@cat-factory/orchestration 0.24.2 → 0.25.0

package/dist/modules/kaizen/KaizenService.js

@@ -0,0 +1,322 @@
+import { DEFAULT_WORKSPACE_SETTINGS, extractJson, inlineModelRef, resolveScopedModelProvider, } from '@cat-factory/kernel';
+import { generateText } from 'ai';
+import { catFactoryObservability, KAIZEN_SYSTEM_PROMPT, promptVersionForKind, } from '@cat-factory/agents';
+import { comboKeyFor, isVerified, nextComboState } from './kaizen.logic.js';
+/** The agent kind keying the workspace default model + observability for the grader. */
+const KAIZEN_AGENT_KIND = 'kaizen';
+/** Newest LLM calls to digest into the grader prompt (the bodies are heavy). */
+const MAX_CALLS_DIGESTED = 40;
+/**
+ * The Kaizen agent: after a run completes, it grades each completed agent step by how
+ * smooth / guided / efficient the interaction was vs confused / chaotic, and recommends
+ * improvements. A combo `(promptVersion, agentKind, model)` that earns a high grade with
+ * no recommendations enough times in a row is VERIFIED and stops being graded.
+ *
+ * Grading is two-phase so it never blocks a run: {@link scheduleForRun} (called from the
+ * engine's terminal hook) only inserts `scheduled` rows; {@link runGrading} (driven by the
+ * background sweep) does the LLM analysis. The grader resolves its model exactly like an
+ * agent step — for the `kaizen` kind — so operators configure it in Model Configuration.
+ */
+export class KaizenService {
+    deps;
+    constructor(deps) {
+        this.deps = deps;
+    }
+    /** Whether the LLM-backed grader is available (else gradings settle as `failed`). */
+    get enabled() {
+        return (!!this.deps.modelProviderResolver || !!this.deps.modelProvider) && !!this.deps.modelRef;
+    }
+    // ---- phase 1: schedule (called at run completion, never blocks) ----------
+    /**
+     * Schedule a grading for each completed agent step of a finished run, skipping verified
+     * combos and steps already graded. Best-effort: a failure here must never derail the
+     * engine's emit (the caller wraps it), and re-driving the same run is idempotent.
+     */
+    async scheduleForRun(workspaceId, instance) {
+        // No grader model wired ⇒ every scheduled row could only ever settle `failed`, so don't
+        // flood the table (and the run-window event stream) with rows for a disabled grader.
+        if (!this.enabled)
+            return;
+        if (!(await this.kaizenEnabled(workspaceId)))
+            return;
+        for (let stepIndex = 0; stepIndex < instance.steps.length; stepIndex++) {
+            const step = instance.steps[stepIndex];
+            if (!step || !this.isGradeable(step) || !step.model)
+                continue;
+            const model = step.model;
+            const promptVersion = promptVersionForKind(step.agentKind);
+            const comboKey = comboKeyFor(step.agentKind, model, promptVersion);
+            const combo = await this.deps.kaizenVerifiedComboRepository.getByKey(workspaceId, comboKey);
+            if (isVerified(combo))
+                continue;
+            const existing = await this.deps.kaizenGradingRepository.getByStep(workspaceId, instance.id, stepIndex);
+            if (existing)
+                continue;
+            const now = this.deps.clock.now();
+            const grading = {
+                id: this.deps.idGenerator.next('kzn'),
+                executionId: instance.id,
+                blockId: instance.blockId,
+                stepIndex,
+                agentKind: step.agentKind,
+                model,
+                promptVersion,
+                comboKey,
+                status: 'scheduled',
+                grade: null,
+                summary: '',
+                recommendations: [],
+                graderModel: null,
+                error: null,
+                createdAt: now,
+                updatedAt: now,
+            };
+            await this.deps.kaizenGradingRepository.upsert(workspaceId, grading);
+            await this.emit(workspaceId, grading);
+        }
+    }
+    // ---- phase 2: run (driven by the background sweep) -----------------------
+    /**
+     * Run a batch of pending gradings: `scheduled` rows plus `running` rows orphaned by a
+     * crashed sweep (older than `staleBefore`). Returns how many it processed.
+     */
+    async runPending(staleBefore, limit) {
+        const pending = await this.deps.kaizenGradingRepository.listPending(staleBefore, limit);
+        let processed = 0;
+        for (const { workspaceId, grading } of pending) {
+            // Atomically claim before working it: a concurrent/overlapping sweep pass that listed
+            // the same row loses the claim and skips it, so a grading is processed at most once.
+            const claimed = await this.deps.kaizenGradingRepository.claim(workspaceId, grading.id, staleBefore, this.deps.clock.now());
+            if (!claimed)
+                continue;
+            try {
+                await this.runGrading(workspaceId, grading);
+                processed++;
+            }
+            catch {
+                // runGrading already records a `failed` row; never let one bad grading abort the batch.
+            }
+        }
+        return processed;
+    }
+    /** Grade one step: mark running, digest telemetry, call the grader, record + update streak. */
+    async runGrading(workspaceId, grading) {
+        const running = {
+            ...grading,
+            status: 'running',
+            updatedAt: this.deps.clock.now(),
+        };
+        await this.deps.kaizenGradingRepository.upsert(workspaceId, running);
+        await this.emit(workspaceId, running);
+        if (!this.enabled) {
+            await this.fail(workspaceId, running, 'No model is configured for the Kaizen agent');
+            return;
+        }
+        try {
+            // Fetch only THIS step kind's calls (filtered in SQL) so the cap is spent on the
+            // graded kind rather than being crowded out by a long run's other kinds. The metric
+            // store keys calls by (execution, agentKind) — there is no per-step discriminator —
+            // so if a run ran the SAME kind in two steps their calls are merged here; the
+            // provided-context snapshot below is still matched precisely by stepIndex.
+            const [snapshot, stepCalls] = await Promise.all([
+                this.snapshotForStep(workspaceId, grading),
+                this.deps.llmCallMetricRepository.listByExecution(workspaceId, grading.executionId, MAX_CALLS_DIGESTED, grading.agentKind),
+            ]);
+            // Don't grade blind. With neither a provided-context snapshot NOR any recorded LLM
+            // calls (e.g. the deployment has prompt recording off), the grader has no evidence to
+            // judge the interaction on — it would be guessing, and a guessed high-grade-with-no-recs
+            // would advance the combo streak toward a bogus `verified`, after which the engine stops
+            // grading that combo for good. Settle `failed` instead so the combo is left untouched.
+            if (!snapshot && stepCalls.length === 0) {
+                await this.fail(workspaceId, running, 'No telemetry was captured for this step (prompt recording may be off), so it cannot be graded');
+                return;
+            }
+            const { ref, provider } = await this.resolveModel(workspaceId, grading.blockId);
+            const model = provider.resolve(ref);
+            const result = await generateText({
+                model,
+                system: KAIZEN_SYSTEM_PROMPT,
+                prompt: buildKaizenPrompt(grading, snapshot, stepCalls),
+                temperature: 0.2,
+                maxOutputTokens: 2000,
+                providerOptions: catFactoryObservability({ agentKind: KAIZEN_AGENT_KIND, workspaceId }),
+            });
+            const verdict = parseVerdict(result.text);
+            const now = this.deps.clock.now();
+            const complete = {
+                ...grading,
+                status: 'complete',
+                grade: verdict.grade,
+                summary: verdict.summary,
+                recommendations: verdict.recommendations,
+                graderModel: `${ref.provider}:${ref.model}`,
+                error: null,
+                updatedAt: now,
+            };
+            await this.deps.kaizenGradingRepository.upsert(workspaceId, complete);
+            await this.updateCombo(workspaceId, complete, now);
+            await this.emit(workspaceId, complete);
+        }
+        catch (e) {
+            await this.fail(workspaceId, running, e instanceof Error ? e.message : String(e));
+        }
+    }
+    // ---- read surface -------------------------------------------------------
+    /** The Kaizen screen payload: recent history + the verified-combo library. */
+    async getOverview(workspaceId, limit = 200) {
+        const [gradings, verified] = await Promise.all([
+            this.deps.kaizenGradingRepository.listByWorkspace(workspaceId, limit),
+            this.deps.kaizenVerifiedComboRepository.listByWorkspace(workspaceId),
+        ]);
+        return { gradings, verified };
+    }
+    /** The gradings recorded for a single run (the run-window status surface). */
+    listForExecution(workspaceId, executionId) {
+        return this.deps.kaizenGradingRepository.listByExecution(workspaceId, executionId);
+    }
+    // ---- internals ----------------------------------------------------------
+    async kaizenEnabled(workspaceId) {
+        if (!this.deps.workspaceSettingsRepository)
+            return DEFAULT_WORKSPACE_SETTINGS.kaizenEnabled;
+        const settings = await this.deps.workspaceSettingsRepository.get(workspaceId);
+        return settings?.kaizenEnabled ?? DEFAULT_WORKSPACE_SETTINGS.kaizenEnabled;
+    }
+    /** A step is gradeable when it ran an LLM to completion (so it has a resolved model). */
+    isGradeable(step) {
+        return step.state === 'done' && !step.skipped && !!step.model;
+    }
+    async snapshotForStep(workspaceId, grading) {
+        const snapshots = await this.deps.agentContextObservability.listByExecution(workspaceId, grading.executionId);
+        return (snapshots.find((s) => s.stepIndex === grading.stepIndex) ??
+            snapshots.find((s) => s.agentKind === grading.agentKind) ??
+            null);
+    }
+    async resolveModel(workspaceId, blockId) {
+        const provider = await resolveScopedModelProvider(workspaceId, this.deps);
+        const ref = await this.modelFor(workspaceId, blockId);
+        if (!provider || !ref)
+            throw new Error('No model is configured for the Kaizen agent');
+        return { provider, ref };
+    }
+    /** Block pin > workspace per-kind default (`kaizen`) > routing default. */
+    async modelFor(workspaceId, blockId) {
+        const fallback = this.deps.modelRef;
+        const resolve = (ref) => inlineModelRef(ref, fallback ?? ref);
+        const block = await this.deps.blockRepository.get(workspaceId, blockId);
+        const fromBlock = this.deps.resolveBlockModel?.(block?.modelId);
+        if (fromBlock)
+            return resolve(fromBlock);
+        const defaultId = await this.deps.resolveWorkspaceModelDefault?.(workspaceId, KAIZEN_AGENT_KIND, block?.modelPresetId);
+        const fromDefault = this.deps.resolveBlockModel?.(defaultId);
+        if (fromDefault)
+            return resolve(fromDefault);
+        return fallback;
+    }
+    async updateCombo(workspaceId, grading, now) {
+        const prev = await this.deps.kaizenVerifiedComboRepository.getByKey(workspaceId, grading.comboKey);
+        const next = nextComboState(prev, grading, now);
+        await this.deps.kaizenVerifiedComboRepository.upsert(workspaceId, next);
+    }
+    async fail(workspaceId, grading, message) {
+        const failed = {
+            ...grading,
+            status: 'failed',
+            error: message,
+            updatedAt: this.deps.clock.now(),
+        };
+        await this.deps.kaizenGradingRepository.upsert(workspaceId, failed);
+        await this.emit(workspaceId, failed);
+    }
+    async emit(workspaceId, grading) {
+        try {
+            await this.deps.events?.kaizenGradingChanged?.(workspaceId, grading);
+        }
+        catch {
+            // Best-effort push; the persisted row is the source of truth.
+        }
+    }
+}
+/** Pull the first JSON object out of the grader's reply and coerce it to a verdict. */
+function parseVerdict(text) {
+    // Use the shared, string-literal-aware extractor (handles prose/braces around the
+    // object and fenced code blocks) rather than a naive first-`{`/last-`}` slice.
+    const parsed = extractJson(text);
+    const json = parsed && typeof parsed === 'object' && !Array.isArray(parsed)
+        ? parsed
+        : null;
+    if (!json)
+        throw new Error('Kaizen grader returned no parseable JSON verdict');
+    const rawGrade = typeof json.grade === 'number' ? json.grade : Number(json.grade);
+    if (!Number.isFinite(rawGrade))
+        throw new Error('Kaizen grader returned no numeric grade');
+    const grade = Math.min(5, Math.max(1, Math.round(rawGrade)));
+    const summary = typeof json.summary === 'string' ? json.summary : '';
+    const recommendations = Array.isArray(json.recommendations)
+        ? json.recommendations.filter((r) => typeof r === 'string' && r.trim().length > 0)
+        : [];
+    return { grade, summary, recommendations };
+}
+/** Build the grader's user prompt: the provided context + an interaction-telemetry digest. */
+function buildKaizenPrompt(grading, snapshot, calls) {
+    const parts = [];
+    parts.push(`Grade this completed agent step.\n` +
+        `Agent kind: ${grading.agentKind}\nModel: ${grading.model}\nPrompt version: ${grading.promptVersion}`);
+    if (snapshot) {
+        if (snapshot.systemPrompt)
+            parts.push(`=== SYSTEM PROMPT GIVEN TO THE AGENT ===\n${snapshot.systemPrompt}`);
+        if (snapshot.userPrompt)
+            parts.push(`=== USER PROMPT GIVEN TO THE AGENT ===\n${snapshot.userPrompt}`);
+        if (snapshot.fragments.length > 0) {
+            parts.push(`=== BEST-PRACTICE FRAGMENTS FOLDED IN ===\n` +
+                snapshot.fragments.map((f) => `- ${f.id}`).join('\n'));
+        }
+        if (snapshot.contextFiles.length > 0) {
+            parts.push(`=== CONTEXT FILES INJECTED ===\n` +
+                snapshot.contextFiles.map((f) => `- ${f.path} (${f.content.length} chars)`).join('\n'));
+        }
+    }
+    else {
+        parts.push('No provided-context snapshot was captured for this step (prompt recording may be off). ' +
+            'Grade primarily from the interaction telemetry below.');
+    }
+    parts.push(`=== INTERACTION TELEMETRY ===\n${digestCalls(calls)}`);
+    return parts.join('\n\n');
+}
+/** A compact, model-readable digest of the per-call telemetry for one step. */
+function digestCalls(calls) {
+    if (calls.length === 0)
+        return 'No LLM calls were recorded for this step.';
+    const truncated = calls.filter((c) => c.finishReason === 'length').length;
+    const errors = calls.filter((c) => !c.ok).length;
+    const promptTokens = calls.reduce((s, c) => s + c.promptTokens, 0);
+    const completionTokens = calls.reduce((s, c) => s + c.completionTokens, 0);
+    const finishReasons = summarizeCounts(calls.map((c) => c.finishReason ?? 'unknown'));
+    const lines = [
+        `Total model calls: ${calls.length}`,
+        `Truncated calls (hit output limit): ${truncated}`,
+        `Failed calls: ${errors}`,
+        `Prompt tokens (sum): ${promptTokens}`,
+        `Completion tokens (sum): ${completionTokens}`,
+        `Finish reasons: ${finishReasons}`,
+    ];
+    const last = calls[0]; // newest first
+    if (last?.responseText) {
+        lines.push(`Final visible response (truncated to 2000 chars):\n${last.responseText.slice(0, 2000)}`);
+    }
+    if (errors > 0) {
+        const messages = calls
+            .filter((c) => !c.ok && c.errorMessage)
+            .slice(0, 5)
+            .map((c) => `- ${c.errorMessage}`);
+        if (messages.length)
+            lines.push(`Error messages:\n${messages.join('\n')}`);
+    }
+    return lines.join('\n');
+}
+function summarizeCounts(values) {
+    const counts = new Map();
+    for (const v of values)
+        counts.set(v, (counts.get(v) ?? 0) + 1);
+    return [...counts.entries()].map(([k, n]) => `${k}×${n}`).join(', ');
+}
+//# sourceMappingURL=KaizenService.js.map

package/dist/modules/kaizen/KaizenService.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"KaizenService.js","sourceRoot":"","sources":["../../../src/modules/kaizen/KaizenService.ts"],"names":[],"mappings":"AAmBA,OAAO,EACL,0BAA0B,EAC1B,WAAW,EACX,cAAc,EACd,0BAA0B,GAC3B,MAAM,qBAAqB,CAAA;AAC5B,OAAO,EAAE,YAAY,EAAE,MAAM,IAAI,CAAA;AACjC,OAAO,EACL,uBAAuB,EACvB,oBAAoB,EACpB,oBAAoB,GACrB,MAAM,qBAAqB,CAAA;AAE5B,OAAO,EAAE,WAAW,EAAE,UAAU,EAAE,cAAc,EAAE,MAAM,mBAAmB,CAAA;AAE3E,wFAAwF;AACxF,MAAM,iBAAiB,GAAG,QAAQ,CAAA;AAElC,gFAAgF;AAChF,MAAM,kBAAkB,GAAG,EAAE,CAAA;AA+B7B;;;;;;;;;;GAUG;AACH,MAAM,OAAO,aAAa;IACK,IAAI;IAAjC,YAA6B,IAA+B;oBAA/B,IAAI;IAA8B,CAAC;IAEhE,qFAAqF;IACrF,IAAI,OAAO;QACT,OAAO,CAAC,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,qBAAqB,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAA;IACjG,CAAC;IAED,6EAA6E;IAE7E;;;;OAIG;IACH,KAAK,CAAC,cAAc,CAAC,WAAmB,EAAE,QAA2B;QACnE,wFAAwF;QACxF,qFAAqF;QACrF,IAAI,CAAC,IAAI,CAAC,OAAO;YAAE,OAAM;QACzB,IAAI,CAAC,CAAC,MAAM,IAAI,CAAC,aAAa,CAAC,WAAW,CAAC,CAAC;YAAE,OAAM;QACpD,KAAK,IAAI,SAAS,GAAG,CAAC,EAAE,SAAS,GAAG,QAAQ,CAAC,KAAK,CAAC,MAAM,EAAE,SAAS,EAAE,EAAE,CAAC;YACvE,MAAM,IAAI,GAAG,QAAQ,CAAC,KAAK,CAAC,SAAS,CAAC,CAAA;YACtC,IAAI,CAAC,IAAI,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK;gBAAE,SAAQ;YAC7D,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAA;YACxB,MAAM,aAAa,GAAG,oBAAoB,CAAC,IAAI,CAAC,SAAS,CAAC,CAAA;YAC1D,MAAM,QAAQ,GAAG,WAAW,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,EAAE,aAAa,CAAC,CAAA;YAClE,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,6BAA6B,CAAC,QAAQ,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAA;YAC3F,IAAI,UAAU,CAAC,KAAK,CAAC;gBAAE,SAAQ;YAC/B,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,SAAS,CAChE,WAAW,EACX,QAAQ,CAAC,EAAE,EACX,SAAS,CACV,CAAA;YACD,IAAI,QAAQ;gBAAE,SAAQ;YACtB,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAA;YACjC,MAAM,OAAO,GAAkB;gBAC7B,EAAE,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,CAAC,IAAI,CAAC,KAAK,CAAC;gBACrC,WAAW,EAAE,QAAQ,CAAC,EAAE;gBACxB,OAAO,EAAE,QAAQ,CAAC,OAAO;gBACzB,SAAS;gBACT,SAAS,EAAE,IAAI,CAAC,SAAS;gBACzB,KAAK;gBACL,aAAa;gBACb,QAAQ;gBACR,MAAM,EAAE,WAAW;gBACnB,KAAK,EAAE,IAAI;gBACX,OAAO,EAAE,EAAE;gBACX,eAAe,EAAE,EAAE;gBACnB,WAAW,EAAE,IAAI;gBACjB,KAAK,EAAE,IAAI;gBACX,SAAS,EAAE,GAAG;gBACd,SAAS,EAAE,GAAG;aACf,CAAA;YACD,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;YACpE,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACvC,CAAC;IACH,CAAC;IAED,6EAA6E;IAE7E;;;OAGG;IACH,KAAK,CAAC,UAAU,CAAC,WAAmB,EAAE,KAAa;QACjD,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,WAAW,CAAC,WAAW,EAAE,KAAK,CAAC,CAAA;QACvF,IAAI,SAAS,GAAG,CAAC,CAAA;QACjB,KAAK,MAAM,EAAE,WAAW,EAAE,OAAO,EAAE,IAAI,OAAO,EAAE,CAAC;YAC/C,sFAAsF;YACtF,qFAAqF;YACrF,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,KAAK,CAC3D,WAAW,EACX,OAAO,CAAC,EAAE,EACV,WAAW,EACX,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CACtB,CAAA;YACD,IAAI,CAAC,OAAO;gBAAE,SAAQ;YACtB,IAAI,CAAC;gBACH,MAAM,IAAI,CAAC,UAAU,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;gBAC3C,SAAS,EAAE,CAAA;YACb,CAAC;YAAC,MAAM,CAAC;gBACP,wFAAwF;YAC1F,CAAC;QACH,CAAC;QACD,OAAO,SAAS,CAAA;IAClB,CAAC;IAED,+FAA+F;IAC/F,KAAK,CAAC,UAAU,CAAC,WAAmB,EAAE,OAAsB;QAC1D,MAAM,OAAO,GAAkB;YAC7B,GAAG,OAAO;YACV,MAAM,EAAE,SAAS;YACjB,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;SACjC,CAAA;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,MAAM,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACpE,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QAErC,IAAI,CAAC,IAAI,CAAC,OAAO,EAAE,CAAC;YAClB,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,6CAA6C,CAAC,CAAA;YACpF,OAAM;QACR,CAAC;QAED,IAAI,CAAC;YACH,iFAAiF;YACjF,oFAAoF;YACpF,oFAAoF;YACpF,8EAA8E;YAC9E,2EAA2E;YAC3E,MAAM,CAAC,QAAQ,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;gBAC9C,IAAI,CAAC,eAAe,CAAC,WAAW,EAAE,OAAO,CAAC;gBAC1C,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,eAAe,CAC/C,WAAW,EACX,OAAO,CAAC,WAAW,EACnB,kBAAkB,EAClB,OAAO,CAAC,SAAS,CAClB;aACF,CAAC,CAAA;YACF,mFAAmF;YACnF,sFAAsF;YACtF,yFAAyF;YACzF,yFAAyF;YACzF,uFAAuF;YACvF,IAAI,CAAC,QAAQ,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;gBACxC,MAAM,IAAI,CAAC,IAAI,CACb,WAAW,EACX,OAAO,EACP,+FAA+F,CAChG,CAAA;gBACD,OAAM;YACR,CAAC;YACD,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,MAAM,IAAI,CAAC,YAAY,CAAC,WAAW,EAAE,OAAO,CAAC,OAAO,CAAC,CAAA;YAC/E,MAAM,KAAK,GAAG,QAAQ,CAAC,OAAO,CAAC,GAAG,CAAC,CAAA;YACnC,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;gBAChC,KAAK;gBACL,MAAM,EAAE,oBAAoB;gBAC5B,MAAM,EAAE,iBAAiB,CAAC,OAAO,EAAE,QAAQ,EAAE,SAAS,CAAC;gBACvD,WAAW,EAAE,GAAG;gBAChB,eAAe,EAAE,IAAI;gBACrB,eAAe,EAAE,uBAAuB,CAAC,EAAE,SAAS,EAAE,iBAAiB,EAAE,WAAW,EAAE,CAAC;aACxF,CAAC,CAAA;YACF,MAAM,OAAO,GAAG,YAAY,CAAC,MAAM,CAAC,IAAI,CAAC,CAAA;YACzC,MAAM,GAAG,GAAG,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE,CAAA;YACjC,MAAM,QAAQ,GAAkB;gBAC9B,GAAG,OAAO;gBACV,MAAM,EAAE,UAAU;gBAClB,KAAK,EAAE,OAAO,CAAC,KAAK;gBACpB,OAAO,EAAE,OAAO,CAAC,OAAO;gBACxB,eAAe,EAAE,OAAO,CAAC,eAAe;gBACxC,WAAW,EAAE,GAAG,GAAG,CAAC,QAAQ,IAAI,GAAG,CAAC,KAAK,EAAE;gBAC3C,KAAK,EAAE,IAAI;gBACX,SAAS,EAAE,GAAG;aACf,CAAA;YACD,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,MAAM,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAA;YACrE,MAAM,IAAI,CAAC,WAAW,CAAC,WAAW,EAAE,QAAQ,EAAE,GAAG,CAAC,CAAA;YAClD,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,QAAQ,CAAC,CAAA;QACxC,CAAC;QAAC,OAAO,CAAC,EAAE,CAAC;YACX,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,OAAO,EAAE,CAAC,YAAY,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAA;QACnF,CAAC;IACH,CAAC;IAED,4EAA4E;IAE5E,8EAA8E;IAC9E,KAAK,CAAC,WAAW,CAAC,WAAmB,EAAE,KAAK,GAAG,GAAG;QAChD,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;YAC7C,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,eAAe,CAAC,WAAW,EAAE,KAAK,CAAC;YACrE,IAAI,CAAC,IAAI,CAAC,6BAA6B,CAAC,eAAe,CAAC,WAAW,CAAC;SACrE,CAAC,CAAA;QACF,OAAO,EAAE,QAAQ,EAAE,QAAQ,EAAE,CAAA;IAC/B,CAAC;IAED,8EAA8E;IAC9E,gBAAgB,CAAC,WAAmB,EAAE,WAAmB;QACvD,OAAO,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,eAAe,CAAC,WAAW,EAAE,WAAW,CAAC,CAAA;IACpF,CAAC;IAED,4EAA4E;IAEpE,KAAK,CAAC,aAAa,CAAC,WAAmB;QAC7C,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,2BAA2B;YAAE,OAAO,0BAA0B,CAAC,aAAa,CAAA;QAC3F,MAAM,QAAQ,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,2BAA2B,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC7E,OAAO,QAAQ,EAAE,aAAa,IAAI,0BAA0B,CAAC,aAAa,CAAA;IAC5E,CAAC;IAED,yFAAyF;IACjF,WAAW,CAAC,IAAkB;QACpC,OAAO,IAAI,CAAC,KAAK,KAAK,MAAM,IAAI,CAAC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,IAAI,CAAC,KAAK,CAAA;IAC/D,CAAC;IAEO,KAAK,CAAC,eAAe,CAC3B,WAAmB,EACnB,OAAsB;QAEtB,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,yBAAyB,CAAC,eAAe,CACzE,WAAW,EACX,OAAO,CAAC,WAAW,CACpB,CAAA;QACD,OAAO,CACL,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,CAAC,SAAS,CAAC;YACxD,SAAS,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,CAAC,SAAS,CAAC;YACxD,IAAI,CACL,CAAA;IACH,CAAC;IAEO,KAAK,CAAC,YAAY,CACxB,WAAmB,EACnB,OAAe;QAEf,MAAM,QAAQ,GAAG,MAAM,0BAA0B,CAAC,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,CAAA;QACzE,MAAM,GAAG,GAAG,MAAM,IAAI,CAAC,QAAQ,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACrD,IAAI,CAAC,QAAQ,IAAI,CAAC,GAAG;YAAE,MAAM,IAAI,KAAK,CAAC,6CAA6C,CAAC,CAAA;QACrF,OAAO,EAAE,QAAQ,EAAE,GAAG,EAAE,CAAA;IAC1B,CAAC;IAED,2EAA2E;IACnE,KAAK,CAAC,QAAQ,CAAC,WAAmB,EAAE,OAAe;QACzD,MAAM,QAAQ,GAAG,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAA;QACnC,MAAM,OAAO,GAAG,CAAC,GAAa,EAAY,EAAE,CAAC,cAAc,CAAC,GAAG,EAAE,QAAQ,IAAI,GAAG,CAAC,CAAA;QACjF,MAAM,KAAK,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,eAAe,CAAC,GAAG,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACvE,MAAM,SAAS,GAAG,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,CAAC,KAAK,EAAE,OAAO,CAAC,CAAA;QAC/D,IAAI,SAAS;YAAE,OAAO,OAAO,CAAC,SAAS,CAAC,CAAA;QACxC,MAAM,SAAS,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,4BAA4B,EAAE,CAC9D,WAAW,EACX,iBAAiB,EACjB,KAAK,EAAE,aAAa,CACrB,CAAA;QACD,MAAM,WAAW,GAAG,IAAI,CAAC,IAAI,CAAC,iBAAiB,EAAE,CAAC,SAAS,CAAC,CAAA;QAC5D,IAAI,WAAW;YAAE,OAAO,OAAO,CAAC,WAAW,CAAC,CAAA;QAC5C,OAAO,QAAQ,CAAA;IACjB,CAAC;IAEO,KAAK,CAAC,WAAW,CACvB,WAAmB,EACnB,OAAsB,EACtB,GAAW;QAEX,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAAC,6BAA6B,CAAC,QAAQ,CACjE,WAAW,EACX,OAAO,CAAC,QAAQ,CACjB,CAAA;QACD,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,EAAE,OAAO,EAAE,GAAG,CAAC,CAAA;QAC/C,MAAM,IAAI,CAAC,IAAI,CAAC,6BAA6B,CAAC,MAAM,CAAC,WAAW,EAAE,IAAI,CAAC,CAAA;IACzE,CAAC;IAEO,KAAK,CAAC,IAAI,CAAC,WAAmB,EAAE,OAAsB,EAAE,OAAe;QAC7E,MAAM,MAAM,GAAkB;YAC5B,GAAG,OAAO;YACV,MAAM,EAAE,QAAQ;YAChB,KAAK,EAAE,OAAO;YACd,SAAS,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC,GAAG,EAAE;SACjC,CAAA;QACD,MAAM,IAAI,CAAC,IAAI,CAAC,uBAAuB,CAAC,MAAM,CAAC,WAAW,EAAE,MAAM,CAAC,CAAA;QACnE,MAAM,IAAI,CAAC,IAAI,CAAC,WAAW,EAAE,MAAM,CAAC,CAAA;IACtC,CAAC;IAEO,KAAK,CAAC,IAAI,CAAC,WAAmB,EAAE,OAAsB;QAC5D,IAAI,CAAC;YACH,MAAM,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,oBAAoB,EAAE,CAAC,WAAW,EAAE,OAAO,CAAC,CAAA;QACtE,CAAC;QAAC,MAAM,CAAC;YACP,8DAA8D;QAChE,CAAC;IACH,CAAC;CACF;AASD,uFAAuF;AACvF,SAAS,YAAY,CAAC,IAAY;IAChC,kFAAkF;IAClF,+EAA+E;IAC/E,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,CAAC,CAAA;IAChC,MAAM,IAAI,GACR,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;QAC5D,CAAC,CAAE,MAAkC;QACrC,CAAC,CAAC,IAAI,CAAA;IACV,IAAI,CAAC,IAAI;QAAE,MAAM,IAAI,KAAK,CAAC,kDAAkD,CAAC,CAAA;IAC9E,MAAM,QAAQ,GAAG,OAAO,IAAI,CAAC,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAA;IACjF,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC;QAAE,MAAM,IAAI,KAAK,CAAC,yCAAyC,CAAC,CAAA;IAC1F,MAAM,KAAK,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,CAAC,CAAA;IAC5D,MAAM,OAAO,GAAG,OAAO,IAAI,CAAC,OAAO,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAA;IACpE,MAAM,eAAe,GAAG,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC,eAAe,CAAC;QACzD,CAAC,CAAC,IAAI,CAAC,eAAe,CAAC,MAAM,CAAC,CAAC,CAAC,EAAe,EAAE,CAAC,OAAO,CAAC,KAAK,QAAQ,IAAI,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAAC;QAC/F,CAAC,CAAC,EAAE,CAAA;IACN,OAAO,EAAE,KAAK,EAAE,OAAO,EAAE,eAAe,EAAE,CAAA;AAC5C,CAAC;AAED,8FAA8F;AAC9F,SAAS,iBAAiB,CACxB,OAAsB,EACtB,QAAqC,EACrC,KAAsB;IAEtB,MAAM,KAAK,GAAa,EAAE,CAAA;IAC1B,KAAK,CAAC,IAAI,CACR,oCAAoC;QAClC,eAAe,OAAO,CAAC,SAAS,YAAY,OAAO,CAAC,KAAK,qBAAqB,OAAO,CAAC,aAAa,EAAE,CACxG,CAAA;IAED,IAAI,QAAQ,EAAE,CAAC;QACb,IAAI,QAAQ,CAAC,YAAY;YACvB,KAAK,CAAC,IAAI,CAAC,6CAA6C,QAAQ,CAAC,YAAY,EAAE,CAAC,CAAA;QAClF,IAAI,QAAQ,CAAC,UAAU;YACrB,KAAK,CAAC,IAAI,CAAC,2CAA2C,QAAQ,CAAC,UAAU,EAAE,CAAC,CAAA;QAC9E,IAAI,QAAQ,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YAClC,KAAK,CAAC,IAAI,CACR,6CAA6C;gBAC3C,QAAQ,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CACxD,CAAA;QACH,CAAC;QACD,IAAI,QAAQ,CAAC,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;YACrC,KAAK,CAAC,IAAI,CACR,kCAAkC;gBAChC,QAAQ,CAAC,YAAY,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,IAAI,KAAK,CAAC,CAAC,OAAO,CAAC,MAAM,SAAS,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CACzF,CAAA;QACH,CAAC;IACH,CAAC;SAAM,CAAC;QACN,KAAK,CAAC,IAAI,CACR,yFAAyF;YACvF,uDAAuD,CAC1D,CAAA;IACH,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,kCAAkC,WAAW,CAAC,KAAK,CAAC,EAAE,CAAC,CAAA;IAClE,OAAO,KAAK,CAAC,IAAI,CAAC,MAAM,CAAC,CAAA;AAC3B,CAAC;AAED,+EAA+E;AAC/E,SAAS,WAAW,CAAC,KAAsB;IACzC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,2CAA2C,CAAA;IAC1E,MAAM,SAAS,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,KAAK,QAAQ,CAAC,CAAC,MAAM,CAAA;IACzE,MAAM,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,MAAM,CAAA;IAChD,MAAM,YAAY,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,YAAY,EAAE,CAAC,CAAC,CAAA;IAClE,MAAM,gBAAgB,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,gBAAgB,EAAE,CAAC,CAAC,CAAA;IAC1E,MAAM,aAAa,GAAG,eAAe,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,YAAY,IAAI,SAAS,CAAC,CAAC,CAAA;IACpF,MAAM,KAAK,GAAG;QACZ,sBAAsB,KAAK,CAAC,MAAM,EAAE;QACpC,uCAAuC,SAAS,EAAE;QAClD,iBAAiB,MAAM,EAAE;QACzB,wBAAwB,YAAY,EAAE;QACtC,4BAA4B,gBAAgB,EAAE;QAC9C,mBAAmB,aAAa,EAAE;KACnC,CAAA;IACD,MAAM,IAAI,GAAG,KAAK,CAAC,CAAC,CAAC,CAAA,CAAC,eAAe;IACrC,IAAI,IAAI,EAAE,YAAY,EAAE,CAAC;QACvB,KAAK,CAAC,IAAI,CACR,sDAAsD,IAAI,CAAC,YAAY,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC,EAAE,CACzF,CAAA;IACH,CAAC;IACD,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;QACf,MAAM,QAAQ,GAAG,KAAK;aACnB,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,CAAC,EAAE,IAAI,CAAC,CAAC,YAAY,CAAC;aACtC,KAAK,CAAC,CAAC,EAAE,CAAC,CAAC;aACX,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,KAAK,CAAC,CAAC,YAAY,EAAE,CAAC,CAAA;QACpC,IAAI,QAAQ,CAAC,MAAM;YAAE,KAAK,CAAC,IAAI,CAAC,oBAAoB,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC,CAAA;IAC5E,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACzB,CAAC;AAED,SAAS,eAAe,CAAC,MAAgB;IACvC,MAAM,MAAM,GAAG,IAAI,GAAG,EAAkB,CAAA;IACxC,KAAK,MAAM,CAAC,IAAI,MAAM;QAAE,MAAM,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,GAAG,CAAC,CAAC,CAAA;IAC/D,OAAO,CAAC,GAAG,MAAM,CAAC,OAAO,EAAE,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,GAAG,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAA;AACtE,CAAC"}

package/dist/modules/kaizen/kaizen.logic.d.ts

@@ -0,0 +1,29 @@
+import type { KaizenGrading, KaizenVerifiedCombo } from '@cat-factory/contracts';
+/** The grade (on the 1..5 scale) at or above which a grading counts as "high". */
+export declare const HIGH_GRADE = 4;
+/** Consecutive high-grade-with-no-recommendations gradings needed to verify a combo. */
+export declare const VERIFICATION_STREAK = 5;
+/** `agentKind|model|promptVersion` — the key a verified combo is tracked under. */
+export declare function comboKeyFor(agentKind: string, model: string, promptVersion: number): string;
+/**
+ * Whether a completed grading is a "high grade": a strong score AND no recommendations.
+ * The no-recommendations gate is the real quality signal — the grader found nothing to
+ * improve — so a grading with ANY recommendation does not advance the streak regardless of
+ * its number. Requiring a *flawless* 5 every time made the streak almost never converge
+ * (so the "stop grading a verified combo" optimization never engaged and good combos were
+ * re-graded on every run forever); a 4-or-5 with nothing to improve is the intended bar.
+ */
+export declare function isHighGrade(grade: number | null, recommendations: readonly string[]): boolean;
+/**
+ * The combo's next verification state after folding in one completed grading.
+ * A high grade increments the streak (and flips `verified` once it reaches
+ * {@link VERIFICATION_STREAK}); anything else resets the streak to 0. Once verified,
+ * a combo stays verified (the engine stops scheduling gradings for it, so no further
+ * grading should arrive — but if one does, re-evaluate from the same rules).
+ *
+ * `prev` is null for a combo's first-ever grading.
+ */
+export declare function nextComboState(prev: KaizenVerifiedCombo | null, grading: Pick<KaizenGrading, 'comboKey' | 'agentKind' | 'model' | 'promptVersion' | 'grade' | 'recommendations'>, now: number): KaizenVerifiedCombo;
+/** Whether a combo is verified and should therefore NOT be graded again. */
+export declare function isVerified(combo: KaizenVerifiedCombo | null): boolean;
+//# sourceMappingURL=kaizen.logic.d.ts.map

package/dist/modules/kaizen/kaizen.logic.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"kaizen.logic.d.ts","sourceRoot":"","sources":["../../../src/modules/kaizen/kaizen.logic.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,mBAAmB,EAAE,MAAM,wBAAwB,CAAA;AAKhF,kFAAkF;AAClF,eAAO,MAAM,UAAU,IAAI,CAAA;AAE3B,wFAAwF;AACxF,eAAO,MAAM,mBAAmB,IAAI,CAAA;AAEpC,mFAAmF;AACnF,wBAAgB,WAAW,CAAC,SAAS,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,GAAG,MAAM,CAE3F;AAED;;;;;;;GAOG;AACH,wBAAgB,WAAW,CAAC,KAAK,EAAE,MAAM,GAAG,IAAI,EAAE,eAAe,EAAE,SAAS,MAAM,EAAE,GAAG,OAAO,CAE7F;AAED;;;;;;;;GAQG;AACH,wBAAgB,cAAc,CAC5B,IAAI,EAAE,mBAAmB,GAAG,IAAI,EAChC,OAAO,EAAE,IAAI,CACX,aAAa,EACb,UAAU,GAAG,WAAW,GAAG,OAAO,GAAG,eAAe,GAAG,OAAO,GAAG,iBAAiB,CACnF,EACD,GAAG,EAAE,MAAM,GACV,mBAAmB,CAgBrB;AAED,4EAA4E;AAC5E,wBAAgB,UAAU,CAAC,KAAK,EAAE,mBAAmB,GAAG,IAAI,GAAG,OAAO,CAErE"}

package/dist/modules/kaizen/kaizen.logic.js

@@ -0,0 +1,52 @@
+// Pure Kaizen decision logic — no I/O, no clock, no LLM. Kept separate from the
+// service so the streak/verification rules are unit-testable in isolation.
+/** The grade (on the 1..5 scale) at or above which a grading counts as "high". */
+export const HIGH_GRADE = 4;
+/** Consecutive high-grade-with-no-recommendations gradings needed to verify a combo. */
+export const VERIFICATION_STREAK = 5;
+/** `agentKind|model|promptVersion` — the key a verified combo is tracked under. */
+export function comboKeyFor(agentKind, model, promptVersion) {
+    return `${agentKind}|${model}|${promptVersion}`;
+}
+/**
+ * Whether a completed grading is a "high grade": a strong score AND no recommendations.
+ * The no-recommendations gate is the real quality signal — the grader found nothing to
+ * improve — so a grading with ANY recommendation does not advance the streak regardless of
+ * its number. Requiring a *flawless* 5 every time made the streak almost never converge
+ * (so the "stop grading a verified combo" optimization never engaged and good combos were
+ * re-graded on every run forever); a 4-or-5 with nothing to improve is the intended bar.
+ */
+export function isHighGrade(grade, recommendations) {
+    return grade != null && grade >= HIGH_GRADE && recommendations.length === 0;
+}
+/**
+ * The combo's next verification state after folding in one completed grading.
+ * A high grade increments the streak (and flips `verified` once it reaches
+ * {@link VERIFICATION_STREAK}); anything else resets the streak to 0. Once verified,
+ * a combo stays verified (the engine stops scheduling gradings for it, so no further
+ * grading should arrive — but if one does, re-evaluate from the same rules).
+ *
+ * `prev` is null for a combo's first-ever grading.
+ */
+export function nextComboState(prev, grading, now) {
+    const high = isHighGrade(grading.grade, grading.recommendations);
+    const priorStreak = prev?.consecutiveHighGrades ?? 0;
+    const consecutiveHighGrades = high ? priorStreak + 1 : 0;
+    const verified = consecutiveHighGrades >= VERIFICATION_STREAK;
+    const verifiedAt = verified ? (prev?.verifiedAt ?? now) : null;
+    return {
+        comboKey: grading.comboKey,
+        agentKind: grading.agentKind,
+        model: grading.model,
+        promptVersion: grading.promptVersion,
+        consecutiveHighGrades,
+        verified,
+        verifiedAt,
+        updatedAt: now,
+    };
+}
+/** Whether a combo is verified and should therefore NOT be graded again. */
+export function isVerified(combo) {
+    return combo?.verified === true;
+}
+//# sourceMappingURL=kaizen.logic.js.map

package/dist/modules/kaizen/kaizen.logic.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"kaizen.logic.js","sourceRoot":"","sources":["../../../src/modules/kaizen/kaizen.logic.ts"],"names":[],"mappings":"AAEA,gFAAgF;AAChF,2EAA2E;AAE3E,kFAAkF;AAClF,MAAM,CAAC,MAAM,UAAU,GAAG,CAAC,CAAA;AAE3B,wFAAwF;AACxF,MAAM,CAAC,MAAM,mBAAmB,GAAG,CAAC,CAAA;AAEpC,mFAAmF;AACnF,MAAM,UAAU,WAAW,CAAC,SAAiB,EAAE,KAAa,EAAE,aAAqB;IACjF,OAAO,GAAG,SAAS,IAAI,KAAK,IAAI,aAAa,EAAE,CAAA;AACjD,CAAC;AAED;;;;;;;GAOG;AACH,MAAM,UAAU,WAAW,CAAC,KAAoB,EAAE,eAAkC;IAClF,OAAO,KAAK,IAAI,IAAI,IAAI,KAAK,IAAI,UAAU,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,CAAA;AAC7E,CAAC;AAED;;;;;;;;GAQG;AACH,MAAM,UAAU,cAAc,CAC5B,IAAgC,EAChC,OAGC,EACD,GAAW;IAEX,MAAM,IAAI,GAAG,WAAW,CAAC,OAAO,CAAC,KAAK,EAAE,OAAO,CAAC,eAAe,CAAC,CAAA;IAChE,MAAM,WAAW,GAAG,IAAI,EAAE,qBAAqB,IAAI,CAAC,CAAA;IACpD,MAAM,qBAAqB,GAAG,IAAI,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAA;IACxD,MAAM,QAAQ,GAAG,qBAAqB,IAAI,mBAAmB,CAAA;IAC7D,MAAM,UAAU,GAAG,QAAQ,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,UAAU,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAA;IAC9D,OAAO;QACL,QAAQ,EAAE,OAAO,CAAC,QAAQ;QAC1B,SAAS,EAAE,OAAO,CAAC,SAAS;QAC5B,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,aAAa,EAAE,OAAO,CAAC,aAAa;QACpC,qBAAqB;QACrB,QAAQ;QACR,UAAU;QACV,SAAS,EAAE,GAAG;KACf,CAAA;AACH,CAAC;AAED,4EAA4E;AAC5E,MAAM,UAAU,UAAU,CAAC,KAAiC;IAC1D,OAAO,KAAK,EAAE,QAAQ,KAAK,IAAI,CAAA;AACjC,CAAC"}

package/dist/modules/settings/WorkspaceSettingsService.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"WorkspaceSettingsService.d.ts","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,4BAA4B,EAC5B,mBAAmB,EACnB,iBAAiB,EACjB,2BAA2B,EAC5B,MAAM,qBAAqB,CAAA;AAG5B,MAAM,WAAW,oCAAoC;IACnD,2BAA2B,EAAE,2BAA2B,CAAA;IACxD,mBAAmB,EAAE,mBAAmB,CAAA;CACzC;AAED;;;;;;GAMG;AACH,qBAAa,wBAAwB;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IAEzD,YAAY,IAAI,EAAE,oCAAoC,EAGrD;IAED,0FAA0F;IACpF,GAAG,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAEzD;IAED,kEAAkE;IAC5D,MAAM,CACV,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,4BAA4B,GAClC,OAAO,CAAC,iBAAiB,CAAC,CA8B5B;CACF"}
+{"version":3,"file":"WorkspaceSettingsService.d.ts","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EACV,4BAA4B,EAC5B,mBAAmB,EACnB,iBAAiB,EACjB,2BAA2B,EAC5B,MAAM,qBAAqB,CAAA;AAG5B,MAAM,WAAW,oCAAoC;IACnD,2BAA2B,EAAE,2BAA2B,CAAA;IACxD,mBAAmB,EAAE,mBAAmB,CAAA;CACzC;AAED;;;;;;GAMG;AACH,qBAAa,wBAAwB;IACnC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,mBAAmB,CAAqB;IAEzD,YAAY,IAAI,EAAE,oCAAoC,EAGrD;IAED,0FAA0F;IACpF,GAAG,CAAC,WAAW,EAAE,MAAM,GAAG,OAAO,CAAC,iBAAiB,CAAC,CAEzD;IAED,kEAAkE;IAC5D,MAAM,CACV,WAAW,EAAE,MAAM,EACnB,KAAK,EAAE,4BAA4B,GAClC,OAAO,CAAC,iBAAiB,CAAC,CA+B5B;CACF"}

package/dist/modules/settings/WorkspaceSettingsService.js

@@ -27,6 +27,7 @@ export class WorkspaceSettingsService {
             taskLimitShared: patch.taskLimitShared !== undefined ? patch.taskLimitShared : current.taskLimitShared,
             taskLimitPerType: patch.taskLimitPerType !== undefined ? patch.taskLimitPerType : current.taskLimitPerType,
             storeAgentContext: patch.storeAgentContext ?? current.storeAgentContext,
+            kaizenEnabled: patch.kaizenEnabled ?? current.kaizenEnabled,
             spendCurrency: patch.spendCurrency !== undefined ? patch.spendCurrency : current.spendCurrency,
             spendMonthlyLimit: patch.spendMonthlyLimit !== undefined ? patch.spendMonthlyLimit : current.spendMonthlyLimit,
         };

package/dist/modules/settings/WorkspaceSettingsService.js.map

@@ -1 +1 @@
-{"version":3,"file":"WorkspaceSettingsService.js","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,0BAA0B,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAA;AAOlF;;;;;;GAMG;AACH,MAAM,OAAO,wBAAwB;IAClB,QAAQ,CAA6B;IACrC,mBAAmB,CAAqB;IAEzD,YAAY,IAA0C;QACpD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,2BAA2B,CAAA;QAChD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,CAAA;IACrD,CAAC;IAED,0FAA0F;IAC1F,KAAK,CAAC,GAAG,CAAC,WAAmB;QAC3B,OAAO,CAAC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,GAAG,0BAA0B,EAAE,CAAA;IACpF,CAAC;IAED,kEAAkE;IAClE,KAAK,CAAC,MAAM,CACV,WAAmB,EACnB,KAAmC;QAEnC,MAAM,gBAAgB,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAC7D,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC3C,MAAM,IAAI,GAAsB;YAC9B,wBAAwB,EAAE,KAAK,CAAC,wBAAwB,IAAI,OAAO,CAAC,wBAAwB;YAC5F,aAAa,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,aAAa;YAC3D,eAAe,EACb,KAAK,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe;YACvF,gBAAgB,EACd,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB;YAC1F,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,IAAI,OAAO,CAAC,iBAAiB;YACvE,aAAa,EACX,KAAK,CAAC,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa;YACjF,iBAAiB,EACf,KAAK,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,OAAO,CAAC,iBAAiB;SAC9F,CAAA;QACD,qFAAqF;QACrF,0CAA0C;QAC1C,IAAI,IAAI,CAAC,aAAa,KAAK,KAAK,EAAE,CAAC;YACjC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;QAC9B,CAAC;aAAM,IAAI,IAAI,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;YAC3C,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;YAC5B,IAAI,IAAI,CAAC,eAAe,IAAI,IAAI;gBAAE,IAAI,CAAC,eAAe,GAAG,CAAC,CAAA;QAC5D,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAAE,IAAI,CAAC,gBAAgB,GAAG,EAAE,CAAA;QAC/D,CAAC;QACD,MAAM,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,IAAI,CAAC,CAAA;QAC7C,OAAO,IAAI,CAAA;IACb,CAAC;CACF"}
+{"version":3,"file":"WorkspaceSettingsService.js","sourceRoot":"","sources":["../../../src/modules/settings/WorkspaceSettingsService.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,0BAA0B,EAAE,gBAAgB,EAAE,MAAM,qBAAqB,CAAA;AAOlF;;;;;;GAMG;AACH,MAAM,OAAO,wBAAwB;IAClB,QAAQ,CAA6B;IACrC,mBAAmB,CAAqB;IAEzD,YAAY,IAA0C;QACpD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,2BAA2B,CAAA;QAChD,IAAI,CAAC,mBAAmB,GAAG,IAAI,CAAC,mBAAmB,CAAA;IACrD,CAAC;IAED,0FAA0F;IAC1F,KAAK,CAAC,GAAG,CAAC,WAAmB;QAC3B,OAAO,CAAC,MAAM,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,WAAW,CAAC,CAAC,IAAI,EAAE,GAAG,0BAA0B,EAAE,CAAA;IACpF,CAAC;IAED,kEAAkE;IAClE,KAAK,CAAC,MAAM,CACV,WAAmB,EACnB,KAAmC;QAEnC,MAAM,gBAAgB,CAAC,IAAI,CAAC,mBAAmB,EAAE,WAAW,CAAC,CAAA;QAC7D,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,GAAG,CAAC,WAAW,CAAC,CAAA;QAC3C,MAAM,IAAI,GAAsB;YAC9B,wBAAwB,EAAE,KAAK,CAAC,wBAAwB,IAAI,OAAO,CAAC,wBAAwB;YAC5F,aAAa,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,aAAa;YAC3D,eAAe,EACb,KAAK,CAAC,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,eAAe,CAAC,CAAC,CAAC,OAAO,CAAC,eAAe;YACvF,gBAAgB,EACd,KAAK,CAAC,gBAAgB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,gBAAgB,CAAC,CAAC,CAAC,OAAO,CAAC,gBAAgB;YAC1F,iBAAiB,EAAE,KAAK,CAAC,iBAAiB,IAAI,OAAO,CAAC,iBAAiB;YACvE,aAAa,EAAE,KAAK,CAAC,aAAa,IAAI,OAAO,CAAC,aAAa;YAC3D,aAAa,EACX,KAAK,CAAC,aAAa,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,aAAa,CAAC,CAAC,CAAC,OAAO,CAAC,aAAa;YACjF,iBAAiB,EACf,KAAK,CAAC,iBAAiB,KAAK,SAAS,CAAC,CAAC,CAAC,KAAK,CAAC,iBAAiB,CAAC,CAAC,CAAC,OAAO,CAAC,iBAAiB;SAC9F,CAAA;QACD,qFAAqF;QACrF,0CAA0C;QAC1C,IAAI,IAAI,CAAC,aAAa,KAAK,KAAK,EAAE,CAAC;YACjC,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;QAC9B,CAAC;aAAM,IAAI,IAAI,CAAC,aAAa,KAAK,QAAQ,EAAE,CAAC;YAC3C,IAAI,CAAC,gBAAgB,GAAG,IAAI,CAAA;YAC5B,IAAI,IAAI,CAAC,eAAe,IAAI,IAAI;gBAAE,IAAI,CAAC,eAAe,GAAG,CAAC,CAAA;QAC5D,CAAC;aAAM,CAAC;YACN,IAAI,CAAC,eAAe,GAAG,IAAI,CAAA;YAC3B,IAAI,IAAI,CAAC,gBAAgB,IAAI,IAAI;gBAAE,IAAI,CAAC,gBAAgB,GAAG,EAAE,CAAA;QAC/D,CAAC;QACD,MAAM,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,WAAW,EAAE,IAAI,CAAC,CAAA;QAC7C,OAAO,IAAI,CAAA;IACb,CAAC;CACF"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@cat-factory/orchestration",
-  "version": "0.24.2",
+  "version": "0.25.0",
   "description": "Delivery-workflow engine for the Agent Architecture Board (execution, bootstrap, pipelines, board, boardScan, requirements, and composition root).",
   "repository": {
     "type": "git",
@@ -25,14 +25,14 @@
   },
   "dependencies": {
     "ai": "^6.0.209",
-    "@cat-factory/agents": "0.15.2",
-    "@cat-factory/contracts": "0.29.0",
-    "@cat-factory/integrations": "0.21.4",
-    "@cat-factory/kernel": "0.32.0",
-    "@cat-factory/prompt-fragments": "0.7.27",
-    "@cat-factory/sandbox": "0.8.5",
-    "@cat-factory/spend": "0.10.0",
-    "@cat-factory/workspaces": "0.7.39"
+    "@cat-factory/agents": "0.16.0",
+    "@cat-factory/contracts": "0.30.0",
+    "@cat-factory/integrations": "0.21.5",
+    "@cat-factory/kernel": "0.33.0",
+    "@cat-factory/prompt-fragments": "0.7.28",
+    "@cat-factory/sandbox": "0.8.6",
+    "@cat-factory/spend": "0.10.1",
+    "@cat-factory/workspaces": "0.7.40"
   },
   "devDependencies": {
     "typescript": "7.0.1-rc",