@miller-tech/uap 1.175.13 → 1.178.0

package/dist/self-harness/evidence-proposer.js

@@ -0,0 +1,132 @@
+/**
+ * Evidence-driven proposer — the D -> B connection (harness plan, 2026-07-31).
+ *
+ * The plan's load-bearing sequence is D (evidence corpus) THEN B (tool search
+ * space): record what the tool layer actually does, then mutate the knobs the
+ * evidence implicates. Shipping both halves without this module would leave the
+ * corpus written but unread and `ToolMod` defined but never emitted — a search
+ * space with nothing searching it.
+ *
+ * Unlike `heuristicProposer`, which maps a run-level FailureKind to an
+ * inference-server knob, this reads per-tool-call outcomes. That is the level
+ * the tool layer fails at: an `edit-miss` never surfaces as a run-level failure
+ * kind, it just quietly costs a turn, which is exactly why the loop could not
+ * see the surface the literature says matters most (tools +3.3pp, arXiv
+ * 2604.25850).
+ *
+ * Every proposal is a HYPOTHESIS with a measurable prediction, not a fix. The
+ * paired validator decides, and the change manifest falsifies it next round.
+ */
+import { TOOL_KNOB_ALLOWLIST, validateMod } from './mods.js';
+import { summarizeToolCalls } from '../telemetry/tool-calls.js';
+/** Read a tool knob's in-effect value: profile first, then env, then default. */
+function currentToolKnob(profile, key, fallback) {
+    const fromProfile = (profile.tool ?? {})[key];
+    if (fromProfile != null)
+        return fromProfile;
+    return process.env[key] ?? fallback;
+}
+function toolMod(profile, key, to) {
+    const spec = TOOL_KNOB_ALLOWLIST[key];
+    const from = currentToolKnob(profile, key, spec.type === 'number' ? '' : '1');
+    if (from === to)
+        return null;
+    const mod = { kind: 'tool', key, from, to };
+    return validateMod(mod).ok ? mod : null;
+}
+/**
+ * Propose tool-layer Mods from the evidence corpus.
+ *
+ * `summary` is injected so this is testable without a database, and so a caller
+ * can scope it to one run.
+ */
+export function proposeFromEvidence(summary, profile, thresholds = {}) {
+    const minAttempts = thresholds.minEditAttempts ?? 10;
+    const successFloor = thresholds.editSuccessFloor ?? 0.8;
+    const tolerantCeiling = thresholds.tolerantShareCeiling ?? 0.3;
+    const out = [];
+    const e = summary.editHealth;
+    if (e.attempts < minAttempts)
+        return out; // not enough evidence to act on
+    const tolerantShare = e.tolerant / e.attempts;
+    // Signal 1 — anchors miss outright. The model's picture of the file is stale,
+    // which a bigger read window may fix and tolerance certainly does not.
+    if (e.successRate < successFloor && e.misses > 0) {
+        const cur = Number(currentToolKnob(profile, 'UAP_READ_WINDOW_BYTES', '8000'));
+        if (Number.isFinite(cur)) {
+            const next = Math.min(32_000, Math.round(cur * 2));
+            const mod = toolMod(profile, 'UAP_READ_WINDOW_BYTES', String(next));
+            if (mod)
+                out.push(mod);
+        }
+        // Diagnostics OFF is the counter-hypothesis worth testing when misses
+        // dominate: the nearest-region report costs context on every miss, and if it
+        // is not converting misses into successes it is pure overhead.
+        if (e.misses > e.attempts * 0.4) {
+            const mod = toolMod(profile, 'UAP_EDIT_DIAGNOSTICS', '0');
+            if (mod)
+                out.push(mod);
+        }
+    }
+    // Signal 2 — a large share of edits land ONLY through tolerance. That is a
+    // success today, but it means the model is routinely wrong about the file, so
+    // it is worth measuring whether tolerance is rescuing real edits or masking
+    // drift that produces subtly wrong ones.
+    if (tolerantShare > tolerantCeiling) {
+        const mod = toolMod(profile, 'UAP_EDIT_TOLERANT', '0');
+        if (mod)
+            out.push(mod);
+    }
+    // Signal 3 — the loop is running out of turns rather than failing on tools.
+    const roundsPressure = summary.topFailures.some((f) => f.outcome === 'timeout');
+    if (roundsPressure) {
+        const cur = Number(currentToolKnob(profile, 'UAP_MAX_TOOL_ROUNDS', '12'));
+        if (Number.isFinite(cur)) {
+            const mod = toolMod(profile, 'UAP_MAX_TOOL_ROUNDS', String(Math.min(40, Math.round(cur * 1.5))));
+            if (mod)
+                out.push(mod);
+        }
+    }
+    return out;
+}
+/**
+ * A `Proposer` that reads the live corpus.
+ *
+ * It ignores `weaknesses` by design: those are run-level failure kinds mined
+ * from bench records, and this proposer's whole point is the signal that never
+ * reaches that level.
+ */
+export function evidenceProposer(opts = {}) {
+    return {
+        id: 'evidence',
+        propose(_weaknesses, profile) {
+            const summary = opts.summary ?? summarizeToolCalls(opts.runId);
+            return proposeFromEvidence(summary, profile, opts.thresholds);
+        },
+    };
+}
+/**
+ * Run several proposers and concatenate, de-duplicated by structural identity.
+ * Lets the loop search the tool layer AND the server layer in one iteration
+ * without either proposer knowing about the other.
+ */
+export function combineProposers(id, proposers) {
+    return {
+        id,
+        propose(weaknesses, profile) {
+            const seen = new Set();
+            const out = [];
+            for (const p of proposers) {
+                for (const mod of p.propose(weaknesses, profile)) {
+                    const key = JSON.stringify(mod);
+                    if (seen.has(key))
+                        continue;
+                    seen.add(key);
+                    out.push(mod);
+                }
+            }
+            return out;
+        },
+    };
+}
+//# sourceMappingURL=evidence-proposer.js.map

package/dist/self-harness/evidence-proposer.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"evidence-proposer.js","sourceRoot":"","sources":["../../src/self-harness/evidence-proposer.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;GAkBG;AAEH,OAAO,EAAO,mBAAmB,EAAE,WAAW,EAAsB,MAAM,WAAW,CAAC;AAItF,OAAO,EAAE,kBAAkB,EAAwB,MAAM,4BAA4B,CAAC;AAEtF,iFAAiF;AACjF,SAAS,eAAe,CAAC,OAAuB,EAAE,GAAkB,EAAE,QAAgB;IACpF,MAAM,WAAW,GAAG,CAAC,OAAO,CAAC,IAAI,IAAI,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9C,IAAI,WAAW,IAAI,IAAI;QAAE,OAAO,WAAW,CAAC;IAC5C,OAAO,OAAO,CAAC,GAAG,CAAC,GAAG,CAAC,IAAI,QAAQ,CAAC;AACtC,CAAC;AAED,SAAS,OAAO,CAAC,OAAuB,EAAE,GAAkB,EAAE,EAAU;IACtE,MAAM,IAAI,GAAG,mBAAmB,CAAC,GAAG,CAAC,CAAC;IACtC,MAAM,IAAI,GAAG,eAAe,CAAC,OAAO,EAAE,GAAG,EAAE,IAAI,CAAC,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC;IAC9E,IAAI,IAAI,KAAK,EAAE;QAAE,OAAO,IAAI,CAAC;IAC7B,MAAM,GAAG,GAAQ,EAAE,IAAI,EAAE,MAAM,EAAE,GAAG,EAAE,IAAI,EAAE,EAAE,EAAE,CAAC;IACjD,OAAO,WAAW,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC;AAC1C,CAAC;AAYD;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CACjC,OAAwB,EACxB,OAAuB,EACvB,aAAiC,EAAE;IAEnC,MAAM,WAAW,GAAG,UAAU,CAAC,eAAe,IAAI,EAAE,CAAC;IACrD,MAAM,YAAY,GAAG,UAAU,CAAC,gBAAgB,IAAI,GAAG,CAAC;IACxD,MAAM,eAAe,GAAG,UAAU,CAAC,oBAAoB,IAAI,GAAG,CAAC;IAE/D,MAAM,GAAG,GAAU,EAAE,CAAC;IACtB,MAAM,CAAC,GAAG,OAAO,CAAC,UAAU,CAAC;IAC7B,IAAI,CAAC,CAAC,QAAQ,GAAG,WAAW;QAAE,OAAO,GAAG,CAAC,CAAC,gCAAgC;IAE1E,MAAM,aAAa,GAAG,CAAC,CAAC,QAAQ,GAAG,CAAC,CAAC,QAAQ,CAAC;IAE9C,8EAA8E;IAC9E,uEAAuE;IACvE,IAAI,CAAC,CAAC,WAAW,GAAG,YAAY,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;QACjD,MAAM,GAAG,GAAG,MAAM,CAAC,eAAe,CAAC,OAAO,EAAE,uBAAuB,EAAE,MAAM,CAAC,CAAC,CAAC;QAC9E,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,MAAM,IAAI,GAAG,IAAI,CAAC,GAAG,CAAC,MAAM,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;YACnD,MAAM,GAAG,GAAG,OAAO,CAAC,OAAO,EAAE,uBAAuB,EAAE,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC;YACpE,IAAI,GAAG;gBAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACzB,CAAC;QACD,sEAAsE;QACtE,6EAA6E;QAC7E,+DAA+D;QAC/D,IAAI,CAAC,CAAC,MAAM,GAAG,CAAC,CAAC,QAAQ,GAAG,GAAG,EAAE,CAAC;YAChC,MAAM,GAAG,GAAG,OAAO,CAAC,OAAO,EAAE,sBAAsB,EAAE,GAAG,CAAC,CAAC;YAC1D,IAAI,GAAG;gBAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACzB,CAAC;IACH,CAAC;IAED,2EAA2E;IAC3E,8EAA8E;IAC9E,4EAA4E;IAC5E,yCAAyC;IACzC,IAAI,aAAa,GAAG,eAAe,EAAE,CAAC;QACpC,MAAM,GAAG,GAAG,OAAO,CAAC,OAAO,EAAE,mBAAmB,EAAE,GAAG,CAAC,CAAC;QACvD,IAAI,GAAG;YAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IACzB,CAAC;IAED,4EAA4E;IAC5E,MAAM,cAAc,GAAG,OAAO,CAAC,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,KAAK,SAAS,CAAC,CAAC;IAChF,IAAI,cAAc,EAAE,CAAC;QACnB,MAAM,GAAG,GAAG,MAAM,CAAC,eAAe,CAAC,OAAO,EAAE,qBAAqB,EAAE,IAAI,CAAC,CAAC,CAAC;QAC1E,IAAI,MAAM,CAAC,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAC;YACzB,MAAM,GAAG,GAAG,OAAO,CAAC,OAAO,EAAE,qBAAqB,EAAE,MAAM,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,KAAK,CAAC,GAAG,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;YACjG,IAAI,GAAG;gBAAE,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;QACzB,CAAC;IACH,CAAC;IAED,OAAO,GAAG,CAAC;AACb,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAuF,EAAE;IAEzF,OAAO;QACL,EAAE,EAAE,UAAU;QACd,OAAO,CAAC,WAA6B,EAAE,OAAuB;YAC5D,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,IAAI,kBAAkB,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC;YAC/D,OAAO,mBAAmB,CAAC,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,UAAU,CAAC,CAAC;QAChE,CAAC;KACF,CAAC;AACJ,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,gBAAgB,CAAC,EAAU,EAAE,SAAqB;IAChE,OAAO;QACL,EAAE;QACF,OAAO,CAAC,UAAU,EAAE,OAAO;YACzB,MAAM,IAAI,GAAG,IAAI,GAAG,EAAU,CAAC;YAC/B,MAAM,GAAG,GAAU,EAAE,CAAC;YACtB,KAAK,MAAM,CAAC,IAAI,SAAS,EAAE,CAAC;gBAC1B,KAAK,MAAM,GAAG,IAAI,CAAC,CAAC,OAAO,CAAC,UAAU,EAAE,OAAO,CAAC,EAAE,CAAC;oBACjD,MAAM,GAAG,GAAG,IAAI,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC;oBAChC,IAAI,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC;wBAAE,SAAS;oBAC5B,IAAI,CAAC,GAAG,CAAC,GAAG,CAAC,CAAC;oBACd,GAAG,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;gBAChB,CAAC;YACH,CAAC;YACD,OAAO,GAAG,CAAC;QACb,CAAC;KACF,CAAC;AACJ,CAAC"}

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

@@ -0,0 +1,130 @@
+/**
+ * Change manifests — every harness edit is a falsifiable contract
+ * (harness plan area C, 2026-07-31).
+ *
+ * Agentic Harness Engineering (arXiv 2604.25850) attributes a large part of its
+ * Terminal-Bench 2 gain (69.7% -> 77.0%) not to smarter proposals but to
+ * DECISION OBSERVABILITY: every edit is paired with a self-declared prediction
+ * about which tasks it will fix and which it might break, that prediction is
+ * checked against the next round's actual per-task outcomes, and edits whose
+ * predictions do not materialise are reverted at file granularity.
+ *
+ * Without this, an accepted Mod is accepted forever on the strength of one
+ * noisy A/B. With it, acceptance is provisional and the loop can un-learn.
+ *
+ * Everything here is pure except the JSON store at the bottom, so the
+ * attribution rules are unit-testable without a bench run.
+ */
+import type { RunRecord } from '../benchmarks/paired/types.js';
+import { Mod } from './mods.js';
+/** A prediction recorded at the moment a Mod is applied. */
+export interface ChangeManifest {
+    /** Stable id — `modKey`-derived by the caller, or any unique string. */
+    id: string;
+    mod: Mod;
+    /** ISO timestamp. Injected, never `Date.now()`, so runs stay reproducible. */
+    createdAt: string;
+    /** Task ids this edit claims it will turn from failing to passing. */
+    predictedFixes: string[];
+    /** Task ids this edit admits it might break. Declaring them is not a free pass. */
+    predictedRisks: string[];
+    /**
+     * Files/knobs the edit touched, for a file-granular revert. For an env Mod
+     * this is the env file; for scaffold/middleware, the component id.
+     */
+    touched: string[];
+    /** Human-readable summary, for logs and the pending queue. */
+    summary: string;
+}
+export interface ManifestAttribution {
+    manifestId: string;
+    /** Predicted fixes that actually flipped fail -> pass. */
+    fixesRealised: string[];
+    /** Predicted fixes that did not flip. The falsification. */
+    fixesMissed: string[];
+    /** Predicted risks that did break (declared, so not a surprise). */
+    risksRealised: string[];
+    /** Tasks that regressed WITHOUT being declared. The expensive kind. */
+    unpredictedRegressions: string[];
+    /** Net task delta across the whole suite (passes gained minus lost). */
+    netDelta: number;
+}
+export type ManifestVerdict = 'keep' | 'revert';
+export interface ManifestDecision {
+    verdict: ManifestVerdict;
+    reason: string;
+    /** The Mod to apply to undo this change, when the verdict is 'revert'. */
+    revert?: Mod;
+}
+/**
+ * Check a manifest's prediction against what actually happened.
+ *
+ * `before` and `after` are the record sets from the round the edit was made and
+ * the round after it.
+ */
+export declare function attributeManifest(manifest: ChangeManifest, before: RunRecord[], after: RunRecord[], opts?: {
+    condition?: string;
+    passRatio?: number;
+}): ManifestAttribution;
+export interface ManifestPolicyOptions {
+    /**
+     * Revert when an edit predicted fixes and delivered NONE of them, even if the
+     * net delta is neutral. Default true — an edit that did not do what it claimed
+     * is unexplained behaviour in the harness, and unexplained changes accumulate.
+     */
+    revertOnZeroRealised?: boolean;
+    /**
+     * Tolerated undeclared regressions before an automatic revert. Default 0: the
+     * whole point of declaring risks is that undeclared breakage is not tolerated.
+     */
+    maxUnpredictedRegressions?: number;
+}
+/**
+ * Decide whether a change survives its own prediction.
+ *
+ * Order matters — undeclared regressions are checked FIRST, because an edit that
+ * broke something nobody predicted is disqualified regardless of what else it
+ * happened to fix.
+ */
+export declare function decideManifest(manifest: ChangeManifest, attribution: ManifestAttribution, opts?: ManifestPolicyOptions): ManifestDecision;
+/** Build a manifest for a Mod. `now` is injected so runs stay reproducible. */
+export declare function makeManifest(params: {
+    id: string;
+    mod: Mod;
+    now: string;
+    predictedFixes?: string[];
+    predictedRisks?: string[];
+    touched?: string[];
+}): ChangeManifest;
+/** The file/knob a Mod edits — the granularity a revert operates at. */
+export declare function defaultTouched(mod: Mod): string[];
+interface StoreShape {
+    open: ChangeManifest[];
+    closed: Array<{
+        manifest: ChangeManifest;
+        attribution: ManifestAttribution;
+        decision: ManifestDecision;
+    }>;
+}
+/**
+ * JSON-backed manifest store under `.uap/self-harness/manifests.json`.
+ *
+ * Deliberately a plain file, not SQLite: manifests are few, human-auditable, and
+ * an operator inspecting why the loop reverted its own change should be able to
+ * read them without a client.
+ */
+export declare class ManifestStore {
+    private readonly path;
+    private data;
+    constructor(projectRoot: string);
+    private load;
+    private persist;
+    /** Manifests awaiting next-round attribution. */
+    open(): ChangeManifest[];
+    record(manifest: ChangeManifest): void;
+    /** Move a manifest to the closed log with its verdict. */
+    close(manifest: ChangeManifest, attribution: ManifestAttribution, decision: ManifestDecision): void;
+    closed(): StoreShape['closed'];
+}
+export {};
+//# sourceMappingURL=manifest.d.ts.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"manifest.d.ts","sourceRoot":"","sources":["../../src/self-harness/manifest.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAIH,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,+BAA+B,CAAC;AAC/D,OAAO,EAAE,GAAG,EAA0B,MAAM,WAAW,CAAC;AAExD,4DAA4D;AAC5D,MAAM,WAAW,cAAc;IAC7B,wEAAwE;IACxE,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,GAAG,CAAC;IACT,8EAA8E;IAC9E,SAAS,EAAE,MAAM,CAAC;IAClB,sEAAsE;IACtE,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB,mFAAmF;IACnF,cAAc,EAAE,MAAM,EAAE,CAAC;IACzB;;;OAGG;IACH,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,8DAA8D;IAC9D,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,MAAM,WAAW,mBAAmB;IAClC,UAAU,EAAE,MAAM,CAAC;IACnB,0DAA0D;IAC1D,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,4DAA4D;IAC5D,WAAW,EAAE,MAAM,EAAE,CAAC;IACtB,oEAAoE;IACpE,aAAa,EAAE,MAAM,EAAE,CAAC;IACxB,uEAAuE;IACvE,sBAAsB,EAAE,MAAM,EAAE,CAAC;IACjC,wEAAwE;IACxE,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,eAAe,GAAG,MAAM,GAAG,QAAQ,CAAC;AAEhD,MAAM,WAAW,gBAAgB;IAC/B,OAAO,EAAE,eAAe,CAAC;IACzB,MAAM,EAAE,MAAM,CAAC;IACf,0EAA0E;IAC1E,MAAM,CAAC,EAAE,GAAG,CAAC;CACd;AA2CD;;;;;GAKG;AACH,wBAAgB,iBAAiB,CAC/B,QAAQ,EAAE,cAAc,EACxB,MAAM,EAAE,SAAS,EAAE,EACnB,KAAK,EAAE,SAAS,EAAE,EAClB,IAAI,GAAE;IAAE,SAAS,CAAC,EAAE,MAAM,CAAC;IAAC,SAAS,CAAC,EAAE,MAAM,CAAA;CAAO,GACpD,mBAAmB,CAkDrB;AAED,MAAM,WAAW,qBAAqB;IACpC;;;;OAIG;IACH,oBAAoB,CAAC,EAAE,OAAO,CAAC;IAC/B;;;OAGG;IACH,yBAAyB,CAAC,EAAE,MAAM,CAAC;CACpC;AAED;;;;;;GAMG;AACH,wBAAgB,cAAc,CAC5B,QAAQ,EAAE,cAAc,EACxB,WAAW,EAAE,mBAAmB,EAChC,IAAI,GAAE,qBAA0B,GAC/B,gBAAgB,CA2ClB;AAED,+EAA+E;AAC/E,wBAAgB,YAAY,CAAC,MAAM,EAAE;IACnC,EAAE,EAAE,MAAM,CAAC;IACX,GAAG,EAAE,GAAG,CAAC;IACT,GAAG,EAAE,MAAM,CAAC;IACZ,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,OAAO,CAAC,EAAE,MAAM,EAAE,CAAC;CACpB,GAAG,cAAc,CAUjB;AAED,wEAAwE;AACxE,wBAAgB,cAAc,CAAC,GAAG,EAAE,GAAG,GAAG,MAAM,EAAE,CAajD;AAMD,UAAU,UAAU;IAClB,IAAI,EAAE,cAAc,EAAE,CAAC;IACvB,MAAM,EAAE,KAAK,CAAC;QAAE,QAAQ,EAAE,cAAc,CAAC;QAAC,WAAW,EAAE,mBAAmB,CAAC;QAAC,QAAQ,EAAE,gBAAgB,CAAA;KAAE,CAAC,CAAC;CAC3G;AAID;;;;;;GAMG;AACH,qBAAa,aAAa;IACxB,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAS;IAC9B,OAAO,CAAC,IAAI,CAAa;gBAEb,WAAW,EAAE,MAAM;IAK/B,OAAO,CAAC,IAAI;IAWZ,OAAO,CAAC,OAAO;IASf,iDAAiD;IACjD,IAAI,IAAI,cAAc,EAAE;IAIxB,MAAM,CAAC,QAAQ,EAAE,cAAc,GAAG,IAAI;IAMtC,0DAA0D;IAC1D,KAAK,CACH,QAAQ,EAAE,cAAc,EACxB,WAAW,EAAE,mBAAmB,EAChC,QAAQ,EAAE,gBAAgB,GACzB,IAAI;IAMP,MAAM,IAAI,UAAU,CAAC,QAAQ,CAAC;CAG/B"}

package/dist/self-harness/manifest.js

@@ -0,0 +1,247 @@
+/**
+ * Change manifests — every harness edit is a falsifiable contract
+ * (harness plan area C, 2026-07-31).
+ *
+ * Agentic Harness Engineering (arXiv 2604.25850) attributes a large part of its
+ * Terminal-Bench 2 gain (69.7% -> 77.0%) not to smarter proposals but to
+ * DECISION OBSERVABILITY: every edit is paired with a self-declared prediction
+ * about which tasks it will fix and which it might break, that prediction is
+ * checked against the next round's actual per-task outcomes, and edits whose
+ * predictions do not materialise are reverted at file granularity.
+ *
+ * Without this, an accepted Mod is accepted forever on the strength of one
+ * noisy A/B. With it, acceptance is provisional and the loop can un-learn.
+ *
+ * Everything here is pure except the JSON store at the bottom, so the
+ * attribution rules are unit-testable without a bench run.
+ */
+import { existsSync, mkdirSync, readFileSync, writeFileSync } from 'node:fs';
+import { dirname, join } from 'node:path';
+import { describeMod, invertMod } from './mods.js';
+/**
+ * Task ids that passed, from one CONDITION's records.
+ *
+ * Scoping to a condition is essential, not tidiness: a real `records.jsonl` from
+ * the paired bench holds BOTH arms, so a task the treatment solves and the
+ * baseline fails would read as "not passing" in every round, every predicted fix
+ * would look unrealised, and `revertOnZeroRealised` would revert every change
+ * the loop ever made.
+ *
+ * `passRatio` is the flakiness bar. Requiring a clean sweep sounds strict but on
+ * a stochastic local model it means a task lifted from 1/5 to 4/5 seeds — a real
+ * improvement — scores as no fix at all. Default 0.6: pass the majority of seeds.
+ */
+function passingTasks(records, condition, passRatio = 0.6) {
+    const seen = new Map();
+    for (const r of records) {
+        if (condition && r.condition !== condition)
+            continue;
+        const acc = seen.get(r.taskId) ?? { pass: 0, total: 0 };
+        acc.total++;
+        if (r.metrics.correct)
+            acc.pass++;
+        seen.set(r.taskId, acc);
+    }
+    const passed = new Set();
+    for (const [taskId, acc] of seen) {
+        if (acc.total > 0 && acc.pass / acc.total >= passRatio)
+            passed.add(taskId);
+    }
+    return passed;
+}
+/** The condition to attribute against when the caller does not name one. */
+function inferCondition(records) {
+    const labels = new Set(records.map((r) => r.condition));
+    // One arm: unambiguous. Several: prefer the shipping surface, else give up and
+    // let the caller name it rather than silently mixing arms.
+    if (labels.size === 1)
+        return [...labels][0];
+    for (const preferred of ['uap-full', 'candidate', 'with-mod']) {
+        if (labels.has(preferred))
+            return preferred;
+    }
+    return undefined;
+}
+/**
+ * Check a manifest's prediction against what actually happened.
+ *
+ * `before` and `after` are the record sets from the round the edit was made and
+ * the round after it.
+ */
+export function attributeManifest(manifest, before, after, opts = {}) {
+    const condition = opts.condition ?? inferCondition(after) ?? inferCondition(before);
+    const passedBefore = passingTasks(before, condition, opts.passRatio);
+    const passedAfter = passingTasks(after, condition, opts.passRatio);
+    // INTERSECTION, not union. A task present in `before` and absent from `after`
+    // (suite edited, run aborted, sampled subset, adapter crash) is not a
+    // regression — absence is not failure — but a union scored it as one and
+    // triggered an immediate revert.
+    const beforeIds = new Set(before.filter((r) => !condition || r.condition === condition).map((r) => r.taskId));
+    const afterIds = new Set(after.filter((r) => !condition || r.condition === condition).map((r) => r.taskId));
+    const allTasks = new Set([...beforeIds].filter((t) => afterIds.has(t)));
+    const fixesRealised = [];
+    const fixesMissed = [];
+    for (const t of manifest.predictedFixes) {
+        // A predicted fix for a task that did not run again is unproven, not missed —
+        // counting it as missed would revert the change on an incomplete round.
+        if (!afterIds.has(t))
+            continue;
+        if (!passedBefore.has(t) && passedAfter.has(t))
+            fixesRealised.push(t);
+        else
+            fixesMissed.push(t);
+    }
+    const risky = new Set(manifest.predictedRisks);
+    const risksRealised = [];
+    const unpredictedRegressions = [];
+    let gained = 0;
+    let lost = 0;
+    for (const t of allTasks) {
+        const wasPassing = passedBefore.has(t);
+        const isPassing = passedAfter.has(t);
+        if (!wasPassing && isPassing)
+            gained++;
+        if (wasPassing && !isPassing) {
+            lost++;
+            if (risky.has(t))
+                risksRealised.push(t);
+            else
+                unpredictedRegressions.push(t);
+        }
+    }
+    return {
+        manifestId: manifest.id,
+        fixesRealised,
+        fixesMissed,
+        risksRealised,
+        unpredictedRegressions,
+        netDelta: gained - lost,
+    };
+}
+/**
+ * Decide whether a change survives its own prediction.
+ *
+ * Order matters — undeclared regressions are checked FIRST, because an edit that
+ * broke something nobody predicted is disqualified regardless of what else it
+ * happened to fix.
+ */
+export function decideManifest(manifest, attribution, opts = {}) {
+    const maxUnpredicted = opts.maxUnpredictedRegressions ?? 0;
+    const revertOnZero = opts.revertOnZeroRealised ?? true;
+    if (attribution.unpredictedRegressions.length > maxUnpredicted) {
+        return {
+            verdict: 'revert',
+            reason: `${attribution.unpredictedRegressions.length} undeclared regression(s) ` +
+                `(${attribution.unpredictedRegressions.slice(0, 5).join(', ')}) — ` +
+                `the manifest did not declare them.`,
+            revert: invertMod(manifest.mod),
+        };
+    }
+    if (attribution.netDelta < 0) {
+        return {
+            verdict: 'revert',
+            reason: `net task delta ${attribution.netDelta} — the change costs more than it wins.`,
+            revert: invertMod(manifest.mod),
+        };
+    }
+    // Gate on what was actually OBSERVED this round (realised + missed), not on
+    // what was predicted: if none of the predicted tasks re-ran, there is no
+    // evidence either way and reverting would punish an incomplete round.
+    const observed = attribution.fixesRealised.length + attribution.fixesMissed.length;
+    if (revertOnZero && observed > 0 && attribution.fixesRealised.length === 0) {
+        return {
+            verdict: 'revert',
+            reason: `predicted ${manifest.predictedFixes.length} fix(es), realised none — ` +
+                `the edit did not do what it claimed.`,
+            revert: invertMod(manifest.mod),
+        };
+    }
+    return {
+        verdict: 'keep',
+        reason: `realised ${attribution.fixesRealised.length}/${manifest.predictedFixes.length} predicted fix(es), ` +
+            `net delta ${attribution.netDelta}, no undeclared regressions.`,
+    };
+}
+/** Build a manifest for a Mod. `now` is injected so runs stay reproducible. */
+export function makeManifest(params) {
+    return {
+        id: params.id,
+        mod: params.mod,
+        createdAt: params.now,
+        predictedFixes: params.predictedFixes ?? [],
+        predictedRisks: params.predictedRisks ?? [],
+        touched: params.touched ?? defaultTouched(params.mod),
+        summary: describeMod(params.mod),
+    };
+}
+/** The file/knob a Mod edits — the granularity a revert operates at. */
+export function defaultTouched(mod) {
+    switch (mod.kind) {
+        case 'env':
+            return [`env:${mod.key}`];
+        case 'scaffold':
+            return [`scaffold:${mod.component}`];
+        case 'middleware':
+            return [`middleware:${mod.id}`];
+        case 'config':
+            return [`config:${mod.key}`];
+        case 'tool':
+            return [`tool:${mod.key}`];
+    }
+}
+const EMPTY_STORE = { open: [], closed: [] };
+/**
+ * JSON-backed manifest store under `.uap/self-harness/manifests.json`.
+ *
+ * Deliberately a plain file, not SQLite: manifests are few, human-auditable, and
+ * an operator inspecting why the loop reverted its own change should be able to
+ * read them without a client.
+ */
+export class ManifestStore {
+    path;
+    data;
+    constructor(projectRoot) {
+        this.path = join(projectRoot, '.uap', 'self-harness', 'manifests.json');
+        this.data = this.load();
+    }
+    load() {
+        try {
+            if (!existsSync(this.path))
+                return { ...EMPTY_STORE, open: [], closed: [] };
+            const parsed = JSON.parse(readFileSync(this.path, 'utf-8'));
+            return { open: parsed.open ?? [], closed: parsed.closed ?? [] };
+        }
+        catch {
+            // A corrupt store must not wedge the loop; start clean and keep going.
+            return { open: [], closed: [] };
+        }
+    }
+    persist() {
+        try {
+            mkdirSync(dirname(this.path), { recursive: true });
+            writeFileSync(this.path, `${JSON.stringify(this.data, null, 2)}\n`, 'utf-8');
+        }
+        catch {
+            /* best-effort */
+        }
+    }
+    /** Manifests awaiting next-round attribution. */
+    open() {
+        return [...this.data.open];
+    }
+    record(manifest) {
+        this.data.open = this.data.open.filter((m) => m.id !== manifest.id);
+        this.data.open.push(manifest);
+        this.persist();
+    }
+    /** Move a manifest to the closed log with its verdict. */
+    close(manifest, attribution, decision) {
+        this.data.open = this.data.open.filter((m) => m.id !== manifest.id);
+        this.data.closed.push({ manifest, attribution, decision });
+        this.persist();
+    }
+    closed() {
+        return [...this.data.closed];
+    }
+}
+//# sourceMappingURL=manifest.js.map

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

@@ -0,0 +1 @@
+{"version":3,"file":"manifest.js","sourceRoot":"","sources":["../../src/self-harness/manifest.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;GAgBG;AAEH,OAAO,EAAE,UAAU,EAAE,SAAS,EAAE,YAAY,EAAE,aAAa,EAAE,MAAM,SAAS,CAAC;AAC7E,OAAO,EAAE,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAE1C,OAAO,EAAO,WAAW,EAAE,SAAS,EAAE,MAAM,WAAW,CAAC;AA6CxD;;;;;;;;;;;;GAYG;AACH,SAAS,YAAY,CAAC,OAAoB,EAAE,SAAkB,EAAE,SAAS,GAAG,GAAG;IAC7E,MAAM,IAAI,GAAG,IAAI,GAAG,EAA2C,CAAC;IAChE,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,IAAI,SAAS,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS;YAAE,SAAS;QACrD,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,CAAC;QACxD,GAAG,CAAC,KAAK,EAAE,CAAC;QACZ,IAAI,CAAC,CAAC,OAAO,CAAC,OAAO;YAAE,GAAG,CAAC,IAAI,EAAE,CAAC;QAClC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC;IAC1B,CAAC;IACD,MAAM,MAAM,GAAG,IAAI,GAAG,EAAU,CAAC;IACjC,KAAK,MAAM,CAAC,MAAM,EAAE,GAAG,CAAC,IAAI,IAAI,EAAE,CAAC;QACjC,IAAI,GAAG,CAAC,KAAK,GAAG,CAAC,IAAI,GAAG,CAAC,IAAI,GAAG,GAAG,CAAC,KAAK,IAAI,SAAS;YAAE,MAAM,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IAC7E,CAAC;IACD,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,4EAA4E;AAC5E,SAAS,cAAc,CAAC,OAAoB;IAC1C,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC;IACxD,+EAA+E;IAC/E,2DAA2D;IAC3D,IAAI,MAAM,CAAC,IAAI,KAAK,CAAC;QAAE,OAAO,CAAC,GAAG,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC;IAC7C,KAAK,MAAM,SAAS,IAAI,CAAC,UAAU,EAAE,WAAW,EAAE,UAAU,CAAC,EAAE,CAAC;QAC9D,IAAI,MAAM,CAAC,GAAG,CAAC,SAAS,CAAC;YAAE,OAAO,SAAS,CAAC;IAC9C,CAAC;IACD,OAAO,SAAS,CAAC;AACnB,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,iBAAiB,CAC/B,QAAwB,EACxB,MAAmB,EACnB,KAAkB,EAClB,OAAmD,EAAE;IAErD,MAAM,SAAS,GAAG,IAAI,CAAC,SAAS,IAAI,cAAc,CAAC,KAAK,CAAC,IAAI,cAAc,CAAC,MAAM,CAAC,CAAC;IACpF,MAAM,YAAY,GAAG,YAAY,CAAC,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IACrE,MAAM,WAAW,GAAG,YAAY,CAAC,KAAK,EAAE,SAAS,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;IACnE,8EAA8E;IAC9E,sEAAsE;IACtE,yEAAyE;IACzE,iCAAiC;IACjC,MAAM,SAAS,GAAG,IAAI,GAAG,CACvB,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CACnF,CAAC;IACF,MAAM,QAAQ,GAAG,IAAI,GAAG,CACtB,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,SAAS,IAAI,CAAC,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,MAAM,CAAC,CAClF,CAAC;IACF,MAAM,QAAQ,GAAG,IAAI,GAAG,CAAS,CAAC,GAAG,SAAS,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAEhF,MAAM,aAAa,GAAa,EAAE,CAAC;IACnC,MAAM,WAAW,GAAa,EAAE,CAAC;IACjC,KAAK,MAAM,CAAC,IAAI,QAAQ,CAAC,cAAc,EAAE,CAAC;QACxC,8EAA8E;QAC9E,wEAAwE;QACxE,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,CAAC;YAAE,SAAS;QAC/B,IAAI,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC;YAAE,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;;YACjE,WAAW,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;IAC3B,CAAC;IAED,MAAM,KAAK,GAAG,IAAI,GAAG,CAAC,QAAQ,CAAC,cAAc,CAAC,CAAC;IAC/C,MAAM,aAAa,GAAa,EAAE,CAAC;IACnC,MAAM,sBAAsB,GAAa,EAAE,CAAC;IAC5C,IAAI,MAAM,GAAG,CAAC,CAAC;IACf,IAAI,IAAI,GAAG,CAAC,CAAC;IACb,KAAK,MAAM,CAAC,IAAI,QAAQ,EAAE,CAAC;QACzB,MAAM,UAAU,GAAG,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACvC,MAAM,SAAS,GAAG,WAAW,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACrC,IAAI,CAAC,UAAU,IAAI,SAAS;YAAE,MAAM,EAAE,CAAC;QACvC,IAAI,UAAU,IAAI,CAAC,SAAS,EAAE,CAAC;YAC7B,IAAI,EAAE,CAAC;YACP,IAAI,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC;gBAAE,aAAa,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;;gBACnC,sBAAsB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACtC,CAAC;IACH,CAAC;IAED,OAAO;QACL,UAAU,EAAE,QAAQ,CAAC,EAAE;QACvB,aAAa;QACb,WAAW;QACX,aAAa;QACb,sBAAsB;QACtB,QAAQ,EAAE,MAAM,GAAG,IAAI;KACxB,CAAC;AACJ,CAAC;AAgBD;;;;;;GAMG;AACH,MAAM,UAAU,cAAc,CAC5B,QAAwB,EACxB,WAAgC,EAChC,OAA8B,EAAE;IAEhC,MAAM,cAAc,GAAG,IAAI,CAAC,yBAAyB,IAAI,CAAC,CAAC;IAC3D,MAAM,YAAY,GAAG,IAAI,CAAC,oBAAoB,IAAI,IAAI,CAAC;IAEvD,IAAI,WAAW,CAAC,sBAAsB,CAAC,MAAM,GAAG,cAAc,EAAE,CAAC;QAC/D,OAAO;YACL,OAAO,EAAE,QAAQ;YACjB,MAAM,EACJ,GAAG,WAAW,CAAC,sBAAsB,CAAC,MAAM,4BAA4B;gBACxE,IAAI,WAAW,CAAC,sBAAsB,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM;gBACnE,oCAAoC;YACtC,MAAM,EAAE,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC;SAChC,CAAC;IACJ,CAAC;IAED,IAAI,WAAW,CAAC,QAAQ,GAAG,CAAC,EAAE,CAAC;QAC7B,OAAO;YACL,OAAO,EAAE,QAAQ;YACjB,MAAM,EAAE,kBAAkB,WAAW,CAAC,QAAQ,wCAAwC;YACtF,MAAM,EAAE,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC;SAChC,CAAC;IACJ,CAAC;IAED,4EAA4E;IAC5E,yEAAyE;IACzE,sEAAsE;IACtE,MAAM,QAAQ,GAAG,WAAW,CAAC,aAAa,CAAC,MAAM,GAAG,WAAW,CAAC,WAAW,CAAC,MAAM,CAAC;IACnF,IAAI,YAAY,IAAI,QAAQ,GAAG,CAAC,IAAI,WAAW,CAAC,aAAa,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC3E,OAAO;YACL,OAAO,EAAE,QAAQ;YACjB,MAAM,EACJ,aAAa,QAAQ,CAAC,cAAc,CAAC,MAAM,4BAA4B;gBACvE,sCAAsC;YACxC,MAAM,EAAE,SAAS,CAAC,QAAQ,CAAC,GAAG,CAAC;SAChC,CAAC;IACJ,CAAC;IAED,OAAO;QACL,OAAO,EAAE,MAAM;QACf,MAAM,EACJ,YAAY,WAAW,CAAC,aAAa,CAAC,MAAM,IAAI,QAAQ,CAAC,cAAc,CAAC,MAAM,sBAAsB;YACpG,aAAa,WAAW,CAAC,QAAQ,8BAA8B;KAClE,CAAC;AACJ,CAAC;AAED,+EAA+E;AAC/E,MAAM,UAAU,YAAY,CAAC,MAO5B;IACC,OAAO;QACL,EAAE,EAAE,MAAM,CAAC,EAAE;QACb,GAAG,EAAE,MAAM,CAAC,GAAG;QACf,SAAS,EAAE,MAAM,CAAC,GAAG;QACrB,cAAc,EAAE,MAAM,CAAC,cAAc,IAAI,EAAE;QAC3C,cAAc,EAAE,MAAM,CAAC,cAAc,IAAI,EAAE;QAC3C,OAAO,EAAE,MAAM,CAAC,OAAO,IAAI,cAAc,CAAC,MAAM,CAAC,GAAG,CAAC;QACrD,OAAO,EAAE,WAAW,CAAC,MAAM,CAAC,GAAG,CAAC;KACjC,CAAC;AACJ,CAAC;AAED,wEAAwE;AACxE,MAAM,UAAU,cAAc,CAAC,GAAQ;IACrC,QAAQ,GAAG,CAAC,IAAI,EAAE,CAAC;QACjB,KAAK,KAAK;YACR,OAAO,CAAC,OAAO,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;QAC5B,KAAK,UAAU;YACb,OAAO,CAAC,YAAY,GAAG,CAAC,SAAS,EAAE,CAAC,CAAC;QACvC,KAAK,YAAY;YACf,OAAO,CAAC,cAAc,GAAG,CAAC,EAAE,EAAE,CAAC,CAAC;QAClC,KAAK,QAAQ;YACX,OAAO,CAAC,UAAU,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;QAC/B,KAAK,MAAM;YACT,OAAO,CAAC,QAAQ,GAAG,CAAC,GAAG,EAAE,CAAC,CAAC;IAC/B,CAAC;AACH,CAAC;AAWD,MAAM,WAAW,GAAe,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;AAEzD;;;;;;GAMG;AACH,MAAM,OAAO,aAAa;IACP,IAAI,CAAS;IACtB,IAAI,CAAa;IAEzB,YAAY,WAAmB;QAC7B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,WAAW,EAAE,MAAM,EAAE,cAAc,EAAE,gBAAgB,CAAC,CAAC;QACxE,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC1B,CAAC;IAEO,IAAI;QACV,IAAI,CAAC;YACH,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC;gBAAE,OAAO,EAAE,GAAG,WAAW,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;YAC5E,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,YAAY,CAAC,IAAI,CAAC,IAAI,EAAE,OAAO,CAAC,CAAwB,CAAC;YACnF,OAAO,EAAE,IAAI,EAAE,MAAM,CAAC,IAAI,IAAI,EAAE,EAAE,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,EAAE,EAAE,CAAC;QAClE,CAAC;QAAC,MAAM,CAAC;YACP,uEAAuE;YACvE,OAAO,EAAE,IAAI,EAAE,EAAE,EAAE,MAAM,EAAE,EAAE,EAAE,CAAC;QAClC,CAAC;IACH,CAAC;IAEO,OAAO;QACb,IAAI,CAAC;YACH,SAAS,CAAC,OAAO,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;YACnD,aAAa,CAAC,IAAI,CAAC,IAAI,EAAE,GAAG,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,EAAE,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QAC/E,CAAC;QAAC,MAAM,CAAC;YACP,iBAAiB;QACnB,CAAC;IACH,CAAC;IAED,iDAAiD;IACjD,IAAI;QACF,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC7B,CAAC;IAED,MAAM,CAAC,QAAwB;QAC7B,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,QAAQ,CAAC,EAAE,CAAC,CAAC;QACpE,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAC9B,IAAI,CAAC,OAAO,EAAE,CAAC;IACjB,CAAC;IAED,0DAA0D;IAC1D,KAAK,CACH,QAAwB,EACxB,WAAgC,EAChC,QAA0B;QAE1B,IAAI,CAAC,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,QAAQ,CAAC,EAAE,CAAC,CAAC;QACpE,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,EAAE,QAAQ,EAAE,WAAW,EAAE,QAAQ,EAAE,CAAC,CAAC;QAC3D,IAAI,CAAC,OAAO,EAAE,CAAC;IACjB,CAAC;IAED,MAAM;QACJ,OAAO,CAAC,GAAG,IAAI,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAC/B,CAAC;CACF"}

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

@@ -12,8 +12,15 @@
  */
 import { UapComponent } from '../benchmarks/paired/types.js';
 export type JsonScalar = string | number | boolean;
+/**
+ * Where a knob takes effect. `executor` was added with ToolMod (harness plan
+ * B1): unlike proxy/llama targets it is read CLIENT-side by the delivery
+ * executor, so changing it needs an env var on the candidate arm rather than an
+ * inference-server restart — which is precisely why tool knobs can auto-validate.
+ */
+export type KnobTarget = 'proxy' | 'llama' | 'executor';
 export interface NumericKnob {
-    target: 'proxy' | 'llama';
+    target: KnobTarget;
     type: 'number';
     /** Inclusive safe range. A proposed value outside this is rejected. */
     min: number;
@@ -23,7 +30,7 @@ export interface NumericKnob {
     description: string;
 }
 export interface EnumKnob {
-    target: 'proxy' | 'llama';
+    target: KnobTarget;
     type: 'enum';
     values: readonly string[];
     description: string;
@@ -90,6 +97,39 @@ 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 declare const TOOL_KNOB_ALLOWLIST: {
+    readonly UAP_EDIT_TOLERANT: {
+        readonly target: "executor";
+        readonly type: "enum";
+        readonly values: readonly ["0", "1"];
+        readonly description: "edit_file falls back to whitespace-tolerant matching on an exact miss";
+    };
+    readonly UAP_EDIT_DIAGNOSTICS: {
+        readonly target: "executor";
+        readonly type: "enum";
+        readonly values: readonly ["0", "1"];
+        readonly description: "a failed edit returns the nearest current region instead of \"not found\"";
+    };
+    readonly UAP_READ_WINDOW_BYTES: {
+        readonly target: "executor";
+        readonly type: "number";
+        readonly min: 2000;
+        readonly max: 32000;
+        readonly integer: true;
+        readonly description: "bytes of a file read_file returns per call";
+    };
+    readonly UAP_MAX_TOOL_ROUNDS: {
+        readonly target: "executor";
+        readonly type: "number";
+        readonly min: 4;
+        readonly max: 40;
+        readonly integer: true;
+        readonly description: "tool-call rounds before a final answer is forced";
+    };
+};
+export type KnownToolKnob = keyof typeof TOOL_KNOB_ALLOWLIST;
+export declare function isKnownToolKnob(k: string): k is KnownToolKnob;
+export declare function toolKnobSpec(key: KnownToolKnob): KnobSpec;
 export interface EnvMod {
     kind: 'env';
     key: KnownKnob;
@@ -131,7 +171,21 @@ export interface ConfigMod {
     /** settings-registry category id (for grouping/diffs). */
     category: string;
 }
-export type Mod = EnvMod | ScaffoldMod | MiddlewareMod | ConfigMod;
+/**
+ * A change to the TOOL SURFACE the model sees (harness plan B1). Same shape as
+ * an EnvMod, different allow-list and a different application seam: tool knobs
+ * are read by the executor process, so an A/B needs an env var on the candidate
+ * arm rather than an inference-server restart.
+ */
+export interface ToolMod {
+    kind: 'tool';
+    key: KnownToolKnob;
+    /** Prior value (for revert + audit). */
+    from: string;
+    /** Proposed value. */
+    to: string;
+}
+export type Mod = EnvMod | ScaffoldMod | MiddlewareMod | ConfigMod | ToolMod;
 export interface ValidationOk {
     ok: true;
 }

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

@@ -1 +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;AAG7E,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;;;;;;;GAOG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,QAAQ,CAAC;IACf,iEAAiE;IACjE,GAAG,EAAE,MAAM,CAAC;IACZ,qDAAqD;IACrD,IAAI,EAAE,MAAM,CAAC;IACb,oCAAoC;IACpC,EAAE,EAAE,MAAM,CAAC;IACX,0DAA0D;IAC1D,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,MAAM,GAAG,GAAG,MAAM,GAAG,WAAW,GAAG,aAAa,GAAG,SAAS,CAAC;AAEnE,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,CAoDtD;AA+BD,kFAAkF;AAClF,wBAAgB,WAAW,CAAC,GAAG,EAAE,GAAG,GAAG,MAAM,CAW5C;AAED;;;GAGG;AACH,wBAAgB,SAAS,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,CAmBvC"}
+{"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;AAG7E,MAAM,MAAM,UAAU,GAAG,MAAM,GAAG,MAAM,GAAG,OAAO,CAAC;AASnD;;;;;GAKG;AACH,MAAM,MAAM,UAAU,GAAG,OAAO,GAAG,OAAO,GAAG,UAAU,CAAC;AAExD,MAAM,WAAW,WAAW;IAC1B,MAAM,EAAE,UAAU,CAAC;IACnB,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,UAAU,CAAC;IACnB,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;AAoBD,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAiBa,CAAC;AAE9C,MAAM,MAAM,aAAa,GAAG,MAAM,OAAO,mBAAmB,CAAC;AAE7D,wBAAgB,eAAe,CAAC,CAAC,EAAE,MAAM,GAAG,CAAC,IAAI,aAAa,CAE7D;AAED,wBAAgB,YAAY,CAAC,GAAG,EAAE,aAAa,GAAG,QAAQ,CAEzD;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;;;;;;;GAOG;AACH,MAAM,WAAW,SAAS;IACxB,IAAI,EAAE,QAAQ,CAAC;IACf,iEAAiE;IACjE,GAAG,EAAE,MAAM,CAAC;IACZ,qDAAqD;IACrD,IAAI,EAAE,MAAM,CAAC;IACb,oCAAoC;IACpC,EAAE,EAAE,MAAM,CAAC;IACX,0DAA0D;IAC1D,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED;;;;;GAKG;AACH,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,MAAM,CAAC;IACb,GAAG,EAAE,aAAa,CAAC;IACnB,wCAAwC;IACxC,IAAI,EAAE,MAAM,CAAC;IACb,sBAAsB;IACtB,EAAE,EAAE,MAAM,CAAC;CACZ;AAED,MAAM,MAAM,GAAG,GAAG,MAAM,GAAG,WAAW,GAAG,aAAa,GAAG,SAAS,GAAG,OAAO,CAAC;AAE7E,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,CA8EtD;AA+BD,kFAAkF;AAClF,wBAAgB,WAAW,CAAC,GAAG,EAAE,GAAG,GAAG,MAAM,CAa5C;AAED;;;GAGG;AACH,wBAAgB,SAAS,CAAC,GAAG,EAAE,GAAG,GAAG,GAAG,CAqBvC"}