@miller-tech/uap 1.136.1 → 1.137.1

package/dist/self-tuning/flag-writer.d.ts

@@ -0,0 +1,48 @@
+/**
+ * LLM Self-Tuning — the Flag Writer (P1): atomic, validated, rollback-safe
+ * application of a tuning proposal to the real UAP configuration.
+ *
+ * A tuning step applies a `FlagChange[]` across three destinations:
+ *   - `.uap.json`      — json-kind settings (recipes.*, memory.*, delivery.*, …)
+ *   - `.uap/proxy.env` — proxyEnv-kind settings (PROXY_* the proxy loads at start)
+ *   - a `shellEnv` map — shell-kind runtime toggles the orchestrator injects into
+ *                        the benchmark child process (no file the proxy sources)
+ *
+ * It reuses the registry's `applySetting` so every write goes through the SAME
+ * bounds/enum validation as `uap config set` — an out-of-range value can never
+ * corrupt `.uap.json` and trigger the strict-parse config wipe. Before touching
+ * anything it snapshots the raw bytes of both files, so `rollback()` restores
+ * the exact prior state (including "file did not exist") after a rejected trial.
+ */
+import { FlagChange, FlagConfig } from './flags.js';
+export interface ApplyFlagsResult {
+    /** Changes that were validated and persisted (or staged into shellEnv). */
+    applied: FlagChange[];
+    /** Changes that could not be applied, with the reason. */
+    skipped: {
+        change: FlagChange;
+        reason: string;
+    }[];
+    /**
+     * Env vars for shell-kind flags. These are NOT persisted to a file (nothing
+     * sources them); the orchestrator injects them into the benchmark child env.
+     */
+    shellEnv: Record<string, string>;
+    /** Restore `.uap.json` + `.uap/proxy.env` to their exact pre-apply bytes. */
+    rollback(): void;
+}
+export interface ApplyFlagsOptions {
+    /** Validate + compute effects without writing anything (rollback is a no-op). */
+    dryRun?: boolean;
+}
+/**
+ * Apply a set of flag changes. json/proxyEnv writes go through `applySetting`
+ * (validated); shell flags are collected for the child env. Every change is
+ * value-coerced to its flag domain first, so a proposal that overshoots a range
+ * is clamped rather than rejected. Returns a `rollback()` that undoes all file
+ * writes atomically.
+ */
+export declare function applyFlagChanges(cwd: string, changes: FlagChange[], opts?: ApplyFlagsOptions): ApplyFlagsResult;
+/** Apply an entire FlagConfig (every flag → its value). Returns the same handle. */
+export declare function applyFlagConfig(cwd: string, config: FlagConfig, opts?: ApplyFlagsOptions): ApplyFlagsResult;
+//# sourceMappingURL=flag-writer.d.ts.map

package/dist/self-tuning/flag-writer.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"flag-writer.d.ts","sourceRoot":"","sources":["../../src/self-tuning/flag-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAOH,OAAO,EAAE,UAAU,EAAE,UAAU,EAAmD,MAAM,YAAY,CAAC;AA6BrG,MAAM,WAAW,gBAAgB;IAC/B,2EAA2E;IAC3E,OAAO,EAAE,UAAU,EAAE,CAAC;IACtB,0DAA0D;IAC1D,OAAO,EAAE;QAAE,MAAM,EAAE,UAAU,CAAC;QAAC,MAAM,EAAE,MAAM,CAAA;KAAE,EAAE,CAAC;IAClD;;;OAGG;IACH,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IACjC,6EAA6E;IAC7E,QAAQ,IAAI,IAAI,CAAC;CAClB;AAED,MAAM,WAAW,iBAAiB;IAChC,iFAAiF;IACjF,MAAM,CAAC,EAAE,OAAO,CAAC;CAClB;AAED;;;;;;GAMG;AACH,wBAAgB,gBAAgB,CAC9B,GAAG,EAAE,MAAM,EACX,OAAO,EAAE,UAAU,EAAE,EACrB,IAAI,GAAE,iBAAsB,GAC3B,gBAAgB,CAmDlB;AAED,oFAAoF;AACpF,wBAAgB,eAAe,CAC7B,GAAG,EAAE,MAAM,EACX,MAAM,EAAE,UAAU,EAClB,IAAI,GAAE,iBAAsB,GAC3B,gBAAgB,CAElB"}

package/dist/self-tuning/flag-writer.js

@@ -0,0 +1,102 @@
+/**
+ * LLM Self-Tuning — the Flag Writer (P1): atomic, validated, rollback-safe
+ * application of a tuning proposal to the real UAP configuration.
+ *
+ * A tuning step applies a `FlagChange[]` across three destinations:
+ *   - `.uap.json`      — json-kind settings (recipes.*, memory.*, delivery.*, …)
+ *   - `.uap/proxy.env` — proxyEnv-kind settings (PROXY_* the proxy loads at start)
+ *   - a `shellEnv` map — shell-kind runtime toggles the orchestrator injects into
+ *                        the benchmark child process (no file the proxy sources)
+ *
+ * It reuses the registry's `applySetting` so every write goes through the SAME
+ * bounds/enum validation as `uap config set` — an out-of-range value can never
+ * corrupt `.uap.json` and trigger the strict-parse config wipe. Before touching
+ * anything it snapshots the raw bytes of both files, so `rollback()` restores
+ * the exact prior state (including "file did not exist") after a rejected trial.
+ */
+import { existsSync, readFileSync, writeFileSync, rmSync, mkdirSync } from 'fs';
+import { dirname, join } from 'path';
+import { findUapConfigPath } from '../utils/config-loader.js';
+import { applySetting } from '../cli/config-command.js';
+import { getSetting } from '../config/settings-registry.js';
+import { coerceToDomain, configToChanges, getTunableFlag } from './flags.js';
+/** Resolve the two files a flag write can touch, given a cwd. */
+function targetPaths(cwd) {
+    const uapJson = findUapConfigPath(cwd) ?? join(cwd, '.uap.json');
+    const root = dirname(uapJson);
+    return { uapJson, proxyEnv: join(root, '.uap', 'proxy.env') };
+}
+function snapshot(path) {
+    const existed = existsSync(path);
+    return { path, existed, content: existed ? readFileSync(path, 'utf-8') : null };
+}
+function restore(snap) {
+    if (snap.existed && snap.content != null) {
+        mkdirSync(dirname(snap.path), { recursive: true });
+        writeFileSync(snap.path, snap.content, 'utf-8');
+    }
+    else if (!snap.existed && existsSync(snap.path)) {
+        rmSync(snap.path, { force: true });
+    }
+}
+/**
+ * Apply a set of flag changes. json/proxyEnv writes go through `applySetting`
+ * (validated); shell flags are collected for the child env. Every change is
+ * value-coerced to its flag domain first, so a proposal that overshoots a range
+ * is clamped rather than rejected. Returns a `rollback()` that undoes all file
+ * writes atomically.
+ */
+export function applyFlagChanges(cwd, changes, opts = {}) {
+    const { uapJson, proxyEnv } = targetPaths(cwd);
+    const snaps = [snapshot(uapJson), snapshot(proxyEnv)];
+    const applied = [];
+    const skipped = [];
+    const shellEnv = {};
+    for (const change of changes) {
+        const flag = getTunableFlag(change.key);
+        const setting = getSetting(change.key);
+        if (!flag || !setting) {
+            skipped.push({ change, reason: `not a tunable/registry flag: ${change.key}` });
+            continue;
+        }
+        const coerced = coerceToDomain(change.key, change.to);
+        if (coerced === null) {
+            skipped.push({ change, reason: `value out of domain: ${String(change.to)}` });
+            continue;
+        }
+        const rawValue = String(coerced);
+        // Shell-kind flags never persist to a file — stage them for the child env.
+        if (setting.kind === 'env' && setting.target === 'shell') {
+            shellEnv[setting.key] = rawValue;
+            applied.push({ ...change, to: coerced });
+            continue;
+        }
+        if (opts.dryRun) {
+            applied.push({ ...change, to: coerced });
+            continue;
+        }
+        const res = applySetting(cwd, setting, rawValue);
+        if (res.ok) {
+            applied.push({ ...change, to: coerced });
+        }
+        else {
+            skipped.push({ change, reason: res.message });
+        }
+    }
+    return {
+        applied,
+        skipped,
+        shellEnv,
+        rollback() {
+            if (opts.dryRun)
+                return;
+            for (const s of snaps)
+                restore(s);
+        },
+    };
+}
+/** Apply an entire FlagConfig (every flag → its value). Returns the same handle. */
+export function applyFlagConfig(cwd, config, opts = {}) {
+    return applyFlagChanges(cwd, configToChanges(config), opts);
+}
+//# sourceMappingURL=flag-writer.js.map

package/dist/self-tuning/flag-writer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"flag-writer.js","sourceRoot":"","sources":["../../src/self-tuning/flag-writer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAEH,OAAO,EAAE,UAAU,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,IAAI,CAAC;AAChF,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,MAAM,CAAC;AACrC,OAAO,EAAE,iBAAiB,EAAE,MAAM,2BAA2B,CAAC;AAC9D,OAAO,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AACxD,OAAO,EAAE,UAAU,EAAE,MAAM,gCAAgC,CAAC;AAC5D,OAAO,EAA0B,cAAc,EAAE,eAAe,EAAE,cAAc,EAAE,MAAM,YAAY,CAAC;AAErG,iEAAiE;AACjE,SAAS,WAAW,CAAC,GAAW;IAC9B,MAAM,OAAO,GAAG,iBAAiB,CAAC,GAAG,CAAC,IAAI,IAAI,CAAC,GAAG,EAAE,WAAW,CAAC,CAAC;IACjE,MAAM,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;IAC9B,OAAO,EAAE,OAAO,EAAE,QAAQ,EAAE,IAAI,CAAC,IAAI,EAAE,MAAM,EAAE,WAAW,CAAC,EAAE,CAAC;AAChE,CAAC;AAQD,SAAS,QAAQ,CAAC,IAAY;IAC5B,MAAM,OAAO,GAAG,UAAU,CAAC,IAAI,CAAC,CAAC;IACjC,OAAO,EAAE,IAAI,EAAE,OAAO,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,YAAY,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC;AAClF,CAAC;AAED,SAAS,OAAO,CAAC,IAAkB;IACjC,IAAI,IAAI,CAAC,OAAO,IAAI,IAAI,CAAC,OAAO,IAAI,IAAI,EAAE,CAAC;QACzC,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;QACnD,aAAa,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;IAClD,CAAC;SAAM,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QAClD,MAAM,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,KAAK,EAAE,IAAI,EAAE,CAAC,CAAC;IACrC,CAAC;AACH,CAAC;AAqBD;;;;;;GAMG;AACH,MAAM,UAAU,gBAAgB,CAC9B,GAAW,EACX,OAAqB,EACrB,OAA0B,EAAE;IAE5B,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC;IAC/C,MAAM,KAAK,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC,EAAE,QAAQ,CAAC,QAAQ,CAAC,CAAC,CAAC;IAEtD,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,MAAM,OAAO,GAA6C,EAAE,CAAC;IAC7D,MAAM,QAAQ,GAA2B,EAAE,CAAC;IAE5C,KAAK,MAAM,MAAM,IAAI,OAAO,EAAE,CAAC;QAC7B,MAAM,IAAI,GAAG,cAAc,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACxC,MAAM,OAAO,GAAG,UAAU,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QACvC,IAAI,CAAC,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;YACtB,OAAO,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,gCAAgC,MAAM,CAAC,GAAG,EAAE,EAAE,CAAC,CAAC;YAC/E,SAAS;QACX,CAAC;QACD,MAAM,OAAO,GAAG,cAAc,CAAC,MAAM,CAAC,GAAG,EAAE,MAAM,CAAC,EAAE,CAAC,CAAC;QACtD,IAAI,OAAO,KAAK,IAAI,EAAE,CAAC;YACrB,OAAO,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,wBAAwB,MAAM,CAAC,MAAM,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC;YAC9E,SAAS;QACX,CAAC;QACD,MAAM,QAAQ,GAAG,MAAM,CAAC,OAAO,CAAC,CAAC;QAEjC,2EAA2E;QAC3E,IAAI,OAAO,CAAC,IAAI,KAAK,KAAK,IAAI,OAAO,CAAC,MAAM,KAAK,OAAO,EAAE,CAAC;YACzD,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,QAAQ,CAAC;YACjC,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;YACzC,SAAS;QACX,CAAC;QAED,IAAI,IAAI,CAAC,MAAM,EAAE,CAAC;YAChB,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;YACzC,SAAS;QACX,CAAC;QAED,MAAM,GAAG,GAAG,YAAY,CAAC,GAAG,EAAE,OAAO,EAAE,QAAQ,CAAC,CAAC;QACjD,IAAI,GAAG,CAAC,EAAE,EAAE,CAAC;YACX,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,MAAM,EAAE,EAAE,EAAE,OAAO,EAAE,CAAC,CAAC;QAC3C,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,IAAI,CAAC,EAAE,MAAM,EAAE,MAAM,EAAE,GAAG,CAAC,OAAO,EAAE,CAAC,CAAC;QAChD,CAAC;IACH,CAAC;IAED,OAAO;QACL,OAAO;QACP,OAAO;QACP,QAAQ;QACR,QAAQ;YACN,IAAI,IAAI,CAAC,MAAM;gBAAE,OAAO;YACxB,KAAK,MAAM,CAAC,IAAI,KAAK;gBAAE,OAAO,CAAC,CAAC,CAAC,CAAC;QACpC,CAAC;KACF,CAAC;AACJ,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,eAAe,CAC7B,GAAW,EACX,MAAkB,EAClB,OAA0B,EAAE;IAE5B,OAAO,gBAAgB,CAAC,GAAG,EAAE,eAAe,CAAC,MAAM,CAAC,EAAE,IAAI,CAAC,CAAC;AAC9D,CAAC"}

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

@@ -0,0 +1,88 @@
+/**
+ * LLM Self-Tuning — the tunable-flag domain model (P1).
+ *
+ * This is the SEARCH SPACE. It selects the subset of the UAP settings registry
+ * that is safe + impactful to tune toward quality, and enriches each flag with
+ * what the optimizer needs: a search domain (bool / enum values / numeric range)
+ * and a DEPENDENCY (a flag only "matters" when its parent is in a given state,
+ * e.g. `recipes.fusionN` only when `recipes.recipe='fusion'`). The dependency
+ * map is the design's §11 flag-dependency graph, and it is what lets the
+ * Bayesian optimizer and the LLM tuner prune the 3^42 blow-up to a tractable
+ * space (design §3.3.3).
+ *
+ * Flag values are drawn straight from `settings-registry.ts` where it declares
+ * them (type, enumValues, min/max) so the catalog can never drift from the
+ * canonical settings; search bounds absent from the registry (e.g. a good range
+ * for `memory.shortTerm.maxEntries`) are supplied here.
+ */
+import { SettingCategoryId } from '../config/settings-registry.js';
+export type FlagValue = string | number | boolean;
+/** A flat tuning configuration: settings-registry key → value. */
+export type FlagConfig = Record<string, FlagValue>;
+/** A change to one flag (the atom the LLM tuner and flag-writer operate on). */
+export interface FlagChange {
+    /** settings-registry key, e.g. `recipes.confidenceThreshold` or `PROXY_LOOP_BREAKER`. */
+    key: string;
+    from: FlagValue | null;
+    to: FlagValue;
+    category: SettingCategoryId;
+}
+export type FlagDomain = {
+    kind: 'bool';
+} | {
+    kind: 'enum';
+    values: readonly string[];
+} | {
+    kind: 'number';
+    min: number;
+    max: number;
+    int: boolean;
+    step: number;
+};
+/** A dependency: this flag only affects behavior when `key` holds one of `values`. */
+export interface FlagDependency {
+    key: string;
+    values: readonly FlagValue[];
+}
+export interface TunableFlag {
+    key: string;
+    category: SettingCategoryId;
+    domain: FlagDomain;
+    /** Default value when unset in a config. */
+    default: FlagValue;
+    /** All dependencies must be satisfied for this flag to be "active". */
+    dependsOn: FlagDependency[];
+}
+/** The resolved tunable-flag catalog. */
+export declare const TUNABLE_FLAGS: readonly TunableFlag[];
+export declare function getTunableFlag(key: string): TunableFlag | undefined;
+/** The all-defaults configuration over every tunable flag. */
+export declare function defaultFlagConfig(): FlagConfig;
+/** Resolve a flag's effective value in a config (its value, or its default). */
+export declare function flagValue(cfg: FlagConfig, key: string): FlagValue | undefined;
+/**
+ * Is `flag` ACTIVE in `cfg` — i.e. do all its dependencies hold, TRANSITIVELY?
+ * A flag is inactive if any dependency value is unsatisfied OR the dependency
+ * flag is itself inactive (e.g. `recipes.fusionN` needs `recipes.recipe` to be
+ * `fusion`/`auto` AND `recipes.recipe` needs `recipes.enabled`). An inactive
+ * flag's value is irrelevant, so the optimizer never spends search on it and the
+ * tuner never proposes changing it. The dependency graph is acyclic.
+ */
+export declare function isFlagActive(flag: TunableFlag, cfg: FlagConfig): boolean;
+/** The keys of every flag active in `cfg`. */
+export declare function activeFlags(cfg: FlagConfig): string[];
+/** Human-readable `key = value` lines for the flags active in `cfg`. */
+export declare function describeActiveFlagsList(cfg: FlagConfig): string[];
+/**
+ * Coerce/clamp a value into a flag's domain. Numbers are clamped + optionally
+ * snapped to int; enums fall back to the default on an unknown value; bools are
+ * truth-tested. Returns null when `key` is not a tunable flag.
+ */
+export declare function coerceToDomain(key: string, value: FlagValue): FlagValue | null;
+/** Diff two configs into the minimal FlagChange[] (only keys that differ). */
+export declare function diffConfigs(from: FlagConfig, to: FlagConfig): FlagChange[];
+/** Every flag of a config expressed as a FlagChange (from unknown), for a full apply. */
+export declare function configToChanges(cfg: FlagConfig): FlagChange[];
+/** Apply FlagChange[] onto a config, returning a new config (coerced to domain). */
+export declare function applyChanges(cfg: FlagConfig, changes: FlagChange[]): FlagConfig;
+//# sourceMappingURL=flags.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"flags.d.ts","sourceRoot":"","sources":["../../src/self-tuning/flags.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAc,iBAAiB,EAAE,MAAM,gCAAgC,CAAC;AAE/E,MAAM,MAAM,SAAS,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;AAElD,kEAAkE;AAClE,MAAM,MAAM,UAAU,GAAG,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,CAAC;AAEnD,gFAAgF;AAChF,MAAM,WAAW,UAAU;IACzB,yFAAyF;IACzF,GAAG,EAAE,MAAM,CAAC;IACZ,IAAI,EAAE,SAAS,GAAG,IAAI,CAAC;IACvB,EAAE,EAAE,SAAS,CAAC;IACd,QAAQ,EAAE,iBAAiB,CAAC;CAC7B;AAED,MAAM,MAAM,UAAU,GAClB;IAAE,IAAI,EAAE,MAAM,CAAA;CAAE,GAChB;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,SAAS,MAAM,EAAE,CAAA;CAAE,GAC3C;IAAE,IAAI,EAAE,QAAQ,CAAC;IAAC,GAAG,EAAE,MAAM,CAAC;IAAC,GAAG,EAAE,MAAM,CAAC;IAAC,GAAG,EAAE,OAAO,CAAC;IAAC,IAAI,EAAE,MAAM,CAAA;CAAE,CAAC;AAE7E,sFAAsF;AACtF,MAAM,WAAW,cAAc;IAC7B,GAAG,EAAE,MAAM,CAAC;IACZ,MAAM,EAAE,SAAS,SAAS,EAAE,CAAC;CAC9B;AAED,MAAM,WAAW,WAAW;IAC1B,GAAG,EAAE,MAAM,CAAC;IACZ,QAAQ,EAAE,iBAAiB,CAAC;IAC5B,MAAM,EAAE,UAAU,CAAC;IACnB,4CAA4C;IAC5C,OAAO,EAAE,SAAS,CAAC;IACnB,uEAAuE;IACvE,SAAS,EAAE,cAAc,EAAE,CAAC;CAC7B;AAqFD,yCAAyC;AACzC,eAAO,MAAM,aAAa,EAAE,SAAS,WAAW,EAG9C,CAAC;AAIH,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,GAAG,WAAW,GAAG,SAAS,CAEnE;AAED,8DAA8D;AAC9D,wBAAgB,iBAAiB,IAAI,UAAU,CAI9C;AAED,gFAAgF;AAChF,wBAAgB,SAAS,CAAC,GAAG,EAAE,UAAU,EAAE,GAAG,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS,CAG7E;AAED;;;;;;;GAOG;AACH,wBAAgB,YAAY,CAAC,IAAI,EAAE,WAAW,EAAE,GAAG,EAAE,UAAU,GAAG,OAAO,CAQxE;AAED,8CAA8C;AAC9C,wBAAgB,WAAW,CAAC,GAAG,EAAE,UAAU,GAAG,MAAM,EAAE,CAErD;AAED,wEAAwE;AACxE,wBAAgB,uBAAuB,CAAC,GAAG,EAAE,UAAU,GAAG,MAAM,EAAE,CAIjE;AAED;;;;GAIG;AACH,wBAAgB,cAAc,CAAC,GAAG,EAAE,MAAM,EAAE,KAAK,EAAE,SAAS,GAAG,SAAS,GAAG,IAAI,CAkB9E;AAED,8EAA8E;AAC9E,wBAAgB,WAAW,CAAC,IAAI,EAAE,UAAU,EAAE,EAAE,EAAE,UAAU,GAAG,UAAU,EAAE,CAU1E;AAED,yFAAyF;AACzF,wBAAgB,eAAe,CAAC,GAAG,EAAE,UAAU,GAAG,UAAU,EAAE,CAO7D;AAED,oFAAoF;AACpF,wBAAgB,YAAY,CAAC,GAAG,EAAE,UAAU,EAAE,OAAO,EAAE,UAAU,EAAE,GAAG,UAAU,CAO/E"}

package/dist/self-tuning/flags.js

@@ -0,0 +1,189 @@
+/**
+ * LLM Self-Tuning — the tunable-flag domain model (P1).
+ *
+ * This is the SEARCH SPACE. It selects the subset of the UAP settings registry
+ * that is safe + impactful to tune toward quality, and enriches each flag with
+ * what the optimizer needs: a search domain (bool / enum values / numeric range)
+ * and a DEPENDENCY (a flag only "matters" when its parent is in a given state,
+ * e.g. `recipes.fusionN` only when `recipes.recipe='fusion'`). The dependency
+ * map is the design's §11 flag-dependency graph, and it is what lets the
+ * Bayesian optimizer and the LLM tuner prune the 3^42 blow-up to a tractable
+ * space (design §3.3.3).
+ *
+ * Flag values are drawn straight from `settings-registry.ts` where it declares
+ * them (type, enumValues, min/max) so the catalog can never drift from the
+ * canonical settings; search bounds absent from the registry (e.g. a good range
+ * for `memory.shortTerm.maxEntries`) are supplied here.
+ */
+import { getSetting } from '../config/settings-registry.js';
+const CATALOG = [
+    // ── Recipes / escalation ──────────────────────────────────────────────────
+    { key: 'recipes.enabled' },
+    { key: 'recipes.recipe', dependsOn: [{ key: 'recipes.enabled', values: [true] }] },
+    {
+        key: 'recipes.confidenceThreshold',
+        search: { min: 0.3, max: 0.8, step: 0.05 },
+        dependsOn: [{ key: 'recipes.recipe', values: ['confidence', 'auto'] }],
+    },
+    {
+        key: 'recipes.fusionN',
+        dependsOn: [{ key: 'recipes.recipe', values: ['fusion', 'auto'] }],
+    },
+    { key: 'recipes.allowSelfJudge', dependsOn: [{ key: 'recipes.enabled', values: [true] }] },
+    // ── Hands-free ────────────────────────────────────────────────────────────
+    { key: 'handsfree.enabled' },
+    { key: 'handsfree.intensity', dependsOn: [{ key: 'handsfree.enabled', values: [true] }] },
+    {
+        key: 'UAP_HANDSFREE_STAGNATION_LIMIT',
+        search: { min: 3, max: 12, int: true, step: 1 },
+        dependsOn: [{ key: 'handsfree.enabled', values: [true] }],
+        defaultOverride: 8,
+    },
+    // ── Concurrency ───────────────────────────────────────────────────────────
+    { key: 'modelConcurrency.slots', search: { min: 1, max: 8, int: true, step: 1 }, defaultOverride: 4 },
+    { key: 'modelConcurrency.adaptive' },
+    // ── Memory ────────────────────────────────────────────────────────────────
+    { key: 'memory.shortTerm.maxEntries', search: { min: 20, max: 120, int: true, step: 10 }, defaultOverride: 50 },
+    { key: 'memory.patternRag.enabled' },
+    // ── Verification / delivery ───────────────────────────────────────────────
+    { key: 'delivery.runtimeVerify' },
+    // ── Proxy guardrails (hot-reloadable-ish; applied via proxy.env) ──────────
+    { key: 'PROXY_RECON_CONVERGENCE_THRESHOLD', search: { min: 20, max: 120, int: true, step: 10 }, defaultOverride: 40 },
+    { key: 'PROXY_LOOP_BREAKER' },
+    { key: 'PROXY_STUCK_BREAK' },
+];
+function domainFor(spec) {
+    const s = getSetting(spec.key);
+    if (!s) {
+        throw new Error(`self-tuning flags: '${spec.key}' is not in the settings registry`);
+    }
+    const category = s.category;
+    if (s.type === 'boolean') {
+        return { domain: { kind: 'bool' }, category, def: spec.defaultOverride ?? (s.default ?? false) };
+    }
+    if (s.type === 'enum') {
+        const values = s.enumValues ?? [];
+        return {
+            domain: { kind: 'enum', values },
+            category,
+            def: spec.defaultOverride ?? s.default ?? values[0],
+        };
+    }
+    // number
+    const min = spec.search?.min ?? s.min ?? 0;
+    const max = spec.search?.max ?? s.max ?? Math.max(1, (Number(s.default) || 1) * 4);
+    const int = spec.search?.int ?? s.int ?? false;
+    const step = spec.search?.step ?? (int ? 1 : (max - min) / 10);
+    const def = spec.defaultOverride ?? (typeof s.default === 'number' ? s.default : (min + max) / 2);
+    return { domain: { kind: 'number', min, max, int, step }, category, def };
+}
+/** The resolved tunable-flag catalog. */
+export const TUNABLE_FLAGS = CATALOG.map((spec) => {
+    const { domain, category, def } = domainFor(spec);
+    return { key: spec.key, category, domain, default: def, dependsOn: spec.dependsOn ?? [] };
+});
+const FLAG_BY_KEY = new Map(TUNABLE_FLAGS.map((f) => [f.key, f]));
+export function getTunableFlag(key) {
+    return FLAG_BY_KEY.get(key);
+}
+/** The all-defaults configuration over every tunable flag. */
+export function defaultFlagConfig() {
+    const cfg = {};
+    for (const f of TUNABLE_FLAGS)
+        cfg[f.key] = f.default;
+    return cfg;
+}
+/** Resolve a flag's effective value in a config (its value, or its default). */
+export function flagValue(cfg, key) {
+    if (key in cfg)
+        return cfg[key];
+    return FLAG_BY_KEY.get(key)?.default;
+}
+/**
+ * Is `flag` ACTIVE in `cfg` — i.e. do all its dependencies hold, TRANSITIVELY?
+ * A flag is inactive if any dependency value is unsatisfied OR the dependency
+ * flag is itself inactive (e.g. `recipes.fusionN` needs `recipes.recipe` to be
+ * `fusion`/`auto` AND `recipes.recipe` needs `recipes.enabled`). An inactive
+ * flag's value is irrelevant, so the optimizer never spends search on it and the
+ * tuner never proposes changing it. The dependency graph is acyclic.
+ */
+export function isFlagActive(flag, cfg) {
+    for (const dep of flag.dependsOn) {
+        const v = flagValue(cfg, dep.key);
+        if (v === undefined || !dep.values.includes(v))
+            return false;
+        const depFlag = FLAG_BY_KEY.get(dep.key);
+        if (depFlag && !isFlagActive(depFlag, cfg))
+            return false; // transitive gate
+    }
+    return true;
+}
+/** The keys of every flag active in `cfg`. */
+export function activeFlags(cfg) {
+    return TUNABLE_FLAGS.filter((f) => isFlagActive(f, cfg)).map((f) => f.key);
+}
+/** Human-readable `key = value` lines for the flags active in `cfg`. */
+export function describeActiveFlagsList(cfg) {
+    return TUNABLE_FLAGS.filter((f) => isFlagActive(f, cfg)).map((f) => `${f.key} = ${JSON.stringify(flagValue(cfg, f.key))}`);
+}
+/**
+ * Coerce/clamp a value into a flag's domain. Numbers are clamped + optionally
+ * snapped to int; enums fall back to the default on an unknown value; bools are
+ * truth-tested. Returns null when `key` is not a tunable flag.
+ */
+export function coerceToDomain(key, value) {
+    const flag = FLAG_BY_KEY.get(key);
+    if (!flag)
+        return null;
+    const d = flag.domain;
+    if (d.kind === 'bool') {
+        if (typeof value === 'boolean')
+            return value;
+        const s = String(value).toLowerCase();
+        return s === 'true' || s === '1' || s === 'yes' || s === 'on';
+    }
+    if (d.kind === 'enum') {
+        return d.values.includes(String(value)) ? String(value) : flag.default;
+    }
+    // number
+    let n = Number(value);
+    if (!Number.isFinite(n))
+        n = Number(flag.default);
+    n = Math.max(d.min, Math.min(d.max, n));
+    if (d.int)
+        n = Math.round(n);
+    return n;
+}
+/** Diff two configs into the minimal FlagChange[] (only keys that differ). */
+export function diffConfigs(from, to) {
+    const changes = [];
+    for (const flag of TUNABLE_FLAGS) {
+        const a = flagValue(from, flag.key);
+        const b = flagValue(to, flag.key);
+        if (b !== undefined && a !== b) {
+            changes.push({ key: flag.key, from: a ?? null, to: b, category: flag.category });
+        }
+    }
+    return changes;
+}
+/** Every flag of a config expressed as a FlagChange (from unknown), for a full apply. */
+export function configToChanges(cfg) {
+    const changes = [];
+    for (const flag of TUNABLE_FLAGS) {
+        const v = flagValue(cfg, flag.key);
+        if (v !== undefined)
+            changes.push({ key: flag.key, from: null, to: v, category: flag.category });
+    }
+    return changes;
+}
+/** Apply FlagChange[] onto a config, returning a new config (coerced to domain). */
+export function applyChanges(cfg, changes) {
+    const next = { ...cfg };
+    for (const c of changes) {
+        const coerced = coerceToDomain(c.key, c.to);
+        if (coerced !== null)
+            next[c.key] = coerced;
+    }
+    return next;
+}
+//# sourceMappingURL=flags.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"flags.js","sourceRoot":"","sources":["../../src/self-tuning/flags.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,UAAU,EAAqB,MAAM,gCAAgC,CAAC;AAoD/E,MAAM,OAAO,GAAkB;IAC7B,6EAA6E;IAC7E,EAAE,GAAG,EAAE,iBAAiB,EAAE;IAC1B,EAAE,GAAG,EAAE,gBAAgB,EAAE,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,iBAAiB,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE;IAClF;QACE,GAAG,EAAE,6BAA6B;QAClC,MAAM,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE;QAC1C,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,gBAAgB,EAAE,MAAM,EAAE,CAAC,YAAY,EAAE,MAAM,CAAC,EAAE,CAAC;KACvE;IACD;QACE,GAAG,EAAE,iBAAiB;QACtB,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,gBAAgB,EAAE,MAAM,EAAE,CAAC,QAAQ,EAAE,MAAM,CAAC,EAAE,CAAC;KACnE;IACD,EAAE,GAAG,EAAE,wBAAwB,EAAE,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,iBAAiB,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE;IAE1F,6EAA6E;IAC7E,EAAE,GAAG,EAAE,mBAAmB,EAAE;IAC5B,EAAE,GAAG,EAAE,qBAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,mBAAmB,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE;IACzF;QACE,GAAG,EAAE,gCAAgC;QACrC,MAAM,EAAE,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE;QAC/C,SAAS,EAAE,CAAC,EAAE,GAAG,EAAE,mBAAmB,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC;QACzD,eAAe,EAAE,CAAC;KACnB;IAED,6EAA6E;IAC7E,EAAE,GAAG,EAAE,wBAAwB,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,eAAe,EAAE,CAAC,EAAE;IACrG,EAAE,GAAG,EAAE,2BAA2B,EAAE;IAEpC,6EAA6E;IAC7E,EAAE,GAAG,EAAE,6BAA6B,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,EAAE,EAAE,eAAe,EAAE,EAAE,EAAE;IAC/G,EAAE,GAAG,EAAE,2BAA2B,EAAE;IAEpC,6EAA6E;IAC7E,EAAE,GAAG,EAAE,wBAAwB,EAAE;IAEjC,6EAA6E;IAC7E,EAAE,GAAG,EAAE,mCAAmC,EAAE,MAAM,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,EAAE,EAAE,eAAe,EAAE,EAAE,EAAE;IACrH,EAAE,GAAG,EAAE,oBAAoB,EAAE;IAC7B,EAAE,GAAG,EAAE,mBAAmB,EAAE;CAC7B,CAAC;AAEF,SAAS,SAAS,CAAC,IAAiB;IAClC,MAAM,CAAC,GAAG,UAAU,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAC/B,IAAI,CAAC,CAAC,EAAE,CAAC;QACP,MAAM,IAAI,KAAK,CAAC,uBAAuB,IAAI,CAAC,GAAG,mCAAmC,CAAC,CAAC;IACtF,CAAC;IACD,MAAM,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC;IAC5B,IAAI,CAAC,CAAC,IAAI,KAAK,SAAS,EAAE,CAAC;QACzB,OAAO,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,IAAI,CAAC,eAAe,IAAI,CAAC,CAAC,CAAC,OAAkB,IAAI,KAAK,CAAC,EAAE,CAAC;IAC9G,CAAC;IACD,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,CAAC,CAAC,UAAU,IAAI,EAAE,CAAC;QAClC,OAAO;YACL,MAAM,EAAE,EAAE,IAAI,EAAE,MAAM,EAAE,MAAM,EAAE;YAChC,QAAQ;YACR,GAAG,EAAE,IAAI,CAAC,eAAe,IAAK,CAAC,CAAC,OAAkB,IAAI,MAAM,CAAC,CAAC,CAAC;SAChE,CAAC;IACJ,CAAC;IACD,SAAS;IACT,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,GAAG,IAAI,CAAC,CAAC;IAC3C,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC;IACnF,MAAM,GAAG,GAAG,IAAI,CAAC,MAAM,EAAE,GAAG,IAAI,CAAC,CAAC,GAAG,IAAI,KAAK,CAAC;IAC/C,MAAM,IAAI,GAAG,IAAI,CAAC,MAAM,EAAE,IAAI,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;IAC/D,MAAM,GAAG,GAAG,IAAI,CAAC,eAAe,IAAI,CAAC,OAAO,CAAC,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,GAAG,GAAG,GAAG,CAAC,GAAG,CAAC,CAAC,CAAC;IAClG,OAAO,EAAE,MAAM,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC;AAC5E,CAAC;AAED,yCAAyC;AACzC,MAAM,CAAC,MAAM,aAAa,GAA2B,OAAO,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE;IACxE,MAAM,EAAE,MAAM,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,SAAS,CAAC,IAAI,CAAC,CAAC;IAClD,OAAO,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,QAAQ,EAAE,MAAM,EAAE,OAAO,EAAE,GAAG,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,IAAI,EAAE,EAAE,CAAC;AAC5F,CAAC,CAAC,CAAC;AAEH,MAAM,WAAW,GAAG,IAAI,GAAG,CAAC,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC;AAElE,MAAM,UAAU,cAAc,CAAC,GAAW;IACxC,OAAO,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;AAC9B,CAAC;AAED,8DAA8D;AAC9D,MAAM,UAAU,iBAAiB;IAC/B,MAAM,GAAG,GAAe,EAAE,CAAC;IAC3B,KAAK,MAAM,CAAC,IAAI,aAAa;QAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC,OAAO,CAAC;IACtD,OAAO,GAAG,CAAC;AACb,CAAC;AAED,gFAAgF;AAChF,MAAM,UAAU,SAAS,CAAC,GAAe,EAAE,GAAW;IACpD,IAAI,GAAG,IAAI,GAAG;QAAE,OAAO,GAAG,CAAC,GAAG,CAAC,CAAC;IAChC,OAAO,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,OAAO,CAAC;AACvC,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,YAAY,CAAC,IAAiB,EAAE,GAAe;IAC7D,KAAK,MAAM,GAAG,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;QACjC,MAAM,CAAC,GAAG,SAAS,CAAC,GAAG,EAAE,GAAG,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,GAAG,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;YAAE,OAAO,KAAK,CAAC;QAC7D,MAAM,OAAO,GAAG,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;QACzC,IAAI,OAAO,IAAI,CAAC,YAAY,CAAC,OAAO,EAAE,GAAG,CAAC;YAAE,OAAO,KAAK,CAAC,CAAC,kBAAkB;IAC9E,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,8CAA8C;AAC9C,MAAM,UAAU,WAAW,CAAC,GAAe;IACzC,OAAO,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;AAC7E,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,uBAAuB,CAAC,GAAe;IACrD,OAAO,aAAa,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,YAAY,CAAC,CAAC,EAAE,GAAG,CAAC,CAAC,CAAC,GAAG,CAC1D,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,CAAC,GAAG,MAAM,IAAI,CAAC,SAAS,CAAC,SAAS,CAAC,GAAG,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,EAAE,CAC7D,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAAC,GAAW,EAAE,KAAgB;IAC1D,MAAM,IAAI,GAAG,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;IAClC,IAAI,CAAC,IAAI;QAAE,OAAO,IAAI,CAAC;IACvB,MAAM,CAAC,GAAG,IAAI,CAAC,MAAM,CAAC;IACtB,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACtB,IAAI,OAAO,KAAK,KAAK,SAAS;YAAE,OAAO,KAAK,CAAC;QAC7C,MAAM,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC,WAAW,EAAE,CAAC;QACtC,OAAO,CAAC,KAAK,MAAM,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,KAAK,IAAI,CAAC,KAAK,IAAI,CAAC;IAChE,CAAC;IACD,IAAI,CAAC,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACtB,OAAO,CAAC,CAAC,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC;IACzE,CAAC;IACD,SAAS;IACT,IAAI,CAAC,GAAG,MAAM,CAAC,KAAK,CAAC,CAAC;IACtB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC;QAAE,CAAC,GAAG,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAClD,CAAC,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,CAAC;IACxC,IAAI,CAAC,CAAC,GAAG;QAAE,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;IAC7B,OAAO,CAAC,CAAC;AACX,CAAC;AAED,8EAA8E;AAC9E,MAAM,UAAU,WAAW,CAAC,IAAgB,EAAE,EAAc;IAC1D,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,aAAa,EAAE,CAAC;QACjC,MAAM,CAAC,GAAG,SAAS,CAAC,IAAI,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QACpC,MAAM,CAAC,GAAG,SAAS,CAAC,EAAE,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,CAAC,KAAK,SAAS,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;YAC/B,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,CAAC,IAAI,IAAI,EAAE,EAAE,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;QACnF,CAAC;IACH,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,yFAAyF;AACzF,MAAM,UAAU,eAAe,CAAC,GAAe;IAC7C,MAAM,OAAO,GAAiB,EAAE,CAAC;IACjC,KAAK,MAAM,IAAI,IAAI,aAAa,EAAE,CAAC;QACjC,MAAM,CAAC,GAAG,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC;QACnC,IAAI,CAAC,KAAK,SAAS;YAAE,OAAO,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC,EAAE,QAAQ,EAAE,IAAI,CAAC,QAAQ,EAAE,CAAC,CAAC;IACnG,CAAC;IACD,OAAO,OAAO,CAAC;AACjB,CAAC;AAED,oFAAoF;AACpF,MAAM,UAAU,YAAY,CAAC,GAAe,EAAE,OAAqB;IACjE,MAAM,IAAI,GAAG,EAAE,GAAG,GAAG,EAAE,CAAC;IACxB,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,OAAO,GAAG,cAAc,CAAC,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,EAAE,CAAC,CAAC;QAC5C,IAAI,OAAO,KAAK,IAAI;YAAE,IAAI,CAAC,CAAC,CAAC,GAAG,CAAC,GAAG,OAAO,CAAC;IAC9C,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC"}

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

@@ -0,0 +1,25 @@
+/**
+ * LLM Self-Tuning — raising a small model (qwen3.6) toward Opus 4.8 by tuning
+ * UAP's own flag surface with a closed, benchmark-validated loop.
+ *
+ * Pipeline (see docs/design/LLM_SELF_TUNING_ANALYSIS.md):
+ *   P0 quality-scorer  — a multi-dimensional quality signal beyond pass/fail
+ *   P1 llm-tuner       — LLM-guided flag proposals over the search space
+ *      search-reducer  — a real Gaussian-process Bayesian optimizer + pruning
+ *      flag-writer     — atomic, validated, rollback-safe config writes
+ *      tuning-profile  — model-specific best-config storage
+ *   P2 orchestrator/run— the propose → apply → validate → decide → learn loop
+ *   P4 realtime-adaptor— per-session flag adaptation over the live signal channel
+ */
+export * from './judge.js';
+export * from './quality-scorer.js';
+export * from './flags.js';
+export * from './flag-writer.js';
+export * from './tuning-profile.js';
+export * from './search-reducer.js';
+export * from './llm-tuner.js';
+export * from './orchestrator.js';
+export * from './paired-validator.js';
+export * from './run.js';
+export * from './realtime-adaptor.js';
+//# sourceMappingURL=index.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/self-tuning/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,cAAc,YAAY,CAAC;AAC3B,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,kBAAkB,CAAC;AACjC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,mBAAmB,CAAC;AAClC,cAAc,uBAAuB,CAAC;AACtC,cAAc,UAAU,CAAC;AACzB,cAAc,uBAAuB,CAAC"}

package/dist/self-tuning/index.js

@@ -0,0 +1,25 @@
+/**
+ * LLM Self-Tuning — raising a small model (qwen3.6) toward Opus 4.8 by tuning
+ * UAP's own flag surface with a closed, benchmark-validated loop.
+ *
+ * Pipeline (see docs/design/LLM_SELF_TUNING_ANALYSIS.md):
+ *   P0 quality-scorer  — a multi-dimensional quality signal beyond pass/fail
+ *   P1 llm-tuner       — LLM-guided flag proposals over the search space
+ *      search-reducer  — a real Gaussian-process Bayesian optimizer + pruning
+ *      flag-writer     — atomic, validated, rollback-safe config writes
+ *      tuning-profile  — model-specific best-config storage
+ *   P2 orchestrator/run— the propose → apply → validate → decide → learn loop
+ *   P4 realtime-adaptor— per-session flag adaptation over the live signal channel
+ */
+export * from './judge.js';
+export * from './quality-scorer.js';
+export * from './flags.js';
+export * from './flag-writer.js';
+export * from './tuning-profile.js';
+export * from './search-reducer.js';
+export * from './llm-tuner.js';
+export * from './orchestrator.js';
+export * from './paired-validator.js';
+export * from './run.js';
+export * from './realtime-adaptor.js';
+//# sourceMappingURL=index.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"index.js","sourceRoot":"","sources":["../../src/self-tuning/index.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,cAAc,YAAY,CAAC;AAC3B,cAAc,qBAAqB,CAAC;AACpC,cAAc,YAAY,CAAC;AAC3B,cAAc,kBAAkB,CAAC;AACjC,cAAc,qBAAqB,CAAC;AACpC,cAAc,qBAAqB,CAAC;AACpC,cAAc,gBAAgB,CAAC;AAC/B,cAAc,mBAAmB,CAAC;AAClC,cAAc,uBAAuB,CAAC;AACtC,cAAc,UAAU,CAAC;AACzB,cAAc,uBAAuB,CAAC"}

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

@@ -0,0 +1,66 @@
+/**
+ * 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 type { ModelConfig } from '../models/types.js';
+/** A minimal prompt→text completion seam for the judge/tuner model. */
+export interface JudgeClient {
+    /** Human-readable id (model id) for logs and provenance. */
+    readonly id: string;
+    /** Complete a prompt and return the raw text. `json` requests grammar-constrained JSON. */
+    complete(prompt: string, opts?: {
+        json?: boolean;
+        temperature?: number;
+        maxTokens?: number;
+    }): Promise<string>;
+}
+export interface ResolveJudgeOptions {
+    /** Explicit judge model: a `ModelPresets` key, a preset `apiModel`, or a bare model id. */
+    judgeModel?: string;
+    /** Endpoint override (e.g. a cloud gateway). Defaults to the client's default. */
+    endpoint?: string;
+    /** Read `recipes.judge.model` from `.uap.json` in this cwd when `judgeModel` is unset. */
+    cwd?: string;
+    /** Fallback preset id when nothing resolves. Default 'opus-4.8'. */
+    fallbackPreset?: string;
+    /** When true, do NOT fall back to a default preset — return null if unresolved. */
+    requireExplicit?: boolean;
+}
+/**
+ * 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 declare function resolveJudgeModelConfig(id: string, presets: Record<string, ModelConfig>): ModelConfig;
+/**
+ * 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 declare function resolveJudgeClient(opts?: ResolveJudgeOptions): Promise<JudgeClient | null>;
+/**
+ * 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 declare function parseJsonLenient<T = unknown>(text: string): T | null;
+//# sourceMappingURL=judge.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../src/self-tuning/judge.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;GAoBG;AAEH,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,oBAAoB,CAAC;AAGtD,uEAAuE;AACvE,MAAM,WAAW,WAAW;IAC1B,4DAA4D;IAC5D,QAAQ,CAAC,EAAE,EAAE,MAAM,CAAC;IACpB,2FAA2F;IAC3F,QAAQ,CAAC,MAAM,EAAE,MAAM,EAAE,IAAI,CAAC,EAAE;QAAE,IAAI,CAAC,EAAE,OAAO,CAAC;QAAC,WAAW,CAAC,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAA;KAAE,GAAG,OAAO,CAAC,MAAM,CAAC,CAAC;CAChH;AAED,MAAM,WAAW,mBAAmB;IAClC,2FAA2F;IAC3F,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,kFAAkF;IAClF,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,0FAA0F;IAC1F,GAAG,CAAC,EAAE,MAAM,CAAC;IACb,oEAAoE;IACpE,cAAc,CAAC,EAAE,MAAM,CAAC;IACxB,mFAAmF;IACnF,eAAe,CAAC,EAAE,OAAO,CAAC;CAC3B;AAED;;;;;GAKG;AACH,wBAAgB,uBAAuB,CACrC,EAAE,EAAE,MAAM,EACV,OAAO,EAAE,MAAM,CAAC,MAAM,EAAE,WAAW,CAAC,GACnC,WAAW,CAeb;AAED;;;;GAIG;AACH,wBAAsB,kBAAkB,CAAC,IAAI,GAAE,mBAAwB,GAAG,OAAO,CAAC,WAAW,GAAG,IAAI,CAAC,CA6BpG;AAED;;;;;GAKG;AACH,wBAAgB,gBAAgB,CAAC,CAAC,GAAG,OAAO,EAAE,IAAI,EAAE,MAAM,GAAG,CAAC,GAAG,IAAI,CAyCpE"}