-tuning/quality-scorer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"quality-scorer.js","sourceRoot":"","sources":["../../src/self-tuning/quality-scorer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,EAEL,kBAAkB,EAElB,gBAAgB,EAChB,uBAAuB,GAGxB,MAAM,+BAA+B,CAAC;AAEvC,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAsC9C,MAAM,kBAAkB,GAAG,KAAK,CAAC;AACjC,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B,SAAS,QAAQ,CAAC,CAAS;IACzB,OAAO,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;AACvC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAAC,OAAqB,EAAE,SAAS,GAAG,kBAAkB;IACnF,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IAC9B,IAAI,MAAM,IAAI,IAAI;QAAE,OAAO,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,6BAA6B;IACnF,MAAM,QAAQ,GAAG,CAAC,SAAS,GAAG,CAAC,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,MAAM,CAAC,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,WAAW;IACnF,OAAO,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,QAAQ,GAAG,GAAG,CAAC,CAAC;AAC/D,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,oBAAoB,CAAC,OAAqB;IACxD,IAAI,IAAY,CAAC;IACjB,IAAI,OAAO,CAAC,UAAU,KAAK,KAAK;QAAE,IAAI,GAAG,EAAE,CAAC;SACvC,IAAI,OAAO,CAAC,UAAU,KAAK,IAAI;QAAE,IAAI,GAAG,EAAE,CAAC;;QAC3C,IAAI,GAAG,EAAE,CAAC,CAAC,UAAU;IAC1B,8EAA8E;IAC9E,MAAM,YAAY,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IAC/C,kEAAkE;IAClE,IAAI,OAAO,CAAC,SAAS,IAAI,IAAI,IAAI,OAAO,CAAC,SAAS,KAAK,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO;QAAE,IAAI,IAAI,GAAG,CAAC;IAC1F,OAAO,QAAQ,CAAC,IAAI,GAAG,YAAY,CAAC,CAAC;AACvC,CAAC;AAED,iFAAiF;AACjF,MAAM,UAAU,aAAa,CAAC,OAAqB,EAAE,QAAQ,GAAG,iBAAiB;IAC/E,MAAM,KAAK,GAAG,OAAO,CAAC,KAAK,CAAC;IAC5B,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACxC,IAAI,KAAK,IAAI,IAAI;QAAE,OAAO,KAAK,CAAC;IAChC,+EAA+E;IAC/E,MAAM,UAAU,GAAG,QAAQ,GAAG,CAAC,QAAQ,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,KAAK,GAAG,QAAQ,CAAC,CAAC,CAAC;IACzE,OAAO,QAAQ,CAAC,KAAK,GAAG,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,CAAC;AACpE,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,KAAwB,EAAE,OAAqB,EAAE;IAChF,MAAM,CAAC,GAAG,KAAK,CAAC,OAAO,CAAC;IACxB,MAAM,WAAW,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5C,6EAA6E;IAC7E,8EAA8E;IAC9E,MAAM,YAAY,GAAG,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,KAAK,KAAK,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU,KAAK,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IAC1G,MAAM,IAAI,GAAqC;QAC7C,WAAW;QACX,OAAO,EAAE,YAAY;QACrB,UAAU,EAAE,eAAe,CAAC,CAAC,EAAE,IAAI,CAAC,SAAS,CAAC;QAC9C,eAAe,EAAE,oBAAoB,CAAC,CAAC,CAAC;QACxC,QAAQ,EAAE,aAAa,CAAC,CAAC,EAAE,IAAI,CAAC,QAAQ,CAAC;KAC1C,CAAC;IACF,MAAM,SAAS,GAAG,gBAAgB,CAAC,IAAI,EAAE,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,CAAC;IAClF,OAAO,kBAAkB,CAAC,KAAK,CAAC,EAAE,GAAG,IAAI,EAAE,SAAS,EAAE,MAAM,EAAE,WAAW,EAAE,CAAC,CAAC;AAC/E,CAAC;AAED,gFAAgF;AAChF,MAAM,UAAU,gBAAgB,CAAC,KAAwB,EAAE,cAAsB;IAC/E,MAAM,MAAM,GAAG,CAAC,KAAK,CAAC,MAAM,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,cAAc,CAAC,CAAC;IAC7D,OAAO;QACL,yEAAyE;QACzE,6EAA6E;QAC7E,gEAAgE;QAChE,EAAE;QACF,aAAa;QACb,sEAAsE;QACtE,+EAA+E;QAC/E,wEAAwE;QACxE,EAAE;QACF,iEAAiE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,UAAU,CAAC,CAAC,CAAC,cAAc,GAAG;QAC/G,8EAA8E;QAC9E,sDAAsD;QACtD,EAAE;QACF,mBAAmB;QACnB,KAAK,CAAC,eAAe;QACrB,GAAG,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,EAAE,qBAAqB,EAAE,KAAK,CAAC,SAAS,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC;QACxE,EAAE;QACF,0CAA0C;QAC1C,MAAM,IAAI,sBAAsB;QAChC,EAAE;QACF,4CAA4C;QAC5C,kGAAkG;KACnG,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AASD;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,KAAwB,EACxB,OAAqB,EAAE;IAEvB,MAAM,SAAS,GAAG,gBAAgB,CAAC,KAAK,EAAE,IAAI,CAAC,CAAC;IAChD,IAAI,CAAC,IAAI,CAAC,KAAK;QAAE,OAAO,SAAS,CAAC;IAElC,MAAM,MAAM,GAAG,IAAI,CAAC,cAAc,IAAI,IAAI,CAAC;IAC3C,IAAI,OAAO,GAAwB,IAAI,CAAC;IACxC,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,KAAK,EAAE,MAAM,CAAC,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC,CAAC;QACzG,OAAO,GAAG,gBAAgB,CAAe,GAAG,CAAC,CAAC;IAChD,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,GAAG,IAAI,CAAC,CAAC,6CAA6C;IAC/D,CAAC;IACD,IAAI,CAAC,OAAO,IAAI,OAAO,OAAO,CAAC,WAAW,KAAK,QAAQ;QAAE,OAAO,SAAS,CAAC;IAE1E,MAAM,GAAG,GAAG,CAAC,CAAU,EAAE,QAAgB,EAAU,EAAE,CACnD,OAAO,CAAC,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC;IAEvE,MAAM,IAAI,GAAqC;QAC7C,0EAA0E;QAC1E,4EAA4E;QAC5E,WAAW,EAAE,QAAQ,CAAC,GAAG,GAAG,GAAG,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAChH,OAAO,EAAE,GAAG,CAAC,OAAO,CAAC,OAAO,EAAE,SAAS,CAAC,OAAO,CAAC;QAChD,UAAU,EAAE,SAAS,CAAC,UAAU;QAChC,eAAe,EAAE,SAAS,CAAC,eAAe;QAC1C,QAAQ,EAAE,GAAG,CAAC,OAAO,CAAC,QAAQ,EAAE,SAAS,CAAC,QAAQ,CAAC;KACpD,CAAC;IACF,MAAM,SAAS,GAAG,gBAAgB,CAAC,IAAI,EAAE,IAAI,CAAC,OAAO,IAAI,uBAAuB,CAAC,CAAC;IAClF,OAAO,kBAAkB,CAAC,KAAK,CAAC;QAC9B,GAAG,IAAI;QACP,SAAS;QACT,MAAM,EAAE,QAAQ;QAChB,GAAG,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;KACrF,CAAC,CAAC;AACL,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,SAAS,CAC7B,OAAoB,EACpB,QAA6C,EAC7C,OAAgD,EAAE;IAElD,MAAM,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,WAAW,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC3E,MAAM,GAAG,GAAgB,IAAI,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC;IACnD,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,UAAU,MAAM;QACnB,SAAS,CAAC;YACR,MAAM,CAAC,GAAG,IAAI,EAAE,CAAC;YACjB,IAAI,CAAC,IAAI,OAAO,CAAC,MAAM;gBAAE,OAAO;YAChC,MAAM,CAAC,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;YACrB,MAAM,KAAK,GAAG,MAAM,YAAY,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC;YACpD,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,YAAY,EAAE,KAAK,EAAE,CAAC;QACzC,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,OAAO,CAAC,MAAM,CAAC,EAAE,EAAE,GAAG,EAAE,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC;IACjG,OAAO,GAAG,CAAC;AACb,CAAC"}

package/dist/self-tuning/realtime-adaptor.d.ts

@@ -0,0 +1,78 @@
+/**
+ * LLM Self-Tuning — the real-time adaptor (P4).
+ *
+ * Offline tuning (P0-P3) finds a good STATIC config. This closes the last gap:
+ * per-session adaptation from LIVE signals, for the failures a static config
+ * can't foresee (design §Phase 4). It maps four session signals to conservative,
+ * hot-reloadable adjustments and emits them on the adaptation-signal channel the
+ * proxy reads per request:
+ *
+ *   - tool-use failure rate ↑           → escalate this turn to the judge (fusion)
+ *   - per-turn quality ↓                → escalate this turn to the judge (fusion)
+ *   - context-window utilization ↑      → converge sooner (lower recon threshold)
+ *   - RECON no-write streak ↑           → force synthesis / deliver now
+ *
+ * `computeAdaptation` is pure and returns null when everything is nominal (no
+ * signal written → the proxy keeps its startup config). Emission is opt-in
+ * (`enabled`, default from UAP_REALTIME_ADAPT) so it can never perturb a session
+ * that didn't ask for it.
+ */
+import { AdaptationSignal } from '../coordination/adaptation-signal.js';
+import { FlagConfig } from './flags.js';
+/** Live session signals (any subset). Absent fields are treated as nominal. */
+export interface SessionSignals {
+    /** Fraction of recent tool calls that failed / were malformed (0-1). */
+    toolFailureRate?: number;
+    /** Fraction of the context window in use (0-1), e.g. from GET /v1/context. */
+    contextUtilization?: number;
+    /** Consecutive no-write ("RECON") turns. */
+    reconStreak?: number;
+    /** Recent per-turn composite quality estimate (0-100). */
+    turnQuality?: number;
+}
+/** Breach thresholds. Defaults are conservative — adapt only on a real signal. */
+export interface AdaptationThresholds {
+    /** Escalate when tool-failure rate exceeds this. Default 0.35. */
+    toolFailureRate?: number;
+    /** Converge sooner when context utilization exceeds this. Default 0.85. */
+    contextUtilization?: number;
+    /** Force synthesis when the no-write streak exceeds this. Default 25. */
+    reconStreak?: number;
+    /** Escalate when turn quality drops below this. Default 45. */
+    turnQualityFloor?: number;
+}
+/**
+ * Map live signals + the current config to an adaptation signal, or null when
+ * nothing breaches a threshold. Pure and deterministic.
+ */
+export declare function computeAdaptation(sessionId: string, signals: SessionSignals, current: FlagConfig, now: number, thresholds?: AdaptationThresholds): AdaptationSignal | null;
+export interface EmitOptions extends AdaptationThresholds {
+    /** Explicit master switch. Default: auto-on unless opted out (env/config). */
+    enabled?: boolean;
+    /** Project dir for the config opt-out check. Default process.cwd(). */
+    cwd?: string;
+    /** Signal directory override (tests). */
+    dir?: string;
+    /** "now" in unix seconds (host may lack Date). */
+    now?: number;
+}
+/**
+ * True when real-time adaptation is enabled. AUTO-ON by default; opt out via
+ * either `UAP_REALTIME_ADAPT` (0/false/off) or `.uap.json` `realtimeAdapt.enabled:
+ * false`. Precedence: explicit arg → env (if set) → config → default(true).
+ * The emitter is the effective master switch — with it off, no signal is written
+ * so the proxy has nothing to honor, regardless of PROXY_REALTIME_ADAPT.
+ */
+export declare function realtimeAdaptEnabled(explicit?: boolean, cwd?: string): boolean;
+/**
+ * Compute + emit an adaptation signal for a session. No-op (returns null) when
+ * disabled or when nothing breaches a threshold. Never throws.
+ */
+export declare function emitAdaptation(sessionId: string, signals: SessionSignals, current: FlagConfig, opts?: EmitOptions): AdaptationSignal | null;
+/**
+ * Best-effort read of the proxy's per-session context state (GET /v1/context) to
+ * populate `contextUtilization`. Injectable fetch for tests. Returns {} on any
+ * failure (the adaptor treats missing signals as nominal).
+ */
+export declare function fetchSessionContext(endpoint?: string, fetchImpl?: typeof fetch): Promise<Pick<SessionSignals, 'contextUtilization'>>;
+//# sourceMappingURL=realtime-adaptor.d.ts.map

package/dist/self-tuning/realtime-adaptor.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"realtime-adaptor.d.ts","sourceRoot":"","sources":["../../src/self-tuning/realtime-adaptor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EACL,gBAAgB,EAEjB,MAAM,sCAAsC,CAAC;AAE9C,OAAO,EAAE,UAAU,EAAa,MAAM,YAAY,CAAC;AAEnD,+EAA+E;AAC/E,MAAM,WAAW,cAAc;IAC7B,wEAAwE;IACxE,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,8EAA8E;IAC9E,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,4CAA4C;IAC5C,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,0DAA0D;IAC1D,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB;AAED,kFAAkF;AAClF,MAAM,WAAW,oBAAoB;IACnC,kEAAkE;IAClE,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,2EAA2E;IAC3E,kBAAkB,CAAC,EAAE,MAAM,CAAC;IAC5B,yEAAyE;IACzE,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,+DAA+D;IAC/D,gBAAgB,CAAC,EAAE,MAAM,CAAC;CAC3B;AASD;;;GAGG;AACH,wBAAgB,iBAAiB,CAC/B,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,UAAU,EACnB,GAAG,EAAE,MAAM,EACX,UAAU,GAAE,oBAAyB,GACpC,gBAAgB,GAAG,IAAI,CAiCzB;AAED,MAAM,WAAW,WAAY,SAAQ,oBAAoB;IACvD,8EAA8E;IAC9E,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,uEAAuE;IACvE,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,yCAAyC;IACzC,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,kDAAkD;IAClD,GAAG,CAAC,EAAE,MAAM,CAAC;CACd;AAED;;;;;;GAMG;AACH,wBAAgB,oBAAoB,CAAC,QAAQ,CAAC,EAAE,OAAO,EAAE,GAAG,CAAC,EAAE,MAAM,GAAG,OAAO,CAa9E;AAED;;;GAGG;AACH,wBAAgB,cAAc,CAC5B,SAAS,EAAE,MAAM,EACjB,OAAO,EAAE,cAAc,EACvB,OAAO,EAAE,UAAU,EACnB,IAAI,GAAE,WAAgB,GACrB,gBAAgB,GAAG,IAAI,CAYzB;AAED;;;;GAIG;AACH,wBAAsB,mBAAmB,CACvC,QAAQ,CAAC,EAAE,MAAM,EACjB,SAAS,GAAE,OAAO,KAAa,GAC9B,OAAO,CAAC,IAAI,CAAC,cAAc,EAAE,oBAAoB,CAAC,CAAC,CAYrD"}

package/dist/self-tuning/realtime-adaptor.js

@@ -0,0 +1,127 @@
+/**
+ * LLM Self-Tuning — the real-time adaptor (P4).
+ *
+ * Offline tuning (P0-P3) finds a good STATIC config. This closes the last gap:
+ * per-session adaptation from LIVE signals, for the failures a static config
+ * can't foresee (design §Phase 4). It maps four session signals to conservative,
+ * hot-reloadable adjustments and emits them on the adaptation-signal channel the
+ * proxy reads per request:
+ *
+ *   - tool-use failure rate ↑           → escalate this turn to the judge (fusion)
+ *   - per-turn quality ↓                → escalate this turn to the judge (fusion)
+ *   - context-window utilization ↑      → converge sooner (lower recon threshold)
+ *   - RECON no-write streak ↑           → force synthesis / deliver now
+ *
+ * `computeAdaptation` is pure and returns null when everything is nominal (no
+ * signal written → the proxy keeps its startup config). Emission is opt-in
+ * (`enabled`, default from UAP_REALTIME_ADAPT) so it can never perturb a session
+ * that didn't ask for it.
+ */
+import { writeAdaptationSignal, } from '../coordination/adaptation-signal.js';
+import { loadUapConfigRaw } from '../utils/config-loader.js';
+import { flagValue } from './flags.js';
+const DEFAULTS = {
+    toolFailureRate: 0.35,
+    contextUtilization: 0.85,
+    reconStreak: 25,
+    turnQualityFloor: 45,
+};
+/**
+ * Map live signals + the current config to an adaptation signal, or null when
+ * nothing breaches a threshold. Pure and deterministic.
+ */
+export function computeAdaptation(sessionId, signals, current, now, thresholds = {}) {
+    const thr = { ...DEFAULTS, ...thresholds };
+    const reasons = [];
+    const out = { ts: now, sessionId, reason: '' };
+    // Quality / reliability degradation → escalate this turn to the judge.
+    if (signals.toolFailureRate != null && signals.toolFailureRate > thr.toolFailureRate) {
+        out.escalate = true;
+        out.recipe = 'fusion';
+        reasons.push(`tool-failure ${(signals.toolFailureRate * 100).toFixed(0)}% > ${(thr.toolFailureRate * 100).toFixed(0)}%`);
+    }
+    if (signals.turnQuality != null && signals.turnQuality < thr.turnQualityFloor) {
+        out.escalate = true;
+        out.recipe = 'fusion';
+        reasons.push(`turn-quality ${signals.turnQuality.toFixed(0)} < ${thr.turnQualityFloor}`);
+    }
+    // Context pressure → converge sooner (lower the recon threshold live).
+    if (signals.contextUtilization != null && signals.contextUtilization > thr.contextUtilization) {
+        const cur = Number(flagValue(current, 'PROXY_RECON_CONVERGENCE_THRESHOLD')) || 40;
+        out.reconThreshold = Math.max(20, Math.min(cur, Math.round(cur * 0.5)));
+        reasons.push(`ctx ${(signals.contextUtilization * 100).toFixed(0)}% > ${(thr.contextUtilization * 100).toFixed(0)}% → recon≤${out.reconThreshold}`);
+    }
+    // RECON loop → force synthesis / deliver now.
+    if (signals.reconStreak != null && signals.reconStreak > thr.reconStreak) {
+        out.forceSynthesis = true;
+        reasons.push(`recon-streak ${signals.reconStreak} > ${thr.reconStreak}`);
+    }
+    if (reasons.length === 0)
+        return null; // nominal — do not emit
+    out.reason = reasons.join('; ');
+    return out;
+}
+/**
+ * True when real-time adaptation is enabled. AUTO-ON by default; opt out via
+ * either `UAP_REALTIME_ADAPT` (0/false/off) or `.uap.json` `realtimeAdapt.enabled:
+ * false`. Precedence: explicit arg → env (if set) → config → default(true).
+ * The emitter is the effective master switch — with it off, no signal is written
+ * so the proxy has nothing to honor, regardless of PROXY_REALTIME_ADAPT.
+ */
+export function realtimeAdaptEnabled(explicit, cwd) {
+    if (explicit !== undefined)
+        return explicit;
+    const v = (process.env.UAP_REALTIME_ADAPT ?? '').toLowerCase();
+    if (v)
+        return v === '1' || v === 'true' || v === 'on' || v === 'yes';
+    try {
+        const raw = loadUapConfigRaw(cwd ?? process.cwd());
+        if (raw?.realtimeAdapt?.enabled === false)
+            return false;
+    }
+    catch {
+        /* fall through to default */
+    }
+    return true; // auto-on
+}
+/**
+ * Compute + emit an adaptation signal for a session. No-op (returns null) when
+ * disabled or when nothing breaches a threshold. Never throws.
+ */
+export function emitAdaptation(sessionId, signals, current, opts = {}) {
+    if (!realtimeAdaptEnabled(opts.enabled, opts.cwd))
+        return null;
+    const now = opts.now ?? Date.now() / 1000;
+    const sig = computeAdaptation(sessionId, signals, current, now, opts);
+    if (sig) {
+        try {
+            writeAdaptationSignal(sig, opts.dir);
+        }
+        catch {
+            /* fail open */
+        }
+    }
+    return sig;
+}
+/**
+ * Best-effort read of the proxy's per-session context state (GET /v1/context) to
+ * populate `contextUtilization`. Injectable fetch for tests. Returns {} on any
+ * failure (the adaptor treats missing signals as nominal).
+ */
+export async function fetchSessionContext(endpoint, fetchImpl = fetch) {
+    const base = (endpoint ?? process.env.UAP_INFERENCE_ENDPOINT ?? 'http://localhost:4000/v1').replace(/\/$/, '');
+    try {
+        const res = await fetchImpl(`${base}/context`);
+        if (!res.ok)
+            return {};
+        const data = (await res.json());
+        let util = data.utilization;
+        if (util == null && data.used != null && data.window)
+            util = data.used / data.window;
+        return util != null ? { contextUtilization: Math.max(0, Math.min(1, util)) } : {};
+    }
+    catch {
+        return {};
+    }
+}
+//# sourceMappingURL=realtime-adaptor.js.map

package/dist/self-tuning/realtime-adaptor.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"realtime-adaptor.js","sourceRoot":"","sources":["../../src/self-tuning/realtime-adaptor.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAEL,qBAAqB,GACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAC7D,OAAO,EAAc,SAAS,EAAE,MAAM,YAAY,CAAC;AA0BnD,MAAM,QAAQ,GAAmC;IAC/C,eAAe,EAAE,IAAI;IACrB,kBAAkB,EAAE,IAAI;IACxB,WAAW,EAAE,EAAE;IACf,gBAAgB,EAAE,EAAE;CACrB,CAAC;AAEF;;;GAGG;AACH,MAAM,UAAU,iBAAiB,CAC/B,SAAiB,EACjB,OAAuB,EACvB,OAAmB,EACnB,GAAW,EACX,aAAmC,EAAE;IAErC,MAAM,GAAG,GAAG,EAAE,GAAG,QAAQ,EAAE,GAAG,UAAU,EAAE,CAAC;IAC3C,MAAM,OAAO,GAAa,EAAE,CAAC;IAC7B,MAAM,GAAG,GAAqB,EAAE,EAAE,EAAE,GAAG,EAAE,SAAS,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;IAEjE,uEAAuE;IACvE,IAAI,OAAO,CAAC,eAAe,IAAI,IAAI,IAAI,OAAO,CAAC,eAAe,GAAG,GAAG,CAAC,eAAe,EAAE,CAAC;QACrF,GAAG,CAAC,QAAQ,GAAG,IAAI,CAAC;QACpB,GAAG,CAAC,MAAM,GAAG,QAAQ,CAAC;QACtB,OAAO,CAAC,IAAI,CAAC,gBAAgB,CAAC,OAAO,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,eAAe,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAC3H,CAAC;IACD,IAAI,OAAO,CAAC,WAAW,IAAI,IAAI,IAAI,OAAO,CAAC,WAAW,GAAG,GAAG,CAAC,gBAAgB,EAAE,CAAC;QAC9E,GAAG,CAAC,QAAQ,GAAG,IAAI,CAAC;QACpB,GAAG,CAAC,MAAM,GAAG,QAAQ,CAAC;QACtB,OAAO,CAAC,IAAI,CAAC,gBAAgB,OAAO,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,GAAG,CAAC,gBAAgB,EAAE,CAAC,CAAC;IAC3F,CAAC;IAED,uEAAuE;IACvE,IAAI,OAAO,CAAC,kBAAkB,IAAI,IAAI,IAAI,OAAO,CAAC,kBAAkB,GAAG,GAAG,CAAC,kBAAkB,EAAE,CAAC;QAC9F,MAAM,GAAG,GAAG,MAAM,CAAC,SAAS,CAAC,OAAO,EAAE,mCAAmC,CAAC,CAAC,IAAI,EAAE,CAAC;QAClF,GAAG,CAAC,cAAc,GAAG,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,GAAG,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;QACxE,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,kBAAkB,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,CAAC,kBAAkB,GAAG,GAAG,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,aAAa,GAAG,CAAC,cAAc,EAAE,CAAC,CAAC;IACtJ,CAAC;IAED,8CAA8C;IAC9C,IAAI,OAAO,CAAC,WAAW,IAAI,IAAI,IAAI,OAAO,CAAC,WAAW,GAAG,GAAG,CAAC,WAAW,EAAE,CAAC;QACzE,GAAG,CAAC,cAAc,GAAG,IAAI,CAAC;QAC1B,OAAO,CAAC,IAAI,CAAC,gBAAgB,OAAO,CAAC,WAAW,MAAM,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC;IAC3E,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,IAAI,CAAC,CAAC,wBAAwB;IAC/D,GAAG,CAAC,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAChC,OAAO,GAAG,CAAC;AACb,CAAC;AAaD;;;;;;GAMG;AACH,MAAM,UAAU,oBAAoB,CAAC,QAAkB,EAAE,GAAY;IACnE,IAAI,QAAQ,KAAK,SAAS;QAAE,OAAO,QAAQ,CAAC;IAC5C,MAAM,CAAC,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,kBAAkB,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IAC/D,IAAI,CAAC;QAAE,OAAO,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,MAAM,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,KAAK,KAAK,CAAC;IACrE,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,gBAAgB,CAAC,GAAG,IAAI,OAAO,CAAC,GAAG,EAAE,CAEzC,CAAC;QACT,IAAI,GAAG,EAAE,aAAa,EAAE,OAAO,KAAK,KAAK;YAAE,OAAO,KAAK,CAAC;IAC1D,CAAC;IAAC,MAAM,CAAC;QACP,6BAA6B;IAC/B,CAAC;IACD,OAAO,IAAI,CAAC,CAAC,UAAU;AACzB,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,cAAc,CAC5B,SAAiB,EACjB,OAAuB,EACvB,OAAmB,EACnB,OAAoB,EAAE;IAEtB,IAAI,CAAC,oBAAoB,CAAC,IAAI,CAAC,OAAO,EAAE,IAAI,CAAC,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC;IAC/D,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC;IAC1C,MAAM,GAAG,GAAG,iBAAiB,CAAC,SAAS,EAAE,OAAO,EAAE,OAAO,EAAE,GAAG,EAAE,IAAI,CAAC,CAAC;IACtE,IAAI,GAAG,EAAE,CAAC;QACR,IAAI,CAAC;YACH,qBAAqB,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QACvC,CAAC;QAAC,MAAM,CAAC;YACP,eAAe;QACjB,CAAC;IACH,CAAC;IACD,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,mBAAmB,CACvC,QAAiB,EACjB,YAA0B,KAAK;IAE/B,MAAM,IAAI,GAAG,CAAC,QAAQ,IAAI,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,0BAA0B,CAAC,CAAC,OAAO,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC;IAC/G,IAAI,CAAC;QACH,MAAM,GAAG,GAAG,MAAM,SAAS,CAAC,GAAG,IAAI,UAAU,CAAC,CAAC;QAC/C,IAAI,CAAC,GAAG,CAAC,EAAE;YAAE,OAAO,EAAE,CAAC;QACvB,MAAM,IAAI,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,EAAE,CAA6D,CAAC;QAC5F,IAAI,IAAI,GAAG,IAAI,CAAC,WAAW,CAAC;QAC5B,IAAI,IAAI,IAAI,IAAI,IAAI,IAAI,CAAC,IAAI,IAAI,IAAI,IAAI,IAAI,CAAC,MAAM;YAAE,IAAI,GAAG,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,MAAM,CAAC;QACrF,OAAO,IAAI,IAAI,IAAI,CAAC,CAAC,CAAC,EAAE,kBAAkB,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;IACpF,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,EAAE,CAAC;IACZ,CAAC;AACH,CAAC"}

package/dist/self-tuning/run.d.ts

@@ -0,0 +1,61 @@
+/**
+ * LLM Self-Tuning — the autonomous run loop + profile persistence (P2/P3).
+ *
+ * `runTuningLoop` drives multiple `runTuningIteration` steps under a search-phase
+ * schedule (coarse → medium → fine → combinatorial, design §3.3.2), commits each
+ * ACCEPTED config to `.uap.json`/`.uap/proxy.env` via the flag-writer (only when
+ * `apply` is set), advances the model's versioned `TuningProfile`, and stops on
+ * budget exhaustion or a quality plateau (N consecutive rejections). Cross-model
+ * bundled/stored profiles seed the optimizer as transfer priors.
+ *
+ * The `validate` dependency is injected, so the whole loop runs end-to-end in a
+ * test with a deterministic stub — no live model required.
+ */
+import { FlagConfig } from './flags.js';
+import type { JudgeClient } from './judge.js';
+import type { SearchPhase } from './search-reducer.js';
+import { TunerOptions } from './llm-tuner.js';
+import { TuningValidator, TuningDecisionOptions, TuningIterationResult } from './orchestrator.js';
+import { TuningProfile, TuningProfileStore } from './tuning-profile.js';
+export interface TuningLoopDeps {
+    /** Executor model family being tuned. */
+    model: string;
+    /** The injected paired-bench validator (candidate vs current). */
+    validate: TuningValidator;
+    /** Judge/tuner model client; null → the GP-BO path. */
+    judge?: JudgeClient | null;
+    /** Starting config; defaults to the model's seeded profile config. */
+    startConfig?: FlagConfig;
+    /** Max iterations (token/time budget). Default 6. */
+    maxIterations?: number;
+    /** Consecutive rejections before declaring a plateau and stopping. Default 3. */
+    plateauLimit?: number;
+    /** Search-phase schedule (cycled). Default coarse→medium→fine→combinatorial. */
+    phaseSchedule?: SearchPhase[];
+    decision?: TuningDecisionOptions;
+    tuner?: TunerOptions;
+    /** Extra transfer priors on top of the cross-model bundled/stored profiles. */
+    priors?: FlagConfig[];
+    validationSuites?: string[];
+    /** Stable timestamp (host may lack Date). */
+    now: string;
+    /** Physically persist accepted configs + the profile snapshot. */
+    apply: boolean;
+    /** Project dir for the flag-writer + profile store base. Required when apply. */
+    cwd?: string;
+    /** Profile store; when omitted, a fresh in-memory profile is used. */
+    profileStore?: TuningProfileStore;
+    log?: (msg: string) => void;
+}
+export interface TuningLoopResult {
+    model: string;
+    iterations: TuningIterationResult[];
+    accepted: TuningIterationResult[];
+    finalConfig: FlagConfig;
+    finalQuality: number;
+    profile: TuningProfile;
+    committed: boolean;
+    stoppedBecause: 'budget' | 'plateau';
+}
+export declare function runTuningLoop(deps: TuningLoopDeps): Promise<TuningLoopResult>;
+//# sourceMappingURL=run.d.ts.map

package/dist/self-tuning/run.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../src/self-tuning/run.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAGH,OAAO,EAAE,UAAU,EAAE,MAAM,YAAY,CAAC;AAExC,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,KAAK,EAAe,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,YAAY,EAAE,MAAM,gBAAgB,CAAC;AAC9C,OAAO,EACL,eAAe,EACf,qBAAqB,EACrB,qBAAqB,EAEtB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EACL,aAAa,EACb,kBAAkB,EAInB,MAAM,qBAAqB,CAAC;AAI7B,MAAM,WAAW,cAAc;IAC7B,yCAAyC;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,kEAAkE;IAClE,QAAQ,EAAE,eAAe,CAAC;IAC1B,uDAAuD;IACvD,KAAK,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;IAC3B,sEAAsE;IACtE,WAAW,CAAC,EAAE,UAAU,CAAC;IACzB,qDAAqD;IACrD,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,iFAAiF;IACjF,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,gFAAgF;IAChF,aAAa,CAAC,EAAE,WAAW,EAAE,CAAC;IAC9B,QAAQ,CAAC,EAAE,qBAAqB,CAAC;IACjC,KAAK,CAAC,EAAE,YAAY,CAAC;IACrB,+EAA+E;IAC/E,MAAM,CAAC,EAAE,UAAU,EAAE,CAAC;IACtB,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;IAC5B,6CAA6C;IAC7C,GAAG,EAAE,MAAM,CAAC;IACZ,kEAAkE;IAClE,KAAK,EAAE,OAAO,CAAC;IACf,iFAAiF;IACjF,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,sEAAsE;IACtE,YAAY,CAAC,EAAE,kBAAkB,CAAC;IAClC,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAC;CAC7B;AAED,MAAM,WAAW,gBAAgB;IAC/B,KAAK,EAAE,MAAM,CAAC;IACd,UAAU,EAAE,qBAAqB,EAAE,CAAC;IACpC,QAAQ,EAAE,qBAAqB,EAAE,CAAC;IAClC,WAAW,EAAE,UAAU,CAAC;IACxB,YAAY,EAAE,MAAM,CAAC;IACrB,OAAO,EAAE,aAAa,CAAC;IACvB,SAAS,EAAE,OAAO,CAAC;IACnB,cAAc,EAAE,QAAQ,GAAG,SAAS,CAAC;CACtC;AAED,wBAAsB,aAAa,CAAC,IAAI,EAAE,cAAc,GAAG,OAAO,CAAC,gBAAgB,CAAC,CAkHnF"}

package/dist/self-tuning/run.js

@@ -0,0 +1,122 @@
+/**
+ * LLM Self-Tuning — the autonomous run loop + profile persistence (P2/P3).
+ *
+ * `runTuningLoop` drives multiple `runTuningIteration` steps under a search-phase
+ * schedule (coarse → medium → fine → combinatorial, design §3.3.2), commits each
+ * ACCEPTED config to `.uap.json`/`.uap/proxy.env` via the flag-writer (only when
+ * `apply` is set), advances the model's versioned `TuningProfile`, and stops on
+ * budget exhaustion or a quality plateau (N consecutive rejections). Cross-model
+ * bundled/stored profiles seed the optimizer as transfer priors.
+ *
+ * The `validate` dependency is injected, so the whole loop runs end-to-end in a
+ * test with a deterministic stub — no live model required.
+ */
+import { normalizeModel } from '../self-harness/weakness.js';
+import { applyFlagChanges } from './flag-writer.js';
+import { runTuningIteration, } from './orchestrator.js';
+import { crossModelPriors, recordStep, seedProfile, } from './tuning-profile.js';
+const DEFAULT_PHASES = ['coarse', 'medium', 'fine', 'combinatorial'];
+export async function runTuningLoop(deps) {
+    const log = deps.log ?? (() => { });
+    const maxIter = deps.maxIterations ?? 6;
+    const plateauLimit = deps.plateauLimit ?? 3;
+    const phases = deps.phaseSchedule ?? DEFAULT_PHASES;
+    let profile = deps.profileStore
+        ? deps.profileStore.loadOrSeed(deps.model, deps.now)
+        : seedProfile(deps.model, deps.now);
+    let currentConfig = deps.startConfig ?? { ...profile.config };
+    let bestQuality = profile.quality;
+    // Observations seed the GP: the profile's accepted history + a synthetic
+    // baseline point for the starting config at its recorded quality.
+    const observations = [];
+    if (bestQuality > 0)
+        observations.push({ config: { ...currentConfig }, quality: bestQuality });
+    // Cross-model priors (bundled + other stored profiles) for transfer learning.
+    const priors = [
+        ...crossModelPriors(deps.model, deps.profileStore),
+        ...(deps.priors ?? []),
+    ];
+    const iterations = [];
+    const accepted = [];
+    let committed = false;
+    let plateau = 0;
+    let stoppedBecause = 'budget';
+    for (let i = 0; i < maxIter; i++) {
+        const phase = phases[i % phases.length];
+        log(`— iteration ${i + 1}/${maxIter} [${phase}] —`);
+        const result = await runTuningIteration({
+            ctx: {
+                model: normalizeModel(deps.model),
+                currentConfig,
+                observations: [...observations],
+                priors,
+                validationSuites: deps.validationSuites,
+            },
+            validate: deps.validate,
+            tuner: { ...deps.tuner, judge: deps.judge },
+            decision: deps.decision,
+            phase,
+            log,
+        });
+        iterations.push(result);
+        if (result.observation)
+            observations.push(result.observation);
+        if (result.accepted && result.decision) {
+            accepted.push(result);
+            plateau = 0;
+            const q = result.observation?.quality ?? bestQuality + result.decision.qualityDelta;
+            bestQuality = Math.max(bestQuality, q);
+            currentConfig = result.config;
+            // Commit: persist the accepted flag changes + advance the profile.
+            if (deps.apply) {
+                if (!deps.cwd)
+                    throw new Error('runTuningLoop: apply set but no cwd for the flag-writer');
+                const res = applyFlagChanges(deps.cwd, result.proposal.changes);
+                if (res.skipped.length)
+                    log(`  (skipped ${res.skipped.length} change(s) on commit)`);
+                committed = true;
+            }
+            profile = recordStep(profile, {
+                at: deps.now,
+                changes: result.proposal.changes,
+                quality: q,
+                delta: result.decision.qualityDelta,
+                accepted: true,
+                provenance: `tune ${phase} via ${result.proposal.source}`,
+            }, currentConfig);
+            if (deps.apply && deps.profileStore)
+                deps.profileStore.save(profile);
+        }
+        else {
+            plateau++;
+            if (result.decision) {
+                profile = recordStep(profile, {
+                    at: deps.now,
+                    changes: result.proposal.changes,
+                    quality: result.observation?.quality ?? NaN,
+                    delta: result.decision.qualityDelta,
+                    accepted: false,
+                    provenance: `tune ${phase} via ${result.proposal.source} (rejected)`,
+                }, currentConfig);
+            }
+            if (plateau >= plateauLimit) {
+                stoppedBecause = 'plateau';
+                log(`plateau: ${plateau} consecutive non-accepts — stopping`);
+                break;
+            }
+        }
+    }
+    if (deps.apply && deps.profileStore)
+        deps.profileStore.save(profile);
+    return {
+        model: normalizeModel(deps.model),
+        iterations,
+        accepted,
+        finalConfig: currentConfig,
+        finalQuality: bestQuality,
+        profile,
+        committed,
+        stoppedBecause,
+    };
+}
+//# sourceMappingURL=run.js.map

package/dist/self-tuning/run.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"run.js","sourceRoot":"","sources":["../../src/self-tuning/run.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,EAAE,cAAc,EAAE,MAAM,6BAA6B,CAAC;AAE7D,OAAO,EAAE,gBAAgB,EAAE,MAAM,kBAAkB,CAAC;AAIpD,OAAO,EAIL,kBAAkB,GACnB,MAAM,mBAAmB,CAAC;AAC3B,OAAO,EAGL,gBAAgB,EAChB,UAAU,EACV,WAAW,GACZ,MAAM,qBAAqB,CAAC;AAE7B,MAAM,cAAc,GAAkB,CAAC,QAAQ,EAAE,QAAQ,EAAE,MAAM,EAAE,eAAe,CAAC,CAAC;AA4CpF,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,IAAoB;IACtD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;IACnC,MAAM,OAAO,GAAG,IAAI,CAAC,aAAa,IAAI,CAAC,CAAC;IACxC,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,CAAC,CAAC;IAC5C,MAAM,MAAM,GAAG,IAAI,CAAC,aAAa,IAAI,cAAc,CAAC;IAEpD,IAAI,OAAO,GAAkB,IAAI,CAAC,YAAY;QAC5C,CAAC,CAAC,IAAI,CAAC,YAAY,CAAC,UAAU,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC;QACpD,CAAC,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;IAEtC,IAAI,aAAa,GAAe,IAAI,CAAC,WAAW,IAAI,EAAE,GAAG,OAAO,CAAC,MAAM,EAAE,CAAC;IAC1E,IAAI,WAAW,GAAG,OAAO,CAAC,OAAO,CAAC;IAElC,yEAAyE;IACzE,kEAAkE;IAClE,MAAM,YAAY,GAAkB,EAAE,CAAC;IACvC,IAAI,WAAW,GAAG,CAAC;QAAE,YAAY,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,EAAE,GAAG,aAAa,EAAE,EAAE,OAAO,EAAE,WAAW,EAAE,CAAC,CAAC;IAE/F,8EAA8E;IAC9E,MAAM,MAAM,GAAiB;QAC3B,GAAG,gBAAgB,CAAC,IAAI,CAAC,KAAK,EAAE,IAAI,CAAC,YAAY,CAAC;QAClD,GAAG,CAAC,IAAI,CAAC,MAAM,IAAI,EAAE,CAAC;KACvB,CAAC;IAEF,MAAM,UAAU,GAA4B,EAAE,CAAC;IAC/C,MAAM,QAAQ,GAA4B,EAAE,CAAC;IAC7C,IAAI,SAAS,GAAG,KAAK,CAAC;IACtB,IAAI,OAAO,GAAG,CAAC,CAAC;IAChB,IAAI,cAAc,GAAyB,QAAQ,CAAC;IAEpD,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,KAAK,GAAG,MAAM,CAAC,CAAC,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC;QACxC,GAAG,CAAC,eAAe,CAAC,GAAG,CAAC,IAAI,OAAO,KAAK,KAAK,KAAK,CAAC,CAAC;QAEpD,MAAM,MAAM,GAAG,MAAM,kBAAkB,CAAC;YACtC,GAAG,EAAE;gBACH,KAAK,EAAE,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC;gBACjC,aAAa;gBACb,YAAY,EAAE,CAAC,GAAG,YAAY,CAAC;gBAC/B,MAAM;gBACN,gBAAgB,EAAE,IAAI,CAAC,gBAAgB;aACxC;YACD,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,EAAE,GAAG,IAAI,CAAC,KAAK,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE;YAC3C,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK;YACL,GAAG;SACJ,CAAC,CAAC;QACH,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;QACxB,IAAI,MAAM,CAAC,WAAW;YAAE,YAAY,CAAC,IAAI,CAAC,MAAM,CAAC,WAAW,CAAC,CAAC;QAE9D,IAAI,MAAM,CAAC,QAAQ,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;YACvC,QAAQ,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;YACtB,OAAO,GAAG,CAAC,CAAC;YACZ,MAAM,CAAC,GAAG,MAAM,CAAC,WAAW,EAAE,OAAO,IAAI,WAAW,GAAG,MAAM,CAAC,QAAQ,CAAC,YAAY,CAAC;YACpF,WAAW,GAAG,IAAI,CAAC,GAAG,CAAC,WAAW,EAAE,CAAC,CAAC,CAAC;YACvC,aAAa,GAAG,MAAM,CAAC,MAAM,CAAC;YAE9B,mEAAmE;YACnE,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;gBACf,IAAI,CAAC,IAAI,CAAC,GAAG;oBAAE,MAAM,IAAI,KAAK,CAAC,yDAAyD,CAAC,CAAC;gBAC1F,MAAM,GAAG,GAAG,gBAAgB,CAAC,IAAI,CAAC,GAAG,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC;gBAChE,IAAI,GAAG,CAAC,OAAO,CAAC,MAAM;oBAAE,GAAG,CAAC,cAAc,GAAG,CAAC,OAAO,CAAC,MAAM,uBAAuB,CAAC,CAAC;gBACrF,SAAS,GAAG,IAAI,CAAC;YACnB,CAAC;YACD,OAAO,GAAG,UAAU,CAClB,OAAO,EACP;gBACE,EAAE,EAAE,IAAI,CAAC,GAAG;gBACZ,OAAO,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO;gBAChC,OAAO,EAAE,CAAC;gBACV,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,YAAY;gBACnC,QAAQ,EAAE,IAAI;gBACd,UAAU,EAAE,QAAQ,KAAK,QAAQ,MAAM,CAAC,QAAQ,CAAC,MAAM,EAAE;aAC1D,EACD,aAAa,CACd,CAAC;YACF,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,YAAY;gBAAE,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;QACvE,CAAC;aAAM,CAAC;YACN,OAAO,EAAE,CAAC;YACV,IAAI,MAAM,CAAC,QAAQ,EAAE,CAAC;gBACpB,OAAO,GAAG,UAAU,CAClB,OAAO,EACP;oBACE,EAAE,EAAE,IAAI,CAAC,GAAG;oBACZ,OAAO,EAAE,MAAM,CAAC,QAAQ,CAAC,OAAO;oBAChC,OAAO,EAAE,MAAM,CAAC,WAAW,EAAE,OAAO,IAAI,GAAG;oBAC3C,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,YAAY;oBACnC,QAAQ,EAAE,KAAK;oBACf,UAAU,EAAE,QAAQ,KAAK,QAAQ,MAAM,CAAC,QAAQ,CAAC,MAAM,aAAa;iBACrE,EACD,aAAa,CACd,CAAC;YACJ,CAAC;YACD,IAAI,OAAO,IAAI,YAAY,EAAE,CAAC;gBAC5B,cAAc,GAAG,SAAS,CAAC;gBAC3B,GAAG,CAAC,YAAY,OAAO,qCAAqC,CAAC,CAAC;gBAC9D,MAAM;YACR,CAAC;QACH,CAAC;IACH,CAAC;IAED,IAAI,IAAI,CAAC,KAAK,IAAI,IAAI,CAAC,YAAY;QAAE,IAAI,CAAC,YAAY,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAErE,OAAO;QACL,KAAK,EAAE,cAAc,CAAC,IAAI,CAAC,KAAK,CAAC;QACjC,UAAU;QACV,QAAQ;QACR,WAAW,EAAE,aAAa;QAC1B,YAAY,EAAE,WAAW;QACzB,OAAO;QACP,SAAS;QACT,cAAc;KACf,CAAC;AACJ,CAAC"}