@miller-tech/uap 1.59.0 → 1.61.0

package/dist/self-harness/mine.js

@@ -0,0 +1,78 @@
+/**
+ * Self-Harness — Stage 1 weakness mining from paired-bench run records.
+ *
+ * P1 mines from the structured `RunRecord`s the paired bench already produces
+ * (correct, error, latency, turns) — a HALO-trace miner is additive (P3). Each
+ * failing record is classified into a `FailureKind`; classifications aggregate
+ * into ranked `WeaknessReport`s keyed by a stable signature. Pure + testable.
+ */
+import { signatureHash, rankWeaknesses, } from './weakness.js';
+/**
+ * Classify a single run's failure into a FailureKind, or null if it succeeded.
+ * Heuristics are derived from the observed Qwen3.6-A3B failure modes
+ * (see memory project_uap_paired_bench): timeout vs verify-fail vs runaway etc.
+ */
+export function classifyFailure(rec) {
+    if (rec.metrics.correct)
+        return null;
+    const err = (rec.metrics.error ?? '').toLowerCase();
+    if (err.includes('timed out') || err.includes('timeout'))
+        return 'agent.timeout';
+    if (err.includes('verify failed')) {
+        // A verify failure with very high turns is a non-termination/garble loop;
+        // otherwise it's a plain wrong answer.
+        if ((rec.metrics.turns ?? 0) >= 40)
+            return 'loop.nonterminate';
+        return 'verify.fail';
+    }
+    if (err.includes('exit') || err.includes('spawn'))
+        return 'agent.error';
+    // Fell through with no error but incorrect → treat as a wrong answer.
+    return 'verify.fail';
+}
+/** Mine ranked weaknesses from a set of paired-bench run records. */
+export function mineFromRecords(records, opts) {
+    const minFreq = opts.minFrequency ?? 1;
+    const byKind = new Map();
+    for (const rec of records) {
+        const kind = classifyFailure(rec);
+        if (!kind)
+            continue;
+        let agg = byKind.get(kind);
+        if (!agg) {
+            agg = { tasks: new Set(), evidence: [], count: 0 };
+            byKind.set(kind, agg);
+        }
+        agg.count += 1;
+        agg.tasks.add(rec.taskId);
+        if (agg.evidence.length < 5) {
+            agg.evidence.push(`${rec.taskId}#${rec.seed} ${rec.condition}: ${rec.metrics.error ?? 'incorrect'}`);
+        }
+    }
+    const reports = [];
+    for (const [kind, agg] of byKind) {
+        if (agg.count < minFreq)
+            continue;
+        reports.push({
+            signature: signatureHash({ kind, model: opts.model }),
+            kind,
+            model: opts.model,
+            frequency: agg.count,
+            affectedTasks: [...agg.tasks],
+            hypothesis: HYPOTHESIS[kind],
+            evidence: agg.evidence,
+        });
+    }
+    return rankWeaknesses(reports);
+}
+const HYPOTHESIS = {
+    'toolcall.path.garbled': 'The model mangles tool-call file paths (case/extension/stray dirs) and never lands a correct edit.',
+    'toolcall.args.truncated': 'Tool-call arguments are cut off mid-value, producing empty/partial writes.',
+    'gen.runaway.npredict': 'A single turn generates up to the n-predict cap, consuming the wall-clock budget.',
+    'loop.nonterminate': 'The agent keeps emitting tool calls without converging on a terminal answer.',
+    'guardrail.poison.recon': 'A guardrail (recon-convergence) wedged tool access across requests.',
+    'verify.fail': 'The agent terminates and submits, but the deterministic verifier rejects the answer.',
+    'agent.timeout': 'The agent exceeded its wall-clock budget without producing a verifiable result.',
+    'agent.error': 'The agent process exited non-zero or failed to spawn.',
+};
+//# sourceMappingURL=mine.js.map

package/dist/self-harness/mine.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mine.js","sourceRoot":"","sources":["../../src/self-harness/mine.ts"],"names":[],"mappings":"AAAA;;;;;;;GAOG;AAGH,OAAO,EAGL,aAAa,EACb,cAAc,GACf,MAAM,eAAe,CAAC;AAEvB;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAAC,GAAc;IAC5C,IAAI,GAAG,CAAC,OAAO,CAAC,OAAO;QAAE,OAAO,IAAI,CAAC;IACrC,MAAM,GAAG,GAAG,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,EAAE,CAAC,CAAC,WAAW,EAAE,CAAC;IACpD,IAAI,GAAG,CAAC,QAAQ,CAAC,WAAW,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,SAAS,CAAC;QAAE,OAAO,eAAe,CAAC;IACjF,IAAI,GAAG,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE,CAAC;QAClC,0EAA0E;QAC1E,uCAAuC;QACvC,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,CAAC,CAAC,IAAI,EAAE;YAAE,OAAO,mBAAmB,CAAC;QAC/D,OAAO,aAAa,CAAC;IACvB,CAAC;IACD,IAAI,GAAG,CAAC,QAAQ,CAAC,MAAM,CAAC,IAAI,GAAG,CAAC,QAAQ,CAAC,OAAO,CAAC;QAAE,OAAO,aAAa,CAAC;IACxE,sEAAsE;IACtE,OAAO,aAAa,CAAC;AACvB,CAAC;AASD,qEAAqE;AACrE,MAAM,UAAU,eAAe,CAAC,OAAoB,EAAE,IAAiB;IACrE,MAAM,OAAO,GAAG,IAAI,CAAC,YAAY,IAAI,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,IAAI,GAAG,EAA0E,CAAC;IAEjG,KAAK,MAAM,GAAG,IAAI,OAAO,EAAE,CAAC;QAC1B,MAAM,IAAI,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;QAClC,IAAI,CAAC,IAAI;YAAE,SAAS;QACpB,IAAI,GAAG,GAAG,MAAM,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;QAC3B,IAAI,CAAC,GAAG,EAAE,CAAC;YACT,GAAG,GAAG,EAAE,KAAK,EAAE,IAAI,GAAG,EAAE,EAAE,QAAQ,EAAE,EAAE,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;YACnD,MAAM,CAAC,GAAG,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC;QACxB,CAAC;QACD,GAAG,CAAC,KAAK,IAAI,CAAC,CAAC;QACf,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;QAC1B,IAAI,GAAG,CAAC,QAAQ,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAC5B,GAAG,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,IAAI,IAAI,GAAG,CAAC,SAAS,KAAK,GAAG,CAAC,OAAO,CAAC,KAAK,IAAI,WAAW,EAAE,CAAC,CAAC;QACvG,CAAC;IACH,CAAC;IAED,MAAM,OAAO,GAAqB,EAAE,CAAC;IACrC,KAAK,MAAM,CAAC,IAAI,EAAE,GAAG,CAAC,IAAI,MAAM,EAAE,CAAC;QACjC,IAAI,GAAG,CAAC,KAAK,GAAG,OAAO;YAAE,SAAS;QAClC,OAAO,CAAC,IAAI,CAAC;YACX,SAAS,EAAE,aAAa,CAAC,EAAE,IAAI,EAAE,KAAK,EAAE,IAAI,CAAC,KAAK,EAAE,CAAC;YACrD,IAAI;YACJ,KAAK,EAAE,IAAI,CAAC,KAAK;YACjB,SAAS,EAAE,GAAG,CAAC,KAAK;YACpB,aAAa,EAAE,CAAC,GAAG,GAAG,CAAC,KAAK,CAAC;YAC7B,UAAU,EAAE,UAAU,CAAC,IAAI,CAAC;YAC5B,QAAQ,EAAE,GAAG,CAAC,QAAQ;SACvB,CAAC,CAAC;IACL,CAAC;IACD,OAAO,cAAc,CAAC,OAAO,CAAC,CAAC;AACjC,CAAC;AAED,MAAM,UAAU,GAAgC;IAC9C,uBAAuB,EACrB,oGAAoG;IACtG,yBAAyB,EACvB,4EAA4E;IAC9E,sBAAsB,EACpB,mFAAmF;IACrF,mBAAmB,EACjB,8EAA8E;IAChF,wBAAwB,EACtB,qEAAqE;IACvE,aAAa,EACX,sFAAsF;IACxF,eAAe,EACb,iFAAiF;IACnF,aAAa,EACX,uDAAuD;CAC1D,CAAC"}

package/dist/self-harness/mods.d.ts

@@ -0,0 +1,136 @@
+/**
+ * Self-Harness — the Modification ("Mod") DSL.
+ *
+ * The autonomous proposer may ONLY emit values of the `Mod` union below — never
+ * free-form code. Each Mod is mechanically applicable and automatically
+ * reversible, and `env` Mods are constrained to an allow-list of knobs with
+ * declared safe ranges. This is the "minimal, targeted modification" surface
+ * from the Self-Harness paper (arXiv:2606.09498), widened to UAP's three layers
+ * (scaffold text, proxy/llama env knobs, proxy middleware).
+ *
+ * See docs/design/SELF_HARNESS.md.
+ */
+import { UapComponent } from '../benchmarks/paired/types.js';
+export type JsonScalar = string | number | boolean;
+export interface NumericKnob {
+    target: 'proxy' | 'llama';
+    type: 'number';
+    /** Inclusive safe range. A proposed value outside this is rejected. */
+    min: number;
+    max: number;
+    /** Whether the value must be an integer. */
+    integer?: boolean;
+    description: string;
+}
+export interface EnumKnob {
+    target: 'proxy' | 'llama';
+    type: 'enum';
+    values: readonly string[];
+    description: string;
+}
+export type KnobSpec = NumericKnob | EnumKnob;
+export declare const KNOB_ALLOWLIST: {
+    readonly LLAMA_N_PREDICT: {
+        readonly target: "llama";
+        readonly type: "number";
+        readonly min: 512;
+        readonly max: 16384;
+        readonly integer: true;
+        readonly description: "per-turn generation cap; low values bound a runaway turn";
+    };
+    readonly LLAMA_REPEAT_PENALTY: {
+        readonly target: "llama";
+        readonly type: "number";
+        readonly min: 1;
+        readonly max: 1.3;
+        readonly description: "repetition penalty; damps token-level loops";
+    };
+    readonly PROXY_HARD_FINALIZE_TURNS: {
+        readonly target: "proxy";
+        readonly type: "number";
+        readonly min: 10;
+        readonly max: 80;
+        readonly integer: true;
+        readonly description: "tool-turn ceiling before the proxy strips tools to force termination";
+    };
+    readonly PROXY_RECON_CONVERGENCE_THRESHOLD: {
+        readonly target: "proxy";
+        readonly type: "number";
+        readonly min: 20;
+        readonly max: 200;
+        readonly integer: true;
+        readonly description: "no-write turns before the recon-convergence directive fires";
+    };
+    readonly PROXY_TOOL_NARROWING_KEEP: {
+        readonly target: "proxy";
+        readonly type: "number";
+        readonly min: 2;
+        readonly max: 24;
+        readonly integer: true;
+        readonly description: "how many tools to retain after relevance narrowing (soft floor)";
+    };
+    readonly PROXY_FORCED_THRESHOLD: {
+        readonly target: "proxy";
+        readonly type: "number";
+        readonly min: 5;
+        readonly max: 40;
+        readonly integer: true;
+        readonly description: "consecutive forced tool_choice=required turns before loop-breaker release";
+    };
+};
+export type KnownKnob = keyof typeof KNOB_ALLOWLIST;
+export declare function isKnownKnob(k: string): k is KnownKnob;
+/**
+ * Look up a knob spec as the full `KnobSpec` union. Indexing KNOB_ALLOWLIST
+ * directly preserves each entry's narrow literal type (all numeric today), which
+ * makes the enum branch in validateMod unreachable (`never`); returning the union
+ * keeps both branches type-checkable as enum knobs are added later.
+ */
+export declare function knobSpec(key: KnownKnob): KnobSpec;
+export declare const MIDDLEWARE_IDS: readonly ["toolcall-path-normalizer"];
+export type MiddlewareId = (typeof MIDDLEWARE_IDS)[number];
+export declare function isMiddlewareId(s: string): s is MiddlewareId;
+export interface EnvMod {
+    kind: 'env';
+    key: KnownKnob;
+    /** Prior value (for revert + audit). */
+    from: string;
+    /** Proposed value, as it will be written to the env file. */
+    to: string;
+}
+export interface ScaffoldMod {
+    kind: 'scaffold';
+    component: UapComponent;
+    op: 'replace' | 'append';
+    /** New text (replace) or text to append. */
+    text: string;
+    /** Prior text of the component block (for revert). */
+    from?: string;
+}
+export interface MiddlewareMod {
+    kind: 'middleware';
+    id: MiddlewareId;
+    params: Record<string, JsonScalar>;
+}
+export type Mod = EnvMod | ScaffoldMod | MiddlewareMod;
+export interface ValidationOk {
+    ok: true;
+}
+export interface ValidationErr {
+    ok: false;
+    reason: string;
+}
+export type ValidationResult = ValidationOk | ValidationErr;
+/**
+ * Validate a Mod against the DSL constraints. A proposer's output that fails
+ * this is rejected at parse time — it never reaches the apply/validate stages.
+ */
+export declare function validateMod(mod: Mod): ValidationResult;
+/** One-line, human-readable description of a Mod (for logs and profile diffs). */
+export declare function describeMod(mod: Mod): string;
+/**
+ * Produce the inverse Mod that undoes `mod` (for rollback). Env/scaffold revert
+ * to their captured prior value; a middleware Mod reverts by disabling it.
+ */
+export declare function invertMod(mod: Mod): Mod;
+//# sourceMappingURL=mods.d.ts.map

package/dist/self-harness/mods.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"mods.d.ts","sourceRoot":"","sources":["../../src/self-harness/mods.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAE,YAAY,EAAkB,MAAM,+BAA+B,CAAC;AAE7E,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;AASnD,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,OAAO,GAAG,OAAO,CAAC;IAC1B,IAAI,EAAE,QAAQ,CAAC;IACf,uEAAuE;IACvE,GAAG,EAAE,MAAM,CAAC;IACZ,GAAG,EAAE,MAAM,CAAC;IACZ,4CAA4C;IAC5C,OAAO,CAAC,EAAE,OAAO,CAAC;IAClB,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,QAAQ;IACvB,MAAM,EAAE,OAAO,GAAG,OAAO,CAAC;IAC1B,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,SAAS,MAAM,EAAE,CAAC;IAC1B,WAAW,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,MAAM,QAAQ,GAAG,WAAW,GAAG,QAAQ,CAAC;AAE9C,eAAO,MAAM,cAAc;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA2BkB,CAAC;AAE9C,MAAM,MAAM,SAAS,GAAG,MAAM,OAAO,cAAc,CAAC;AAEpD,wBAAgB,WAAW,CAAC,CAAC,EAAE,MAAM,GAAG,CAAC,IAAI,SAAS,CAErD;AAED;;;;;GAKG;AACH,wBAAgB,QAAQ,CAAC,GAAG,EAAE,SAAS,GAAG,QAAQ,CAEjD;AASD,eAAO,MAAM,cAAc,uCAAwC,CAAC;AACpE,MAAM,MAAM,YAAY,GAAG,CAAC,OAAO,cAAc,CAAC,CAAC,MAAM,CAAC,CAAC;AAE3D,wBAAgB,cAAc,CAAC,CAAC,EAAE,MAAM,GAAG,CAAC,IAAI,YAAY,CAE3D;AAMD,MAAM,WAAW,MAAM;IACrB,IAAI,EAAE,KAAK,CAAC;IACZ,GAAG,EAAE,SAAS,CAAC;IACf,wCAAwC;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,6DAA6D;IAC7D,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,MAAM,WAAW,WAAW;IAC1B,IAAI,EAAE,UAAU,CAAC;IACjB,SAAS,EAAE,YAAY,CAAC;IACxB,EAAE,EAAE,SAAS,GAAG,QAAQ,CAAC;IACzB,4CAA4C;IAC5C,IAAI,EAAE,MAAM,CAAC;IACb,sDAAsD;IACtD,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,MAAM,WAAW,aAAa;IAC5B,IAAI,EAAE,YAAY,CAAC;IACnB,EAAE,EAAE,YAAY,CAAC;IACjB,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,UAAU,CAAC,CAAC;CACpC;AAED,MAAM,MAAM,GAAG,GAAG,MAAM,GAAG,WAAW,GAAG,aAAa,CAAC;AAEvD,MAAM,WAAW,YAAY;IAC3B,EAAE,EAAE,IAAI,CAAC;CACV;AACD,MAAM,WAAW,aAAa;IAC5B,EAAE,EAAE,KAAK,CAAC;IACV,MAAM,EAAE,MAAM,CAAC;CAChB;AACD,MAAM,MAAM,gBAAgB,GAAG,YAAY,GAAG,aAAa,CAAC;AAE5D;;;GAGG;AACH,wBAAgB,WAAW,CAAC,GAAG,EAAE,GAAG,GAAG,gBAAgB,CAkDtD;AAED,kFAAkF;AAClF,wBAAgB,WAAW,CAAC,GAAG,EAAE,GAAG,GAAG,MAAM,CAS5C;AAED;;;GAGG;AACH,wBAAgB,SAAS,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,CAiBvC"}

package/dist/self-harness/mods.js

@@ -0,0 +1,152 @@
+/**
+ * Self-Harness — the Modification ("Mod") DSL.
+ *
+ * The autonomous proposer may ONLY emit values of the `Mod` union below — never
+ * free-form code. Each Mod is mechanically applicable and automatically
+ * reversible, and `env` Mods are constrained to an allow-list of knobs with
+ * declared safe ranges. This is the "minimal, targeted modification" surface
+ * from the Self-Harness paper (arXiv:2606.09498), widened to UAP's three layers
+ * (scaffold text, proxy/llama env knobs, proxy middleware).
+ *
+ * See docs/design/SELF_HARNESS.md.
+ */
+import { UAP_COMPONENTS } from '../benchmarks/paired/types.js';
+export const KNOB_ALLOWLIST = {
+    // --- llama sampler knobs (hot-reload via env + server restart, no slot clear) ---
+    LLAMA_N_PREDICT: {
+        target: 'llama', type: 'number', min: 512, max: 16384, integer: true,
+        description: 'per-turn generation cap; low values bound a runaway turn',
+    },
+    LLAMA_REPEAT_PENALTY: {
+        target: 'llama', type: 'number', min: 1.0, max: 1.3,
+        description: 'repetition penalty; damps token-level loops',
+    },
+    // --- proxy guardrail knobs (live per-request, no restart) ---
+    PROXY_HARD_FINALIZE_TURNS: {
+        target: 'proxy', type: 'number', min: 10, max: 80, integer: true,
+        description: 'tool-turn ceiling before the proxy strips tools to force termination',
+    },
+    PROXY_RECON_CONVERGENCE_THRESHOLD: {
+        target: 'proxy', type: 'number', min: 20, max: 200, integer: true,
+        description: 'no-write turns before the recon-convergence directive fires',
+    },
+    PROXY_TOOL_NARROWING_KEEP: {
+        target: 'proxy', type: 'number', min: 2, max: 24, integer: true,
+        description: 'how many tools to retain after relevance narrowing (soft floor)',
+    },
+    PROXY_FORCED_THRESHOLD: {
+        target: 'proxy', type: 'number', min: 5, max: 40, integer: true,
+        description: 'consecutive forced tool_choice=required turns before loop-breaker release',
+    },
+};
+export function isKnownKnob(k) {
+    return Object.prototype.hasOwnProperty.call(KNOB_ALLOWLIST, k);
+}
+/**
+ * Look up a knob spec as the full `KnobSpec` union. Indexing KNOB_ALLOWLIST
+ * directly preserves each entry's narrow literal type (all numeric today), which
+ * makes the enum branch in validateMod unreachable (`never`); returning the union
+ * keeps both branches type-checkable as enum knobs are added later.
+ */
+export function knobSpec(key) {
+    return KNOB_ALLOWLIST[key];
+}
+// ---------------------------------------------------------------------------
+// Middleware Mods (Phase 2). Purpose-built proxy interceptors the proposer may
+// toggle/parameterize. The first is the tool-call path-normalizer — a
+// mechanical fix for the filename-garbling failure that prompt/param Mods can't
+// reach (see SELF_HARNESS.md §4, §11 P2).
+// ---------------------------------------------------------------------------
+export const MIDDLEWARE_IDS = ['toolcall-path-normalizer'];
+export function isMiddlewareId(s) {
+    return MIDDLEWARE_IDS.includes(s);
+}
+/**
+ * Validate a Mod against the DSL constraints. A proposer's output that fails
+ * this is rejected at parse time — it never reaches the apply/validate stages.
+ */
+export function validateMod(mod) {
+    switch (mod.kind) {
+        case 'env': {
+            if (!isKnownKnob(mod.key)) {
+                return { ok: false, reason: `unknown/non-allow-listed knob: ${mod.key}` };
+            }
+            const spec = knobSpec(mod.key);
+            if (spec.type === 'number') {
+                const n = Number(mod.to);
+                if (!Number.isFinite(n)) {
+                    return { ok: false, reason: `${mod.key}: "${mod.to}" is not numeric` };
+                }
+                if (spec.integer && !Number.isInteger(n)) {
+                    return { ok: false, reason: `${mod.key}: must be an integer, got ${mod.to}` };
+                }
+                if (n < spec.min || n > spec.max) {
+                    return {
+                        ok: false,
+                        reason: `${mod.key}=${mod.to} out of safe range [${spec.min}, ${spec.max}]`,
+                    };
+                }
+                return { ok: true };
+            }
+            // enum
+            if (!spec.values.includes(mod.to)) {
+                return { ok: false, reason: `${mod.key}="${mod.to}" not in {${spec.values.join(', ')}}` };
+            }
+            return { ok: true };
+        }
+        case 'scaffold': {
+            if (!UAP_COMPONENTS.includes(mod.component)) {
+                return { ok: false, reason: `unknown UAP component: ${mod.component}` };
+            }
+            if (mod.op !== 'replace' && mod.op !== 'append') {
+                return { ok: false, reason: `scaffold op must be replace|append` };
+            }
+            if (!mod.text.trim()) {
+                return { ok: false, reason: `scaffold text is empty` };
+            }
+            return { ok: true };
+        }
+        case 'middleware': {
+            if (!isMiddlewareId(mod.id)) {
+                return { ok: false, reason: `unknown middleware id: ${mod.id}` };
+            }
+            return { ok: true };
+        }
+        default:
+            return { ok: false, reason: `unknown mod kind` };
+    }
+}
+/** One-line, human-readable description of a Mod (for logs and profile diffs). */
+export function describeMod(mod) {
+    switch (mod.kind) {
+        case 'env':
+            return `env ${mod.key}: ${mod.from} -> ${mod.to}`;
+        case 'scaffold':
+            return `scaffold ${mod.component} (${mod.op}, ${mod.text.length} chars)`;
+        case 'middleware':
+            return `middleware ${mod.id}(${JSON.stringify(mod.params)})`;
+    }
+}
+/**
+ * Produce the inverse Mod that undoes `mod` (for rollback). Env/scaffold revert
+ * to their captured prior value; a middleware Mod reverts by disabling it.
+ */
+export function invertMod(mod) {
+    switch (mod.kind) {
+        case 'env':
+            return { kind: 'env', key: mod.key, from: mod.to, to: mod.from };
+        case 'scaffold':
+            // Only a captured prior text can be restored; an append has no clean
+            // inverse without the original, so callers must supply `from` to revert.
+            return {
+                kind: 'scaffold',
+                component: mod.component,
+                op: 'replace',
+                text: mod.from ?? '',
+                from: mod.op === 'replace' ? mod.text : undefined,
+            };
+        case 'middleware':
+            return { kind: 'middleware', id: mod.id, params: { ...mod.params, enabled: false } };
+    }
+}
+//# sourceMappingURL=mods.js.map

package/dist/self-harness/mods.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"mods.js","sourceRoot":"","sources":["../../src/self-harness/mods.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;GAWG;AAEH,OAAO,EAAgB,cAAc,EAAE,MAAM,+BAA+B,CAAC;AA+B7E,MAAM,CAAC,MAAM,cAAc,GAAG;IAC5B,mFAAmF;IACnF,eAAe,EAAE;QACf,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,KAAK,EAAE,OAAO,EAAE,IAAI;QACpE,WAAW,EAAE,0DAA0D;KACxE;IACD,oBAAoB,EAAE;QACpB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG;QACnD,WAAW,EAAE,6CAA6C;KAC3D;IACD,+DAA+D;IAC/D,yBAAyB,EAAE;QACzB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI;QAChE,WAAW,EAAE,sEAAsE;KACpF;IACD,iCAAiC,EAAE;QACjC,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,EAAE,EAAE,GAAG,EAAE,GAAG,EAAE,OAAO,EAAE,IAAI;QACjE,WAAW,EAAE,6DAA6D;KAC3E;IACD,yBAAyB,EAAE;QACzB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI;QAC/D,WAAW,EAAE,iEAAiE;KAC/E;IACD,sBAAsB,EAAE;QACtB,MAAM,EAAE,OAAO,EAAE,IAAI,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAC,EAAE,GAAG,EAAE,EAAE,EAAE,OAAO,EAAE,IAAI;QAC/D,WAAW,EAAE,2EAA2E;KACzF;CAC0C,CAAC;AAI9C,MAAM,UAAU,WAAW,CAAC,CAAS;IACnC,OAAO,MAAM,CAAC,SAAS,CAAC,cAAc,CAAC,IAAI,CAAC,cAAc,EAAE,CAAC,CAAC,CAAC;AACjE,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,QAAQ,CAAC,GAAc;IACrC,OAAO,cAAc,CAAC,GAAG,CAAC,CAAC;AAC7B,CAAC;AAED,8EAA8E;AAC9E,+EAA+E;AAC/E,sEAAsE;AACtE,gFAAgF;AAChF,0CAA0C;AAC1C,8EAA8E;AAE9E,MAAM,CAAC,MAAM,cAAc,GAAG,CAAC,0BAA0B,CAAU,CAAC;AAGpE,MAAM,UAAU,cAAc,CAAC,CAAS;IACtC,OAAQ,cAAoC,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAC;AAC3D,CAAC;AA0CD;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAC,GAAQ;IAClC,QAAQ,GAAG,CAAC,IAAI,EAAE,CAAC;QACjB,KAAK,KAAK,CAAC,CAAC,CAAC;YACX,IAAI,CAAC,WAAW,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,CAAC;gBAC1B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,kCAAkC,GAAG,CAAC,GAAG,EAAE,EAAE,CAAC;YAC5E,CAAC;YACD,MAAM,IAAI,GAAG,QAAQ,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;YAC/B,IAAI,IAAI,CAAC,IAAI,KAAK,QAAQ,EAAE,CAAC;gBAC3B,MAAM,CAAC,GAAG,MAAM,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;gBACzB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;oBACxB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,GAAG,CAAC,GAAG,MAAM,GAAG,CAAC,EAAE,kBAAkB,EAAE,CAAC;gBACzE,CAAC;gBACD,IAAI,IAAI,CAAC,OAAO,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC;oBACzC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,GAAG,CAAC,GAAG,6BAA6B,GAAG,CAAC,EAAE,EAAE,EAAE,CAAC;gBAChF,CAAC;gBACD,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;oBACjC,OAAO;wBACL,EAAE,EAAE,KAAK;wBACT,MAAM,EAAE,GAAG,GAAG,CAAC,GAAG,IAAI,GAAG,CAAC,EAAE,uBAAuB,IAAI,CAAC,GAAG,KAAK,IAAI,CAAC,GAAG,GAAG;qBAC5E,CAAC;gBACJ,CAAC;gBACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;YACtB,CAAC;YACD,OAAO;YACP,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAClC,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,GAAG,GAAG,CAAC,GAAG,KAAK,GAAG,CAAC,EAAE,aAAa,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG,EAAE,CAAC;YAC5F,CAAC;YACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;QACtB,CAAC;QACD,KAAK,UAAU,CAAC,CAAC,CAAC;YAChB,IAAI,CAAE,cAAoC,CAAC,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC;gBACnE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,0BAA0B,GAAG,CAAC,SAAS,EAAE,EAAE,CAAC;YAC1E,CAAC;YACD,IAAI,GAAG,CAAC,EAAE,KAAK,SAAS,IAAI,GAAG,CAAC,EAAE,KAAK,QAAQ,EAAE,CAAC;gBAChD,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,oCAAoC,EAAE,CAAC;YACrE,CAAC;YACD,IAAI,CAAC,GAAG,CAAC,IAAI,CAAC,IAAI,EAAE,EAAE,CAAC;gBACrB,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,wBAAwB,EAAE,CAAC;YACzD,CAAC;YACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;QACtB,CAAC;QACD,KAAK,YAAY,CAAC,CAAC,CAAC;YAClB,IAAI,CAAC,cAAc,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,CAAC;gBAC5B,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,0BAA2B,GAAqB,CAAC,EAAE,EAAE,EAAE,CAAC;YACtF,CAAC;YACD,OAAO,EAAE,EAAE,EAAE,IAAI,EAAE,CAAC;QACtB,CAAC;QACD;YACE,OAAO,EAAE,EAAE,EAAE,KAAK,EAAE,MAAM,EAAE,kBAAkB,EAAE,CAAC;IACrD,CAAC;AACH,CAAC;AAED,kFAAkF;AAClF,MAAM,UAAU,WAAW,CAAC,GAAQ;IAClC,QAAQ,GAAG,CAAC,IAAI,EAAE,CAAC;QACjB,KAAK,KAAK;YACR,OAAO,OAAO,GAAG,CAAC,GAAG,KAAK,GAAG,CAAC,IAAI,OAAO,GAAG,CAAC,EAAE,EAAE,CAAC;QACpD,KAAK,UAAU;YACb,OAAO,YAAY,GAAG,CAAC,SAAS,KAAK,GAAG,CAAC,EAAE,KAAK,GAAG,CAAC,IAAI,CAAC,MAAM,SAAS,CAAC;QAC3E,KAAK,YAAY;YACf,OAAO,cAAc,GAAG,CAAC,EAAE,IAAI,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,MAAM,CAAC,GAAG,CAAC;IACjE,CAAC;AACH,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,SAAS,CAAC,GAAQ;IAChC,QAAQ,GAAG,CAAC,IAAI,EAAE,CAAC;QACjB,KAAK,KAAK;YACR,OAAO,EAAE,IAAI,EAAE,KAAK,EAAE,GAAG,EAAE,GAAG,CAAC,GAAG,EAAE,IAAI,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,EAAE,GAAG,CAAC,IAAI,EAAE,CAAC;QACnE,KAAK,UAAU;YACb,qEAAqE;YACrE,yEAAyE;YACzE,OAAO;gBACL,IAAI,EAAE,UAAU;gBAChB,SAAS,EAAE,GAAG,CAAC,SAAS;gBACxB,EAAE,EAAE,SAAS;gBACb,IAAI,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE;gBACpB,IAAI,EAAE,GAAG,CAAC,EAAE,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC,SAAS;aAClD,CAAC;QACJ,KAAK,YAAY;YACf,OAAO,EAAE,IAAI,EAAE,YAAY,EAAE,EAAE,EAAE,GAAG,CAAC,EAAE,EAAE,MAAM,EAAE,EAAE,GAAG,GAAG,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,EAAE,CAAC;IACzF,CAAC;AACH,CAAC"}

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

@@ -0,0 +1,61 @@
+/**
+ * Self-Harness — the P1 orchestrator: one autonomous mine -> propose -> validate
+ * -> decide iteration.
+ *
+ * The `validate` dependency is INJECTED so the loop is testable without the slow
+ * paired bench: tests pass a stub; the CLI wires a real validator that runs
+ * `runPaired` + `analyze` on the validation and held-out suites (and, for env
+ * Mods, physically applies the env edit + restarts the server around the run).
+ * Each candidate is validated in isolation; rejected Mods are reverted, accepted
+ * Mods are recorded into the profile and logged. Budget-bounded.
+ *
+ * See docs/design/SELF_HARNESS.md §3, §7, §11 (P1).
+ */
+import type { RunRecord } from '../benchmarks/paired/types.js';
+import type { Comparison } from '../benchmarks/paired/report.js';
+import { Mod } from './mods.js';
+import { WeaknessReport } from './weakness.js';
+import { Proposer } from './propose.js';
+import { Decision, DecisionOptions } from './decide.js';
+import { HarnessProfile } from './profile.js';
+import { TransferStore } from './transfer.js';
+export interface ValidationOutcome {
+    validation: Comparison;
+    /** Held-out comparison; null when the held-out suite was skipped. */
+    heldout: Comparison | null;
+}
+export type Validator = (mod: Mod) => Promise<ValidationOutcome>;
+export interface OrchestratorOptions {
+    model: string;
+    /** Records mined for weaknesses (e.g. a prior baseline bench). */
+    records: RunRecord[];
+    profile: HarnessProfile;
+    validate: Validator;
+    proposer?: Proposer;
+    /** Max candidate Mods to validate this iteration (token/time budget). Default 3. */
+    maxCandidates?: number;
+    decision?: DecisionOptions;
+    minFrequency?: number;
+    log?: (msg: string) => void;
+    /** Optional cross-model transfer store; outcomes are recorded into it (P3). */
+    transferStore?: TransferStore;
+    /** ISO timestamp + provenance string stamped on recorded transfer entries. */
+    validatedAt?: string;
+    provenance?: string;
+}
+export interface CandidateOutcome {
+    mod: Mod;
+    decision: Decision;
+    accepted: boolean;
+}
+export interface IterationResult {
+    weaknesses: WeaknessReport[];
+    proposed: Mod[];
+    outcomes: CandidateOutcome[];
+    accepted: Mod[];
+    /** Profile after applying all accepted Mods. */
+    profile: HarnessProfile;
+}
+/** Run one self-harness iteration. */
+export declare function runIteration(opts: OrchestratorOptions): Promise<IterationResult>;
+//# sourceMappingURL=orchestrator.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"orchestrator.d.ts","sourceRoot":"","sources":["../../src/self-harness/orchestrator.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAEH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,+BAA+B,CAAC;AAC/D,OAAO,KAAK,EAAE,UAAU,EAAE,MAAM,gCAAgC,CAAC;AACjE,OAAO,EAAE,GAAG,EAAe,MAAM,WAAW,CAAC;AAC7C,OAAO,EAAE,cAAc,EAAE,MAAM,eAAe,CAAC;AAE/C,OAAO,EAAE,QAAQ,EAAqB,MAAM,cAAc,CAAC;AAC3D,OAAO,EAAgB,QAAQ,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AACtE,OAAO,EAAE,cAAc,EAAE,MAAM,cAAc,CAAC;AAC9C,OAAO,EAAE,aAAa,EAAqB,MAAM,eAAe,CAAC;AAEjE,MAAM,WAAW,iBAAiB;IAChC,UAAU,EAAE,UAAU,CAAC;IACvB,qEAAqE;IACrE,OAAO,EAAE,UAAU,GAAG,IAAI,CAAC;CAC5B;AAED,MAAM,MAAM,SAAS,GAAG,CAAC,GAAG,EAAE,GAAG,KAAK,OAAO,CAAC,iBAAiB,CAAC,CAAC;AAEjE,MAAM,WAAW,mBAAmB;IAClC,KAAK,EAAE,MAAM,CAAC;IACd,kEAAkE;IAClE,OAAO,EAAE,SAAS,EAAE,CAAC;IACrB,OAAO,EAAE,cAAc,CAAC;IACxB,QAAQ,EAAE,SAAS,CAAC;IACpB,QAAQ,CAAC,EAAE,QAAQ,CAAC;IACpB,oFAAoF;IACpF,aAAa,CAAC,EAAE,MAAM,CAAC;IACvB,QAAQ,CAAC,EAAE,eAAe,CAAC;IAC3B,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAC;IAC5B,+EAA+E;IAC/E,aAAa,CAAC,EAAE,aAAa,CAAC;IAC9B,8EAA8E;IAC9E,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,UAAU,CAAC,EAAE,MAAM,CAAC;CACrB;AAED,MAAM,WAAW,gBAAgB;IAC/B,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,EAAE,QAAQ,CAAC;IACnB,QAAQ,EAAE,OAAO,CAAC;CACnB;AAED,MAAM,WAAW,eAAe;IAC9B,UAAU,EAAE,cAAc,EAAE,CAAC;IAC7B,QAAQ,EAAE,GAAG,EAAE,CAAC;IAChB,QAAQ,EAAE,gBAAgB,EAAE,CAAC;IAC7B,QAAQ,EAAE,GAAG,EAAE,CAAC;IAChB,gDAAgD;IAChD,OAAO,EAAE,cAAc,CAAC;CACzB;AAED,sCAAsC;AACtC,wBAAsB,YAAY,CAAC,IAAI,EAAE,mBAAmB,GAAG,OAAO,CAAC,eAAe,CAAC,CAiEtF"}

package/dist/self-harness/orchestrator.js

@@ -0,0 +1,110 @@
+/**
+ * Self-Harness — the P1 orchestrator: one autonomous mine -> propose -> validate
+ * -> decide iteration.
+ *
+ * The `validate` dependency is INJECTED so the loop is testable without the slow
+ * paired bench: tests pass a stub; the CLI wires a real validator that runs
+ * `runPaired` + `analyze` on the validation and held-out suites (and, for env
+ * Mods, physically applies the env edit + restarts the server around the run).
+ * Each candidate is validated in isolation; rejected Mods are reverted, accepted
+ * Mods are recorded into the profile and logged. Budget-bounded.
+ *
+ * See docs/design/SELF_HARNESS.md §3, §7, §11 (P1).
+ */
+import { describeMod } from './mods.js';
+import { mineFromRecords } from './mine.js';
+import { heuristicProposer } from './propose.js';
+import { decideAccept } from './decide.js';
+import { modKey, makeEntry } from './transfer.js';
+/** Run one self-harness iteration. */
+export async function runIteration(opts) {
+    const log = opts.log ?? (() => { });
+    const proposer = opts.proposer ?? heuristicProposer;
+    const maxC = opts.maxCandidates ?? 3;
+    // Stage 1 — mine
+    const weaknesses = mineFromRecords(opts.records, {
+        model: opts.model,
+        minFrequency: opts.minFrequency,
+    });
+    log(`mined ${weaknesses.length} weakness(es): ${weaknesses.map((w) => w.kind).join(', ') || '(none)'}`);
+    // Stage 2 — propose (bounded, minimal)
+    const proposed = proposer.propose(weaknesses, opts.profile).slice(0, maxC);
+    log(`proposed ${proposed.length} candidate Mod(s) via ${proposer.id}`);
+    // Stage 3+4 — validate + decide, each candidate in isolation
+    const outcomes = [];
+    const accepted = [];
+    let profile = opts.profile;
+    for (const mod of proposed) {
+        log(`validating: ${describeMod(mod)}`);
+        let decision;
+        try {
+            const { validation, heldout } = await opts.validate(mod);
+            decision = decideAccept(validation, heldout, opts.decision);
+        }
+        catch (e) {
+            decision = {
+                verdict: 'reject',
+                reason: `validation error: ${e instanceof Error ? e.message : String(e)}`,
+                validationDelta: NaN,
+                heldoutDelta: null,
+            };
+        }
+        const ok = decision.verdict === 'accept';
+        log(`  -> ${decision.verdict.toUpperCase()}: ${decision.reason}`);
+        outcomes.push({ mod, decision, accepted: ok });
+        if (ok) {
+            accepted.push(mod);
+            profile = applyAcceptedToProfile(profile, mod);
+        }
+        // Record the outcome (accept OR reject) into the transfer store, attributed
+        // to the weakness whose heuristic produced this Mod, so the fix transfers to
+        // future models and known-bad Mods aren't re-proposed.
+        if (opts.transferStore) {
+            const w = attributeWeakness(mod, weaknesses, proposer, profile);
+            if (w) {
+                opts.transferStore.record(makeEntry({
+                    signature: w.signature,
+                    kind: w.kind,
+                    model: opts.model,
+                    mod,
+                    delta: Number.isFinite(decision.validationDelta) ? decision.validationDelta : 0,
+                    accepted: ok,
+                    validatedAt: opts.validatedAt ?? '',
+                    provenance: opts.provenance ?? 'self-harness iteration',
+                }));
+            }
+        }
+    }
+    return { weaknesses, proposed, outcomes, accepted, profile };
+}
+/**
+ * Attribute a Mod back to the weakness that produced it, by re-running the
+ * proposer on each single weakness and matching by structural key. Works for the
+ * heuristic and transfer proposers (deterministic kind -> Mod mapping).
+ */
+function attributeWeakness(mod, weaknesses, proposer, profile) {
+    const target = modKey(mod);
+    for (const w of weaknesses) {
+        if (proposer.propose([w], profile).some((m) => modKey(m) === target))
+            return w;
+    }
+    return null;
+}
+/** Fold an accepted Mod into the in-memory profile (env/scaffold/middleware). */
+function applyAcceptedToProfile(profile, mod) {
+    switch (mod.kind) {
+        case 'env':
+            return { ...profile, env: { ...profile.env, [mod.key]: mod.to } };
+        case 'scaffold': {
+            const prev = profile.scaffold[mod.component] ?? '';
+            const text = mod.op === 'append' ? `${prev}\n${mod.text}`.trim() : mod.text;
+            return { ...profile, scaffold: { ...profile.scaffold, [mod.component]: text } };
+        }
+        case 'middleware':
+            return {
+                ...profile,
+                middleware: { ...profile.middleware, [mod.id]: { ...mod.params } },
+            };
+    }
+}
+//# sourceMappingURL=orchestrator.js.map

package/dist/self-harness/orchestrator.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"orchestrator.js","sourceRoot":"","sources":["../../src/self-harness/orchestrator.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;GAYG;AAIH,OAAO,EAAO,WAAW,EAAE,MAAM,WAAW,CAAC;AAE7C,OAAO,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAC5C,OAAO,EAAY,iBAAiB,EAAE,MAAM,cAAc,CAAC;AAC3D,OAAO,EAAE,YAAY,EAA6B,MAAM,aAAa,CAAC;AAEtE,OAAO,EAAiB,MAAM,EAAE,SAAS,EAAE,MAAM,eAAe,CAAC;AA4CjE,sCAAsC;AACtC,MAAM,CAAC,KAAK,UAAU,YAAY,CAAC,IAAyB;IAC1D,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,IAAI,CAAC,GAAG,EAAE,GAAE,CAAC,CAAC,CAAC;IACnC,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;IACpD,MAAM,IAAI,GAAG,IAAI,CAAC,aAAa,IAAI,CAAC,CAAC;IAErC,iBAAiB;IACjB,MAAM,UAAU,GAAG,eAAe,CAAC,IAAI,CAAC,OAAO,EAAE;QAC/C,KAAK,EAAE,IAAI,CAAC,KAAK;QACjB,YAAY,EAAE,IAAI,CAAC,YAAY;KAChC,CAAC,CAAC;IACH,GAAG,CAAC,SAAS,UAAU,CAAC,MAAM,kBAAkB,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,QAAQ,EAAE,CAAC,CAAC;IAExG,uCAAuC;IACvC,MAAM,QAAQ,GAAG,QAAQ,CAAC,OAAO,CAAC,UAAU,EAAE,IAAI,CAAC,OAAO,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC;IAC3E,GAAG,CAAC,YAAY,QAAQ,CAAC,MAAM,yBAAyB,QAAQ,CAAC,EAAE,EAAE,CAAC,CAAC;IAEvE,6DAA6D;IAC7D,MAAM,QAAQ,GAAuB,EAAE,CAAC;IACxC,MAAM,QAAQ,GAAU,EAAE,CAAC;IAC3B,IAAI,OAAO,GAAG,IAAI,CAAC,OAAO,CAAC;IAE3B,KAAK,MAAM,GAAG,IAAI,QAAQ,EAAE,CAAC;QAC3B,GAAG,CAAC,eAAe,WAAW,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC;QACvC,IAAI,QAAkB,CAAC;QACvB,IAAI,CAAC;YACH,MAAM,EAAE,UAAU,EAAE,OAAO,EAAE,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC;YACzD,QAAQ,GAAG,YAAY,CAAC,UAAU,EAAE,OAAO,EAAE,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9D,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,QAAQ,GAAG;gBACT,OAAO,EAAE,QAAQ;gBACjB,MAAM,EAAE,qBAAqB,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE;gBACzE,eAAe,EAAE,GAAG;gBACpB,YAAY,EAAE,IAAI;aACnB,CAAC;QACJ,CAAC;QACD,MAAM,EAAE,GAAG,QAAQ,CAAC,OAAO,KAAK,QAAQ,CAAC;QACzC,GAAG,CAAC,QAAQ,QAAQ,CAAC,OAAO,CAAC,WAAW,EAAE,KAAK,QAAQ,CAAC,MAAM,EAAE,CAAC,CAAC;QAClE,QAAQ,CAAC,IAAI,CAAC,EAAE,GAAG,EAAE,QAAQ,EAAE,QAAQ,EAAE,EAAE,EAAE,CAAC,CAAC;QAC/C,IAAI,EAAE,EAAE,CAAC;YACP,QAAQ,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;YACnB,OAAO,GAAG,sBAAsB,CAAC,OAAO,EAAE,GAAG,CAAC,CAAC;QACjD,CAAC;QACD,4EAA4E;QAC5E,6EAA6E;QAC7E,uDAAuD;QACvD,IAAI,IAAI,CAAC,aAAa,EAAE,CAAC;YACvB,MAAM,CAAC,GAAG,iBAAiB,CAAC,GAAG,EAAE,UAAU,EAAE,QAAQ,EAAE,OAAO,CAAC,CAAC;YAChE,IAAI,CAAC,EAAE,CAAC;gBACN,IAAI,CAAC,aAAa,CAAC,MAAM,CACvB,SAAS,CAAC;oBACR,SAAS,EAAE,CAAC,CAAC,SAAS;oBACtB,IAAI,EAAE,CAAC,CAAC,IAAI;oBACZ,KAAK,EAAE,IAAI,CAAC,KAAK;oBACjB,GAAG;oBACH,KAAK,EAAE,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,QAAQ,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC;oBAC/E,QAAQ,EAAE,EAAE;oBACZ,WAAW,EAAE,IAAI,CAAC,WAAW,IAAI,EAAE;oBACnC,UAAU,EAAE,IAAI,CAAC,UAAU,IAAI,wBAAwB;iBACxD,CAAC,CACH,CAAC;YACJ,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,EAAE,UAAU,EAAE,QAAQ,EAAE,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,CAAC;AAC/D,CAAC;AAED;;;;GAIG;AACH,SAAS,iBAAiB,CACxB,GAAQ,EACR,UAA4B,EAC5B,QAAkB,EAClB,OAAuB;IAEvB,MAAM,MAAM,GAAG,MAAM,CAAC,GAAG,CAAC,CAAC;IAC3B,KAAK,MAAM,CAAC,IAAI,UAAU,EAAE,CAAC;QAC3B,IAAI,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC;YAAE,OAAO,CAAC,CAAC;IACjF,CAAC;IACD,OAAO,IAAI,CAAC;AACd,CAAC;AAED,iFAAiF;AACjF,SAAS,sBAAsB,CAAC,OAAuB,EAAE,GAAQ;IAC/D,QAAQ,GAAG,CAAC,IAAI,EAAE,CAAC;QACjB,KAAK,KAAK;YACR,OAAO,EAAE,GAAG,OAAO,EAAE,GAAG,EAAE,EAAE,GAAG,OAAO,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,GAAG,CAAC,EAAE,GAAG,CAAC,EAAE,EAAE,EAAE,CAAC;QACpE,KAAK,UAAU,CAAC,CAAC,CAAC;YAChB,MAAM,IAAI,GAAG,OAAO,CAAC,QAAQ,CAAC,GAAG,CAAC,SAAS,CAAC,IAAI,EAAE,CAAC;YACnD,MAAM,IAAI,GAAG,GAAG,CAAC,EAAE,KAAK,QAAQ,CAAC,CAAC,CAAC,GAAG,IAAI,KAAK,GAAG,CAAC,IAAI,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,IAAI,CAAC;YAC5E,OAAO,EAAE,GAAG,OAAO,EAAE,QAAQ,EAAE,EAAE,GAAG,OAAO,CAAC,QAAQ,EAAE,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,IAAI,EAAE,EAAE,CAAC;QAClF,CAAC;QACD,KAAK,YAAY;YACf,OAAO;gBACL,GAAG,OAAO;gBACV,UAAU,EAAE,EAAE,GAAG,OAAO,CAAC,UAAU,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,GAAG,GAAG,CAAC,MAAM,EAAE,EAAE;aACnE,CAAC;IACN,CAAC;AACH,CAAC"}

package/dist/self-harness/pending.d.ts

@@ -0,0 +1,57 @@
+/**
+ * Self-Harness — gated promotion queue (P3).
+ *
+ * Proposals mined from PRODUCTION traces are never auto-applied: they are
+ * enqueued here as `PendingProposal`s awaiting validation + a human gate. The
+ * promotion policy decides what may auto-promote after validation (low-risk env
+ * knobs) vs what always needs human review (scaffold/middleware). docs/design/
+ * SELF_HARNESS.md §9, §10.
+ */
+import { Mod } from './mods.js';
+import { FailureKind } from './weakness.js';
+export type PromotionGate = 'auto-after-validation' | 'human';
+/**
+ * Promotion policy: env-knob Mods are low-risk and may auto-promote after they
+ * pass validation; scaffold + middleware Mods change behavior more broadly and
+ * always require a human gate. This is the safety boundary for online learning.
+ */
+export declare function promotionGate(mod: Mod): PromotionGate;
+export interface PendingProposal {
+    id: string;
+    signature: string;
+    kind: FailureKind;
+    model: string;
+    mod: Mod;
+    gate: PromotionGate;
+    /** Where it came from: 'halo' | 'proxy-log' | transfer source. */
+    source: string;
+    frequency: number;
+    createdAt: string;
+    status: 'pending' | 'validated' | 'rejected' | 'promoted';
+    /** Set once validated. */
+    validationDelta?: number;
+    note?: string;
+}
+export declare class PendingQueue {
+    private readonly path;
+    private items;
+    constructor(path: string);
+    private persist;
+    /** Enqueue a proposal, de-duped by (model, signature, mod-shape) while pending. */
+    enqueue(p: Omit<PendingProposal, 'gate' | 'status' | 'createdAt'> & {
+        createdAt: string;
+    }): PendingProposal;
+    list(status?: PendingProposal['status']): PendingProposal[];
+    update(id: string, patch: Partial<PendingProposal>): void;
+    /**
+     * Drop terminal (promoted/rejected) entries older than `maxAgeDays`, and stale
+     * still-pending entries older than `maxPendingDays` (never validated → likely
+     * obsolete). Returns the removed items.
+     */
+    prune(opts?: {
+        maxAgeDays?: number;
+        maxPendingDays?: number;
+        now?: number;
+    }): PendingProposal[];
+}
+//# sourceMappingURL=pending.d.ts.map

package/dist/self-harness/pending.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"pending.d.ts","sourceRoot":"","sources":["../../src/self-harness/pending.ts"],"names":[],"mappings":"AAAA;;;;;;;;GAQG;AAIH,OAAO,EAAE,GAAG,EAAE,MAAM,WAAW,CAAC;AAChC,OAAO,EAAE,WAAW,EAAE,MAAM,eAAe,CAAC;AAE5C,MAAM,MAAM,aAAa,GAAG,uBAAuB,GAAG,OAAO,CAAC;AAE9D;;;;GAIG;AACH,wBAAgB,aAAa,CAAC,GAAG,EAAE,GAAG,GAAG,aAAa,CAErD;AAED,MAAM,WAAW,eAAe;IAC9B,EAAE,EAAE,MAAM,CAAC;IACX,SAAS,EAAE,MAAM,CAAC;IAClB,IAAI,EAAE,WAAW,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,GAAG,EAAE,GAAG,CAAC;IACT,IAAI,EAAE,aAAa,CAAC;IACpB,kEAAkE;IAClE,MAAM,EAAE,MAAM,CAAC;IACf,SAAS,EAAE,MAAM,CAAC;IAClB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,EAAE,SAAS,GAAG,WAAW,GAAG,UAAU,GAAG,UAAU,CAAC;IAC1D,0BAA0B;IAC1B,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,IAAI,CAAC,EAAE,MAAM,CAAC;CACf;AAED,qBAAa,YAAY;IAGX,OAAO,CAAC,QAAQ,CAAC,IAAI;IAFjC,OAAO,CAAC,KAAK,CAAyB;gBAET,IAAI,EAAE,MAAM;IAUzC,OAAO,CAAC,OAAO;IAKf,mFAAmF;IACnF,OAAO,CAAC,CAAC,EAAE,IAAI,CAAC,eAAe,EAAE,MAAM,GAAG,QAAQ,GAAG,WAAW,CAAC,GAAG;QAAE,SAAS,EAAE,MAAM,CAAA;KAAE,GAAG,eAAe;IAgB3G,IAAI,CAAC,MAAM,CAAC,EAAE,eAAe,CAAC,QAAQ,CAAC,GAAG,eAAe,EAAE;IAI3D,MAAM,CAAC,EAAE,EAAE,MAAM,EAAE,KAAK,EAAE,OAAO,CAAC,eAAe,CAAC,GAAG,IAAI;IAQzD;;;;OAIG;IACH,KAAK,CAAC,IAAI,GAAE;QAAE,UAAU,CAAC,EAAE,MAAM,CAAC;QAAC,cAAc,CAAC,EAAE,MAAM,CAAC;QAAC,GAAG,CAAC,EAAE,MAAM,CAAA;KAAO,GAAG,eAAe,EAAE;CAkBpG"}