@miller-tech/uap 1.136.2 → 1.137.2

package/dist/self-tuning/judge.js

@@ -0,0 +1,145 @@
+/**
+ * LLM Self-Tuning — the shared judge/tuner LLM client seam.
+ *
+ * Both the Quality Scorer (P0) and the LLM Tuner (P1) need to make a REAL call
+ * to a stronger "judge" model given a prompt and get text back. This module is
+ * the single, injectable seam for that:
+ *
+ *   - `JudgeClient` is a tiny interface (id + complete) so every consumer is
+ *     testable with a deterministic stub and never hard-wires the network.
+ *   - `resolveJudgeClient` builds a production client over `OpenAICompatClient`
+ *     + `ModelPresets`, resolving the judge model from an explicit id, the
+ *     `recipes.judge.model` config field, or an Opus fallback — the same
+ *     construction the `uap verify` acceptance judge uses (src/cli/verify.ts).
+ *   - `parseJsonLenient` recovers a JSON object from a completion even when the
+ *     model wraps it in prose / ``` fences (the local-model reality), while the
+ *     `x-uap-json-response` grammar keeps the happy path deterministic.
+ *
+ * The recipe/confidence-escalation engine that consumes `recipes.judge.model`
+ * at serving time lives in the Python proxy; this is the TypeScript-side judge
+ * call the offline tuning loop uses. See docs/design/LLM_SELF_TUNING_ANALYSIS.md.
+ */
+import { loadUapConfigRaw } from '../utils/config-loader.js';
+/**
+ * Resolve a judge model id into a concrete `ModelConfig`. Order of preference:
+ *   1. an exact `ModelPresets` key,
+ *   2. a preset whose `apiModel` equals the id (accepts wire ids like `claude-opus-4-8`),
+ *   3. a synthesized Anthropic-style config using the id verbatim as the apiModel.
+ */
+export function resolveJudgeModelConfig(id, presets) {
+    if (presets[id])
+        return presets[id];
+    const byApiModel = Object.values(presets).find((m) => m.apiModel === id);
+    if (byApiModel)
+        return byApiModel;
+    // Synthesize: treat the id as an OpenAI/Anthropic-compatible apiModel. The
+    // endpoint (and thus provider translation) is the proxy's concern.
+    return {
+        id,
+        name: id,
+        provider: id.startsWith('claude') ? 'anthropic' : 'custom',
+        apiModel: id,
+        apiKeyEnvVar: id.startsWith('claude') ? 'ANTHROPIC_API_KEY' : undefined,
+        maxContextTokens: 128000,
+        capabilities: [],
+    };
+}
+/**
+ * Build a production `JudgeClient`, or null when no judge model can be resolved
+ * and `requireExplicit`/no-fallback is in effect. Lazily imports the model layer
+ * so callers that never score/tune don't pay for it.
+ */
+export async function resolveJudgeClient(opts = {}) {
+    let id = opts.judgeModel;
+    if (!id && opts.cwd !== undefined) {
+        const raw = loadUapConfigRaw(opts.cwd);
+        const configured = raw?.recipes?.judge?.model;
+        if (typeof configured === 'string' && configured.trim())
+            id = configured.trim();
+    }
+    if (!id) {
+        if (opts.requireExplicit)
+            return null;
+        id = opts.fallbackPreset ?? 'opus-4.8';
+    }
+    const { OpenAICompatClient } = await import('../models/openai-compat-client.js');
+    const { ModelPresets } = await import('../models/types.js');
+    const base = resolveJudgeModelConfig(id, ModelPresets);
+    const model = opts.endpoint ? { ...base, endpoint: opts.endpoint } : base;
+    const client = new OpenAICompatClient();
+    return {
+        id: model.apiModel,
+        async complete(prompt, o = {}) {
+            const r = await client.complete(model, prompt, {
+                temperature: o.temperature ?? 0.1,
+                jsonResponse: o.json,
+                ...(o.maxTokens ? { maxTokens: o.maxTokens } : {}),
+            });
+            return r.content;
+        },
+    };
+}
+/**
+ * Extract the first well-formed JSON object/array from a model completion.
+ * Tolerates ``` fences, a `<think>` preamble, and leading/trailing prose — the
+ * failure modes a local model produces even under the JSON grammar. Returns the
+ * parsed value, or null if no balanced JSON literal is found.
+ */
+export function parseJsonLenient(text) {
+    if (!text)
+        return null;
+    // Fast path: the whole thing parses.
+    const trimmed = text.trim();
+    try {
+        return JSON.parse(trimmed);
+    }
+    catch {
+        /* fall through to scan */
+    }
+    // Strip a leading <think>…</think> block if present.
+    const afterThink = trimmed.replace(/^[\s\S]*?<\/think>/i, '').trim();
+    // Strip ``` fences.
+    const unfenced = afterThink.replace(/```(?:json)?/gi, '');
+    // Scan for the first balanced {...} or [...] literal.
+    for (const [open, close] of [['{', '}'], ['[', ']']]) {
+        const start = unfenced.indexOf(open);
+        if (start < 0)
+            continue;
+        let depth = 0;
+        let inStr = false;
+        let esc = false;
+        for (let i = start; i < unfenced.length; i++) {
+            const ch = unfenced[i];
+            if (esc) {
+                esc = false;
+                continue;
+            }
+            if (ch === '\\') {
+                esc = true;
+                continue;
+            }
+            if (ch === '"') {
+                inStr = !inStr;
+                continue;
+            }
+            if (inStr)
+                continue;
+            if (ch === open)
+                depth++;
+            else if (ch === close) {
+                depth--;
+                if (depth === 0) {
+                    const candidate = unfenced.slice(start, i + 1);
+                    try {
+                        return JSON.parse(candidate);
+                    }
+                    catch {
+                        break; // malformed; give up on this bracket type
+                    }
+                }
+            }
+        }
+    }
+    return null;
+}
+//# sourceMappingURL=judge.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"judge.js","sourceRoot":"","sources":["../../src/self-tuning/judge.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAGH,OAAO,EAAE,gBAAgB,EAAE,MAAM,2BAA2B,CAAC;AAuB7D;;;;;GAKG;AACH,MAAM,UAAU,uBAAuB,CACrC,EAAU,EACV,OAAoC;IAEpC,IAAI,OAAO,CAAC,EAAE,CAAC;QAAE,OAAO,OAAO,CAAC,EAAE,CAAC,CAAC;IACpC,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,EAAE,CAAC,CAAC;IACzE,IAAI,UAAU;QAAE,OAAO,UAAU,CAAC;IAClC,2EAA2E;IAC3E,mEAAmE;IACnE,OAAO;QACL,EAAE;QACF,IAAI,EAAE,EAAE;QACR,QAAQ,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,CAAC,CAAC,QAAQ;QAC1D,QAAQ,EAAE,EAAE;QACZ,YAAY,EAAE,EAAE,CAAC,UAAU,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC,mBAAmB,CAAC,CAAC,CAAC,SAAS;QACvE,gBAAgB,EAAE,MAAM;QACxB,YAAY,EAAE,EAAE;KACjB,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,kBAAkB,CAAC,OAA4B,EAAE;IACrE,IAAI,EAAE,GAAG,IAAI,CAAC,UAAU,CAAC;IACzB,IAAI,CAAC,EAAE,IAAI,IAAI,CAAC,GAAG,KAAK,SAAS,EAAE,CAAC;QAClC,MAAM,GAAG,GAAG,gBAAgB,CAAC,IAAI,CAAC,GAAG,CAAyD,CAAC;QAC/F,MAAM,UAAU,GAAG,GAAG,EAAE,OAAO,EAAE,KAAK,EAAE,KAAK,CAAC;QAC9C,IAAI,OAAO,UAAU,KAAK,QAAQ,IAAI,UAAU,CAAC,IAAI,EAAE;YAAE,EAAE,GAAG,UAAU,CAAC,IAAI,EAAE,CAAC;IAClF,CAAC;IACD,IAAI,CAAC,EAAE,EAAE,CAAC;QACR,IAAI,IAAI,CAAC,eAAe;YAAE,OAAO,IAAI,CAAC;QACtC,EAAE,GAAG,IAAI,CAAC,cAAc,IAAI,UAAU,CAAC;IACzC,CAAC;IAED,MAAM,EAAE,kBAAkB,EAAE,GAAG,MAAM,MAAM,CAAC,mCAAmC,CAAC,CAAC;IACjF,MAAM,EAAE,YAAY,EAAE,GAAG,MAAM,MAAM,CAAC,oBAAoB,CAAC,CAAC;IAC5D,MAAM,IAAI,GAAG,uBAAuB,CAAC,EAAE,EAAE,YAAY,CAAC,CAAC;IACvD,MAAM,KAAK,GAAgB,IAAI,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,GAAG,IAAI,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC;IACvF,MAAM,MAAM,GAAG,IAAI,kBAAkB,EAAE,CAAC;IAExC,OAAO;QACL,EAAE,EAAE,KAAK,CAAC,QAAQ;QAClB,KAAK,CAAC,QAAQ,CAAC,MAAM,EAAE,CAAC,GAAG,EAAE;YAC3B,MAAM,CAAC,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,KAAK,EAAE,MAAM,EAAE;gBAC7C,WAAW,EAAE,CAAC,CAAC,WAAW,IAAI,GAAG;gBACjC,YAAY,EAAE,CAAC,CAAC,IAAI;gBACpB,GAAG,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,CAAC,CAAC,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC;aACnD,CAAC,CAAC;YACH,OAAO,CAAC,CAAC,OAAO,CAAC;QACnB,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAAc,IAAY;IACxD,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IACvB,qCAAqC;IACrC,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,IAAI,CAAC;QACH,OAAO,IAAI,CAAC,KAAK,CAAC,OAAO,CAAM,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,0BAA0B;IAC5B,CAAC;IACD,qDAAqD;IACrD,MAAM,UAAU,GAAG,OAAO,CAAC,OAAO,CAAC,qBAAqB,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;IACrE,oBAAoB;IACpB,MAAM,QAAQ,GAAG,UAAU,CAAC,OAAO,CAAC,gBAAgB,EAAE,EAAE,CAAC,CAAC;IAC1D,sDAAsD;IACtD,KAAK,MAAM,CAAC,IAAI,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAU,EAAE,CAAC;QAC9D,MAAM,KAAK,GAAG,QAAQ,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;QACrC,IAAI,KAAK,GAAG,CAAC;YAAE,SAAS;QACxB,IAAI,KAAK,GAAG,CAAC,CAAC;QACd,IAAI,KAAK,GAAG,KAAK,CAAC;QAClB,IAAI,GAAG,GAAG,KAAK,CAAC;QAChB,KAAK,IAAI,CAAC,GAAG,KAAK,EAAE,CAAC,GAAG,QAAQ,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;YAC7C,MAAM,EAAE,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC;YACvB,IAAI,GAAG,EAAE,CAAC;gBAAC,GAAG,GAAG,KAAK,CAAC;gBAAC,SAAS;YAAC,CAAC;YACnC,IAAI,EAAE,KAAK,IAAI,EAAE,CAAC;gBAAC,GAAG,GAAG,IAAI,CAAC;gBAAC,SAAS;YAAC,CAAC;YAC1C,IAAI,EAAE,KAAK,GAAG,EAAE,CAAC;gBAAC,KAAK,GAAG,CAAC,KAAK,CAAC;gBAAC,SAAS;YAAC,CAAC;YAC7C,IAAI,KAAK;gBAAE,SAAS;YACpB,IAAI,EAAE,KAAK,IAAI;gBAAE,KAAK,EAAE,CAAC;iBACpB,IAAI,EAAE,KAAK,KAAK,EAAE,CAAC;gBACtB,KAAK,EAAE,CAAC;gBACR,IAAI,KAAK,KAAK,CAAC,EAAE,CAAC;oBAChB,MAAM,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,KAAK,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;oBAC/C,IAAI,CAAC;wBACH,OAAO,IAAI,CAAC,KAAK,CAAC,SAAS,CAAM,CAAC;oBACpC,CAAC;oBAAC,MAAM,CAAC;wBACP,MAAM,CAAC,0CAA0C;oBACnD,CAAC;gBACH,CAAC;YACH,CAAC;QACH,CAAC;IACH,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC"}

package/dist/self-tuning/llm-tuner.d.ts

@@ -0,0 +1,83 @@
+/**
+ * LLM Self-Tuning — the LLM Tuner (P1): LLM-guided flag-space exploration.
+ *
+ * The tuner is the "LLM as tuner, not executor" principle (design §3.1): a
+ * stronger model reasons over the quality history + flag interactions and
+ * proposes the next configuration; the small model then executes under it and
+ * the paired bench validates. Two production paths:
+ *
+ *   - LLM path (a judge model configured): the tuner sends the active flag
+ *     domains, the current config, and the past (config → quality) attempts, and
+ *     asks for a small, interpretable change set with a rationale. Every
+ *     proposed change is validated against the flag domain + dependency graph
+ *     before it can reach the flag-writer — a hallucinated key or out-of-range
+ *     value is dropped, never applied.
+ *   - GP path (no judge, or the LLM proposes nothing usable): the Gaussian-
+ *     process Bayesian optimizer (`search-reducer`) picks the next config by
+ *     Expected Improvement. This is the deterministic, network-free fallback.
+ *
+ * Both return the same `TuningProposal`, so the orchestrator is agnostic to how
+ * the next config was chosen.
+ */
+import type { JudgeClient } from './judge.js';
+import { FlagChange, FlagConfig } from './flags.js';
+import { Observation, SearchPhase } from './search-reducer.js';
+export interface TuningProposal {
+    /** The flags to change and their new values (validated, ≤ maxChanges). */
+    changes: FlagChange[];
+    /** Why (LLM rationale, or a GP summary). */
+    rationale: string;
+    /** Expected composite-quality improvement (LLM estimate or GP mean gain). */
+    expectedDelta: number;
+    /** Confidence in the proposal, 0-1. */
+    confidence: number;
+    /** Model family this proposal is optimized for. */
+    targetModel: string;
+    /** Benchmark suites the orchestrator should validate against. */
+    validationSuites: string[];
+    /** How the proposal was produced. */
+    source: 'llm' | 'gp' | 'gp-fallback';
+}
+export interface TuningContext {
+    /** Executor model family being tuned. */
+    model: string;
+    /** The configuration currently in effect. */
+    currentConfig: FlagConfig;
+    /** Past (config, quality) trials — the GP conditioning set + LLM history. */
+    observations: Observation[];
+    /** Cross-model transfer priors (good configs from other models). */
+    priors?: FlagConfig[];
+    /** Suites to stamp on the proposal for validation. */
+    validationSuites?: string[];
+}
+export interface TunerOptions {
+    /** A resolved judge/tuner client. When null/omitted, the GP path is used. */
+    judge?: JudgeClient | null;
+    /** Max flags to change per proposal (interpretability bound). Default 4. */
+    maxChanges?: number;
+    /** Search phase (biases the GP candidate generation + the LLM guidance). */
+    phase?: SearchPhase;
+    /** RNG seed for the GP path. Default 1. */
+    seed?: number;
+}
+/** Compact human/LLM-readable description of the ACTIVE flag domains. */
+export declare function describeActiveFlags(cfg: FlagConfig): string;
+/** Build the tuner prompt (design §Phase 1). */
+export declare function buildTunerPrompt(ctx: TuningContext, opts: TunerOptions): string;
+/**
+ * Validate raw LLM-proposed changes into a clean `FlagChange[]`: keep only
+ * tunable keys, coerce each value into its domain, drop no-ops and changes to a
+ * flag whose parent won't be active after the change set is applied, and cap to
+ * `maxChanges`.
+ */
+export declare function sanitizeChanges(currentConfig: FlagConfig, raw: Array<{
+    key?: unknown;
+    to?: unknown;
+}>, maxChanges: number): FlagChange[];
+/**
+ * Propose the next configuration to try. Uses the judge (LLM) when available and
+ * it returns usable changes; otherwise the Gaussian-process optimizer. Never
+ * throws — an LLM/parse failure degrades to the GP path.
+ */
+export declare function proposeTuning(ctx: TuningContext, opts?: TunerOptions): Promise<TuningProposal>;
+//# sourceMappingURL=llm-tuner.d.ts.map

package/dist/self-tuning/llm-tuner.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm-tuner.d.ts","sourceRoot":"","sources":["../../src/self-tuning/llm-tuner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,YAAY,CAAC;AAE9C,OAAO,EACL,UAAU,EACV,UAAU,EAQX,MAAM,YAAY,CAAC;AACpB,OAAO,EACL,WAAW,EACX,WAAW,EAEZ,MAAM,qBAAqB,CAAC;AAE7B,MAAM,WAAW,cAAc;IAC7B,0EAA0E;IAC1E,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,4CAA4C;IAC5C,SAAS,EAAE,MAAM,CAAC;IAClB,6EAA6E;IAC7E,aAAa,EAAE,MAAM,CAAC;IACtB,uCAAuC;IACvC,UAAU,EAAE,MAAM,CAAC;IACnB,mDAAmD;IACnD,WAAW,EAAE,MAAM,CAAC;IACpB,iEAAiE;IACjE,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC3B,qCAAqC;IACrC,MAAM,EAAE,KAAK,GAAG,IAAI,GAAG,aAAa,CAAC;CACtC;AAED,MAAM,WAAW,aAAa;IAC5B,yCAAyC;IACzC,KAAK,EAAE,MAAM,CAAC;IACd,6CAA6C;IAC7C,aAAa,EAAE,UAAU,CAAC;IAC1B,6EAA6E;IAC7E,YAAY,EAAE,WAAW,EAAE,CAAC;IAC5B,oEAAoE;IACpE,MAAM,CAAC,EAAE,UAAU,EAAE,CAAC;IACtB,sDAAsD;IACtD,gBAAgB,CAAC,EAAE,MAAM,EAAE,CAAC;CAC7B;AAED,MAAM,WAAW,YAAY;IAC3B,6EAA6E;IAC7E,KAAK,CAAC,EAAE,WAAW,GAAG,IAAI,CAAC;IAC3B,4EAA4E;IAC5E,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,4EAA4E;IAC5E,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,2CAA2C;IAC3C,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAQD,yEAAyE;AACzE,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,UAAU,GAAG,MAAM,CAc3D;AAeD,gDAAgD;AAChD,wBAAgB,gBAAgB,CAAC,GAAG,EAAE,aAAa,EAAE,IAAI,EAAE,YAAY,GAAG,MAAM,CAwB/E;AASD;;;;;GAKG;AACH,wBAAgB,eAAe,CAC7B,aAAa,EAAE,UAAU,EACzB,GAAG,EAAE,KAAK,CAAC;IAAE,GAAG,CAAC,EAAE,OAAO,CAAC;IAAC,EAAE,CAAC,EAAE,OAAO,CAAA;CAAE,CAAC,EAC3C,UAAU,EAAE,MAAM,GACjB,UAAU,EAAE,CAoBd;AAqCD;;;;GAIG;AACH,wBAAsB,aAAa,CAAC,GAAG,EAAE,aAAa,EAAE,IAAI,GAAE,YAAiB,GAAG,OAAO,CAAC,cAAc,CAAC,CA+BxG"}

package/dist/self-tuning/llm-tuner.js

@@ -0,0 +1,191 @@
+/**
+ * LLM Self-Tuning — the LLM Tuner (P1): LLM-guided flag-space exploration.
+ *
+ * The tuner is the "LLM as tuner, not executor" principle (design §3.1): a
+ * stronger model reasons over the quality history + flag interactions and
+ * proposes the next configuration; the small model then executes under it and
+ * the paired bench validates. Two production paths:
+ *
+ *   - LLM path (a judge model configured): the tuner sends the active flag
+ *     domains, the current config, and the past (config → quality) attempts, and
+ *     asks for a small, interpretable change set with a rationale. Every
+ *     proposed change is validated against the flag domain + dependency graph
+ *     before it can reach the flag-writer — a hallucinated key or out-of-range
+ *     value is dropped, never applied.
+ *   - GP path (no judge, or the LLM proposes nothing usable): the Gaussian-
+ *     process Bayesian optimizer (`search-reducer`) picks the next config by
+ *     Expected Improvement. This is the deterministic, network-free fallback.
+ *
+ * Both return the same `TuningProposal`, so the orchestrator is agnostic to how
+ * the next config was chosen.
+ */
+import { parseJsonLenient } from './judge.js';
+import { TUNABLE_FLAGS, applyChanges, coerceToDomain, diffConfigs, flagValue, getTunableFlag, isFlagActive, } from './flags.js';
+import { proposeNext, } from './search-reducer.js';
+/** The current best-known quality (from observations), for delta estimates. */
+function currentQuality(ctx) {
+    if (ctx.observations.length === 0)
+        return 0;
+    return ctx.observations.reduce((a, o) => Math.max(a, o.quality), 0);
+}
+/** Compact human/LLM-readable description of the ACTIVE flag domains. */
+export function describeActiveFlags(cfg) {
+    const lines = [];
+    for (const f of TUNABLE_FLAGS) {
+        if (!isFlagActive(f, cfg))
+            continue;
+        const d = f.domain;
+        const dom = d.kind === 'bool'
+            ? 'true|false'
+            : d.kind === 'enum'
+                ? d.values.join('|')
+                : `${d.min}..${d.max}${d.int ? ' (int)' : ''}`;
+        lines.push(`- ${f.key} = ${JSON.stringify(flagValue(cfg, f.key))}  [${dom}]`);
+    }
+    return lines.join('\n');
+}
+function historyForPrompt(observations, limit = 8) {
+    if (!observations.length)
+        return '(none yet)';
+    return observations
+        .slice(-limit)
+        .map((o, i) => {
+        const changed = TUNABLE_FLAGS.filter((f) => isFlagActive(f, o.config))
+            .map((f) => `${f.key}=${flagValue(o.config, f.key)}`)
+            .join(', ');
+        return `#${i + 1} quality=${o.quality.toFixed(1)} :: ${changed.slice(0, 240)}`;
+    })
+        .join('\n');
+}
+/** Build the tuner prompt (design §Phase 1). */
+export function buildTunerPrompt(ctx, opts) {
+    const maxChanges = opts.maxChanges ?? 4;
+    return [
+        'You are a configuration optimizer for the Universal Agent Protocol (UAP).',
+        `Your goal: raise the composite quality (0-100) of the executor model by tuning UAP flags.`,
+        '',
+        `EXECUTOR MODEL: ${ctx.model}`,
+        `PHASE: ${opts.phase ?? 'combinatorial'} (coarse=toggle enablers, medium=enums, fine=numbers, combinatorial=interactions)`,
+        '',
+        'ACTIVE FLAGS (key = current value  [domain]):',
+        describeActiveFlags(ctx.currentConfig),
+        '',
+        'PAST ATTEMPTS (most recent last):',
+        historyForPrompt(ctx.observations),
+        '',
+        'RULES:',
+        `- Change only ${maxChanges} or fewer flags (keep it interpretable).`,
+        '- Only use keys from ACTIVE FLAGS above; respect each domain.',
+        '- Prefer changes that plausibly lift quality given the history.',
+        '- A dependent flag (e.g. recipes.fusionN) only helps if its parent is enabled.',
+        '',
+        'Respond with ONLY this JSON, no prose:',
+        '{"changes":[{"key":"<flag>","to":<value>}],"rationale":"<why>","expectedDelta":<number>,"confidence":<0-1>}',
+    ].join('\n');
+}
+/**
+ * Validate raw LLM-proposed changes into a clean `FlagChange[]`: keep only
+ * tunable keys, coerce each value into its domain, drop no-ops and changes to a
+ * flag whose parent won't be active after the change set is applied, and cap to
+ * `maxChanges`.
+ */
+export function sanitizeChanges(currentConfig, raw, maxChanges) {
+    const cleaned = [];
+    for (const c of raw) {
+        if (typeof c?.key !== 'string')
+            continue;
+        const flag = getTunableFlag(c.key);
+        if (!flag)
+            continue;
+        if (c.to === undefined || c.to === null)
+            continue;
+        const to = coerceToDomain(c.key, c.to);
+        if (to === null)
+            continue;
+        const from = flagValue(currentConfig, c.key) ?? null;
+        if (to === from)
+            continue; // no-op
+        cleaned.push({ key: c.key, from, to, category: flag.category });
+        if (cleaned.length >= maxChanges)
+            break;
+    }
+    // Drop changes to a flag whose dependencies aren't satisfied by the RESULT.
+    const result = applyChanges(currentConfig, cleaned);
+    return cleaned.filter((c) => {
+        const flag = getTunableFlag(c.key);
+        return isFlagActive(flag, result);
+    });
+}
+/** GP-path proposal: next config by Expected Improvement. */
+function gpProposal(ctx, opts, source) {
+    const suggestion = proposeNext(ctx.observations, {
+        seed: opts.seed,
+        phase: opts.phase,
+        priors: ctx.priors,
+        seedConfig: ctx.currentConfig,
+    });
+    const base = currentQuality(ctx);
+    if (!suggestion) {
+        return {
+            changes: [],
+            rationale: 'no candidates to explore (empty search space or no observations)',
+            expectedDelta: 0,
+            confidence: 0,
+            targetModel: ctx.model,
+            validationSuites: ctx.validationSuites ?? [],
+            source,
+        };
+    }
+    const changes = diffConfigs(ctx.currentConfig, suggestion.config).slice(0, opts.maxChanges ?? 4);
+    const expectedDelta = Number.isFinite(suggestion.mean) ? suggestion.mean - base : 0;
+    // Confidence from predictive certainty: tighter std → higher confidence.
+    const confidence = suggestion.std > 0 ? Math.max(0, Math.min(1, 1 - suggestion.std / 50)) : 0.5;
+    return {
+        changes,
+        rationale: `GP-BO: EI=${Number.isFinite(suggestion.ei) ? suggestion.ei.toFixed(3) : '∞'}, predicted quality ${suggestion.mean.toFixed(1)}±${suggestion.std.toFixed(1)}`,
+        expectedDelta,
+        confidence,
+        targetModel: ctx.model,
+        validationSuites: ctx.validationSuites ?? [],
+        source,
+    };
+}
+/**
+ * Propose the next configuration to try. Uses the judge (LLM) when available and
+ * it returns usable changes; otherwise the Gaussian-process optimizer. Never
+ * throws — an LLM/parse failure degrades to the GP path.
+ */
+export async function proposeTuning(ctx, opts = {}) {
+    const maxChanges = opts.maxChanges ?? 4;
+    if (!opts.judge)
+        return gpProposal(ctx, opts, 'gp');
+    let raw = null;
+    try {
+        const text = await opts.judge.complete(buildTunerPrompt(ctx, opts), { json: true, temperature: 0.2 });
+        raw = parseJsonLenient(text);
+    }
+    catch {
+        raw = null;
+    }
+    if (!raw || !Array.isArray(raw.changes))
+        return gpProposal(ctx, opts, 'gp-fallback');
+    const changes = sanitizeChanges(ctx.currentConfig, raw.changes, maxChanges);
+    if (changes.length === 0)
+        return gpProposal(ctx, opts, 'gp-fallback');
+    const expectedDelta = typeof raw.expectedDelta === 'number' && Number.isFinite(raw.expectedDelta)
+        ? raw.expectedDelta
+        : 0;
+    const confidence = typeof raw.confidence === 'number' && Number.isFinite(raw.confidence)
+        ? Math.max(0, Math.min(1, raw.confidence))
+        : 0.5;
+    return {
+        changes,
+        rationale: typeof raw.rationale === 'string' ? raw.rationale.slice(0, 400) : 'LLM tuner proposal',
+        expectedDelta,
+        confidence,
+        targetModel: ctx.model,
+        validationSuites: ctx.validationSuites ?? [],
+        source: 'llm',
+    };
+}
+//# sourceMappingURL=llm-tuner.js.map

package/dist/self-tuning/llm-tuner.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"llm-tuner.js","sourceRoot":"","sources":["../../src/self-tuning/llm-tuner.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAGH,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,EAGL,aAAa,EACb,YAAY,EACZ,cAAc,EACd,WAAW,EACX,SAAS,EACT,cAAc,EACd,YAAY,GACb,MAAM,YAAY,CAAC;AACpB,OAAO,EAGL,WAAW,GACZ,MAAM,qBAAqB,CAAC;AA2C7B,+EAA+E;AAC/E,SAAS,cAAc,CAAC,GAAkB;IACxC,IAAI,GAAG,CAAC,YAAY,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,CAAC,CAAC;IAC5C,OAAO,GAAG,CAAC,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC,CAAC;AACtE,CAAC;AAED,yEAAyE;AACzE,MAAM,UAAU,mBAAmB,CAAC,GAAe;IACjD,MAAM,KAAK,GAAa,EAAE,CAAC;IAC3B,KAAK,MAAM,CAAC,IAAI,aAAa,EAAE,CAAC;QAC9B,IAAI,CAAC,YAAY,CAAC,CAAC,EAAE,GAAG,CAAC;YAAE,SAAS;QACpC,MAAM,CAAC,GAAG,CAAC,CAAC,MAAM,CAAC;QACnB,MAAM,GAAG,GACP,CAAC,CAAC,IAAI,KAAK,MAAM;YACf,CAAC,CAAC,YAAY;YACd,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM;gBACjB,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC;gBACpB,CAAC,CAAC,GAAG,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,GAAG,GAAG,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC;QACrD,KAAK,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,MAAM,GAAG,GAAG,CAAC,CAAC;IAChF,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC1B,CAAC;AAED,SAAS,gBAAgB,CAAC,YAA2B,EAAE,KAAK,GAAG,CAAC;IAC9D,IAAI,CAAC,YAAY,CAAC,MAAM;QAAE,OAAO,YAAY,CAAC;IAC9C,OAAO,YAAY;SAChB,KAAK,CAAC,CAAC,KAAK,CAAC;SACb,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE;QACZ,MAAM,OAAO,GAAG,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC;aACnE,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,GAAG,IAAI,SAAS,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC;aACpD,IAAI,CAAC,IAAI,CAAC,CAAC;QACd,OAAO,IAAI,CAAC,GAAG,CAAC,YAAY,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;IACjF,CAAC,CAAC;SACD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,gDAAgD;AAChD,MAAM,UAAU,gBAAgB,CAAC,GAAkB,EAAE,IAAkB;IACrE,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC;IACxC,OAAO;QACL,2EAA2E;QAC3E,2FAA2F;QAC3F,EAAE;QACF,mBAAmB,GAAG,CAAC,KAAK,EAAE;QAC9B,UAAU,IAAI,CAAC,KAAK,IAAI,eAAe,mFAAmF;QAC1H,EAAE;QACF,+CAA+C;QAC/C,mBAAmB,CAAC,GAAG,CAAC,aAAa,CAAC;QACtC,EAAE;QACF,mCAAmC;QACnC,gBAAgB,CAAC,GAAG,CAAC,YAAY,CAAC;QAClC,EAAE;QACF,QAAQ;QACR,iBAAiB,UAAU,0CAA0C;QACrE,+DAA+D;QAC/D,iEAAiE;QACjE,gFAAgF;QAChF,EAAE;QACF,wCAAwC;QACxC,6GAA6G;KAC9G,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AACf,CAAC;AASD;;;;;GAKG;AACH,MAAM,UAAU,eAAe,CAC7B,aAAyB,EACzB,GAA2C,EAC3C,UAAkB;IAElB,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,CAAC,IAAI,GAAG,EAAE,CAAC;QACpB,IAAI,OAAO,CAAC,EAAE,GAAG,KAAK,QAAQ;YAAE,SAAS;QACzC,MAAM,IAAI,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;QACnC,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,IAAI,CAAC,CAAC,EAAE,KAAK,SAAS,IAAI,CAAC,CAAC,EAAE,KAAK,IAAI;YAAE,SAAS;QAClD,MAAM,EAAE,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,EAAW,CAAC,CAAC;QAChD,IAAI,EAAE,KAAK,IAAI;YAAE,SAAS;QAC1B,MAAM,IAAI,GAAG,SAAS,CAAC,aAAa,EAAE,CAAC,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC;QACrD,IAAI,EAAE,KAAK,IAAI;YAAE,SAAS,CAAC,QAAQ;QACnC,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,CAAC,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QAChE,IAAI,OAAO,CAAC,MAAM,IAAI,UAAU;YAAE,MAAM;IAC1C,CAAC;IACD,4EAA4E;IAC5E,MAAM,MAAM,GAAG,YAAY,CAAC,aAAa,EAAE,OAAO,CAAC,CAAC;IACpD,OAAO,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE;QAC1B,MAAM,IAAI,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,CAAE,CAAC;QACpC,OAAO,YAAY,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACpC,CAAC,CAAC,CAAC;AACL,CAAC;AAED,6DAA6D;AAC7D,SAAS,UAAU,CAAC,GAAkB,EAAE,IAAkB,EAAE,MAA4B;IACtF,MAAM,UAAU,GAAG,WAAW,CAAC,GAAG,CAAC,YAAY,EAAE;QAC/C,IAAI,EAAE,IAAI,CAAC,IAAI;QACf,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,MAAM,EAAE,GAAG,CAAC,MAAM;QAClB,UAAU,EAAE,GAAG,CAAC,aAAa;KAC9B,CAAC,CAAC;IACH,MAAM,IAAI,GAAG,cAAc,CAAC,GAAG,CAAC,CAAC;IACjC,IAAI,CAAC,UAAU,EAAE,CAAC;QAChB,OAAO;YACL,OAAO,EAAE,EAAE;YACX,SAAS,EAAE,kEAAkE;YAC7E,aAAa,EAAE,CAAC;YAChB,UAAU,EAAE,CAAC;YACb,WAAW,EAAE,GAAG,CAAC,KAAK;YACtB,gBAAgB,EAAE,GAAG,CAAC,gBAAgB,IAAI,EAAE;YAC5C,MAAM;SACP,CAAC;IACJ,CAAC;IACD,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,aAAa,EAAE,UAAU,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC,CAAC;IACjG,MAAM,aAAa,GAAG,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,IAAI,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IACpF,yEAAyE;IACzE,MAAM,UAAU,GAAG,UAAU,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,UAAU,CAAC,GAAG,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IAChG,OAAO;QACL,OAAO;QACP,SAAS,EAAE,aAAa,MAAM,CAAC,QAAQ,CAAC,UAAU,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,UAAU,CAAC,EAAE,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,uBAAuB,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,IAAI,UAAU,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE;QACvK,aAAa;QACb,UAAU;QACV,WAAW,EAAE,GAAG,CAAC,KAAK;QACtB,gBAAgB,EAAE,GAAG,CAAC,gBAAgB,IAAI,EAAE;QAC5C,MAAM;KACP,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,aAAa,CAAC,GAAkB,EAAE,OAAqB,EAAE;IAC7E,MAAM,UAAU,GAAG,IAAI,CAAC,UAAU,IAAI,CAAC,CAAC;IACxC,IAAI,CAAC,IAAI,CAAC,KAAK;QAAE,OAAO,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;IAEpD,IAAI,GAAG,GAAuB,IAAI,CAAC;IACnC,IAAI,CAAC;QACH,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,gBAAgB,CAAC,GAAG,EAAE,IAAI,CAAC,EAAE,EAAE,IAAI,EAAE,IAAI,EAAE,WAAW,EAAE,GAAG,EAAE,CAAC,CAAC;QACtG,GAAG,GAAG,gBAAgB,CAAc,IAAI,CAAC,CAAC;IAC5C,CAAC;IAAC,MAAM,CAAC;QACP,GAAG,GAAG,IAAI,CAAC;IACb,CAAC;IACD,IAAI,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,CAAC,OAAO,CAAC;QAAE,OAAO,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,aAAa,CAAC,CAAC;IAErF,MAAM,OAAO,GAAG,eAAe,CAAC,GAAG,CAAC,aAAa,EAAE,GAAG,CAAC,OAAO,EAAE,UAAU,CAAC,CAAC;IAC5E,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,UAAU,CAAC,GAAG,EAAE,IAAI,EAAE,aAAa,CAAC,CAAC;IAEtE,MAAM,aAAa,GAAG,OAAO,GAAG,CAAC,aAAa,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,aAAa,CAAC;QAC/F,CAAC,CAAC,GAAG,CAAC,aAAa;QACnB,CAAC,CAAC,CAAC,CAAC;IACN,MAAM,UAAU,GAAG,OAAO,GAAG,CAAC,UAAU,KAAK,QAAQ,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,UAAU,CAAC;QACtF,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,GAAG,CAAC,UAAU,CAAC,CAAC;QAC1C,CAAC,CAAC,GAAG,CAAC;IACR,OAAO;QACL,OAAO;QACP,SAAS,EAAE,OAAO,GAAG,CAAC,SAAS,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,oBAAoB;QACjG,aAAa;QACb,UAAU;QACV,WAAW,EAAE,GAAG,CAAC,KAAK;QACtB,gBAAgB,EAAE,GAAG,CAAC,gBAAgB,IAAI,EAAE;QAC5C,MAAM,EAAE,KAAK;KACd,CAAC;AACJ,CAAC"}

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

@@ -0,0 +1,90 @@
+/**
+ * LLM Self-Tuning — the closed-loop orchestrator (P2).
+ *
+ * One tuning iteration: propose → apply → validate → decide → learn.
+ *
+ *   1. PROPOSE  the next config (LLM tuner, or GP-BO fallback).
+ *   2. APPLY    the change set onto the current config (in-memory here; the
+ *               committing loop in run.ts persists it via the flag-writer).
+ *   3. VALIDATE it with the INJECTED paired-bench validator — it runs the
+ *               candidate config against the current config over the validation
+ *               (and held-out) suites and returns the quality-scored comparison.
+ *   4. DECIDE   accept iff the composite-QUALITY delta clears a hysteresis band,
+ *               is statistically supported, and the held-out suite doesn't
+ *               regress — the "conservative validation preventing regressions"
+ *               rule, keyed to quality rather than raw correctness.
+ *   5. LEARN    accepted configs advance the model's TuningProfile; the loop's
+ *               observations feed the GP so the next proposal is better-informed.
+ *
+ * The validator is injected so the loop is unit-testable without a live model,
+ * exactly like the self-harness orchestrator.
+ */
+import type { Comparison } from '../benchmarks/paired/report.js';
+import { FlagChange, FlagConfig } from './flags.js';
+import { TuningContext, TunerOptions, TuningProposal } from './llm-tuner.js';
+import type { Observation, SearchPhase } from './search-reducer.js';
+/** The paired-bench result for a candidate config vs the current config. */
+export interface TuningValidationOutcome {
+    validation: Comparison;
+    /** Held-out comparison; null when the held-out suite was skipped. */
+    heldout: Comparison | null;
+    /** Measured composite quality of the candidate arm (for the observation set). */
+    candidateQuality: number;
+}
+/**
+ * Runs a candidate config against the current config and returns the scored
+ * comparison. The real implementation (run.ts) applies the candidate via the
+ * flag-writer, runs the paired bench with quality scoring, and reverts; tests
+ * inject a deterministic stub.
+ */
+export type TuningValidator = (candidate: FlagConfig, current: FlagConfig) => Promise<TuningValidationOutcome>;
+export interface TuningDecisionOptions {
+    /**
+     * Minimum composite-quality improvement (points, 0-100) required to accept — a
+     * hysteresis band that stops the config oscillating on noise (design §6). The
+     * delta must ALSO be statistically significant. Default 2.
+     */
+    minQualityGain?: number;
+    /** Tolerated held-out quality regression before rejecting. Default 2. */
+    heldoutRegressionTolerance?: number;
+}
+export type TuningVerdict = 'accept' | 'reject';
+export interface TuningDecision {
+    verdict: TuningVerdict;
+    reason: string;
+    qualityDelta: number;
+    heldoutQualityDelta: number | null;
+}
+/**
+ * Accept a candidate iff its validation quality delta clears the hysteresis band
+ * AND is significant AND the held-out suite shows no significant quality
+ * regression. Falls back to the correctness delta when quality scores are
+ * absent (a bench run without scoring).
+ */
+export declare function decideTuning(validation: Comparison, heldout: Comparison | null, opts?: TuningDecisionOptions): TuningDecision;
+export interface TuningIterationOptions {
+    ctx: TuningContext;
+    validate: TuningValidator;
+    tuner?: TunerOptions;
+    decision?: TuningDecisionOptions;
+    phase?: SearchPhase;
+    log?: (msg: string) => void;
+}
+export interface TuningIterationResult {
+    proposal: TuningProposal;
+    /** The config that would be adopted if accepted. */
+    candidateConfig: FlagConfig;
+    decision: TuningDecision | null;
+    accepted: boolean;
+    /** New best config after this iteration (candidate if accepted, else current). */
+    config: FlagConfig;
+    /** Observation appended for the GP (the candidate's measured quality). */
+    observation: Observation | null;
+    outcome: 'accepted' | 'rejected' | 'plateau' | 'error';
+    error?: string;
+}
+/** Run ONE tuning iteration (propose → validate → decide). Never throws. */
+export declare function runTuningIteration(opts: TuningIterationOptions): Promise<TuningIterationResult>;
+/** Convert an accepted proposal's changes into a compact one-line audit string. */
+export declare function describeChanges(changes: FlagChange[]): string;
+//# sourceMappingURL=orchestrator.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"orchestrator.d.ts","sourceRoot":"","sources":["../../src/self-tuning/orchestrator.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gCAAgC,CAAC;AACjE,OAAO,EAAE,UAAU,EAAE,UAAU,EAAgB,MAAM,YAAY,CAAC;AAClE,OAAO,EACL,aAAa,EACb,YAAY,EACZ,cAAc,EAEf,MAAM,gBAAgB,CAAC;AACxB,OAAO,KAAK,EAAE,WAAW,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AAEpE,4EAA4E;AAC5E,MAAM,WAAW,uBAAuB;IACtC,UAAU,EAAE,UAAU,CAAC;IACvB,qEAAqE;IACrE,OAAO,EAAE,UAAU,GAAG,IAAI,CAAC;IAC3B,iFAAiF;IACjF,gBAAgB,EAAE,MAAM,CAAC;CAC1B;AAED;;;;;GAKG;AACH,MAAM,MAAM,eAAe,GAAG,CAC5B,SAAS,EAAE,UAAU,EACrB,OAAO,EAAE,UAAU,KAChB,OAAO,CAAC,uBAAuB,CAAC,CAAC;AAEtC,MAAM,WAAW,qBAAqB;IACpC;;;;OAIG;IACH,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,yEAAyE;IACzE,0BAA0B,CAAC,EAAE,MAAM,CAAC;CACrC;AAED,MAAM,MAAM,aAAa,GAAG,QAAQ,GAAG,QAAQ,CAAC;AAEhD,MAAM,WAAW,cAAc;IAC7B,OAAO,EAAE,aAAa,CAAC;IACvB,MAAM,EAAE,MAAM,CAAC;IACf,YAAY,EAAE,MAAM,CAAC;IACrB,mBAAmB,EAAE,MAAM,GAAG,IAAI,CAAC;CACpC;AAED;;;;;GAKG;AACH,wBAAgB,YAAY,CAC1B,UAAU,EAAE,UAAU,EACtB,OAAO,EAAE,UAAU,GAAG,IAAI,EAC1B,IAAI,GAAE,qBAA0B,GAC/B,cAAc,CAyChB;AAED,MAAM,WAAW,sBAAsB;IACrC,GAAG,EAAE,aAAa,CAAC;IACnB,QAAQ,EAAE,eAAe,CAAC;IAC1B,KAAK,CAAC,EAAE,YAAY,CAAC;IACrB,QAAQ,CAAC,EAAE,qBAAqB,CAAC;IACjC,KAAK,CAAC,EAAE,WAAW,CAAC;IACpB,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAC;CAC7B;AAED,MAAM,WAAW,qBAAqB;IACpC,QAAQ,EAAE,cAAc,CAAC;IACzB,oDAAoD;IACpD,eAAe,EAAE,UAAU,CAAC;IAC5B,QAAQ,EAAE,cAAc,GAAG,IAAI,CAAC;IAChC,QAAQ,EAAE,OAAO,CAAC;IAClB,kFAAkF;IAClF,MAAM,EAAE,UAAU,CAAC;IACnB,0EAA0E;IAC1E,WAAW,EAAE,WAAW,GAAG,IAAI,CAAC;IAChC,OAAO,EAAE,UAAU,GAAG,UAAU,GAAG,SAAS,GAAG,OAAO,CAAC;IACvD,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,4EAA4E;AAC5E,wBAAsB,kBAAkB,CAAC,IAAI,EAAE,sBAAsB,GAAG,OAAO,CAAC,qBAAqB,CAAC,CA0DrG;AAED,mFAAmF;AACnF,wBAAgB,eAAe,CAAC,OAAO,EAAE,UAAU,EAAE,GAAG,MAAM,CAE7D"}