@hmharness/evolution 0.14.19 → 0.14.20

package/dist/index.d.ts

@@ -19,3 +19,4 @@ export * from './knowledge.ts';
 export * from './radar.ts';
 export * from './external.ts';
 export * from './knowledge-graph.ts';
+export * from './router-evolution.ts';

package/dist/index.js

@@ -19,3 +19,4 @@ export * from "./knowledge.js";
 export * from "./radar.js";
 export * from "./external.js";
 export * from "./knowledge-graph.js";
+export * from "./router-evolution.js";

package/dist/router-evolution.d.ts

@@ -0,0 +1,60 @@
+/**
+ * @hmharness/evolution - Router Evolution (P2-05)
+ *
+ * The audit called for: "让 model routing 本身成为可实验、可回滚策略"
+ *
+ * Makes routing decisions experimentable: a routing policy can be registered
+ * as a candidate, A/B tested against the current policy, and promoted or
+ * rolled back based on SPRT results.
+ */
+import type { RouteHistoryEntry } from '@hmharness/kernel';
+export interface RoutingPolicy {
+    id: string;
+    name: string;
+    description: string;
+    /** the routing function */
+    decide: (history: RouteHistoryEntry[], taskType: string, candidates: string[]) => string;
+}
+export interface RoutingExperiment {
+    id: string;
+    /** the challenger policy */
+    challenger: RoutingPolicy;
+    /** incumbent policy (or 'default' for current behavior) */
+    incumbent: string;
+    /** control arm results */
+    controlSuccesses: number;
+    controlTotal: number;
+    /** treatment arm results */
+    treatmentSuccesses: number;
+    treatmentTotal: number;
+    status: 'running' | 'promoted' | 'rolled-back' | 'inconclusive';
+    startedAt: string;
+}
+/**
+ * Evaluate a routing experiment using simple SPRT logic.
+ * Pure - testable.
+ */
+export declare function evaluateExperiment(exp: RoutingExperiment): {
+    decision: 'continue' | 'promote' | 'rollback';
+    reason: string;
+};
+/**
+ * The routing policy registry - manages candidate and active policies.
+ */
+export declare class RoutingPolicyRegistry {
+    private policies;
+    private activePolicyId;
+    register(policy: RoutingPolicy): {
+        ok: boolean;
+        reason?: string;
+    };
+    get(id: string): RoutingPolicy | undefined;
+    list(): RoutingPolicy[];
+    /** Get the currently active policy */
+    getActive(): RoutingPolicy | undefined;
+    /** Promote a policy to active (rollback = set back to 'default') */
+    promote(policyId: string): boolean;
+    /** Rollback to default policy */
+    rollback(): void;
+    get activeId(): string;
+}

package/dist/router-evolution.js

@@ -0,0 +1,67 @@
+/**
+ * @hmharness/evolution - Router Evolution (P2-05)
+ *
+ * The audit called for: "让 model routing 本身成为可实验、可回滚策略"
+ *
+ * Makes routing decisions experimentable: a routing policy can be registered
+ * as a candidate, A/B tested against the current policy, and promoted or
+ * rolled back based on SPRT results.
+ */
+/**
+ * Evaluate a routing experiment using simple SPRT logic.
+ * Pure - testable.
+ */
+export function evaluateExperiment(exp) {
+    const minN = 20; // minimum samples per arm
+    if (exp.controlTotal < minN || exp.treatmentTotal < minN) {
+        return { decision: 'continue', reason: `insufficient samples (control ${exp.controlTotal}/${minN}, treatment ${exp.treatmentTotal}/${minN})` };
+    }
+    const controlRate = exp.controlSuccesses / exp.controlTotal;
+    const treatmentRate = exp.treatmentSuccesses / exp.treatmentTotal;
+    const diff = treatmentRate - controlRate;
+    // promote if treatment is 5pp better, rollback if 5pp worse
+    if (diff >= 0.05) {
+        return { decision: 'promote', reason: `treatment ${treatmentRate.toFixed(2)} vs control ${controlRate.toFixed(2)} (+${(diff * 100).toFixed(1)}pp)` };
+    }
+    if (diff <= -0.05) {
+        return { decision: 'rollback', reason: `treatment ${treatmentRate.toFixed(2)} vs control ${controlRate.toFixed(2)} (${(diff * 100).toFixed(1)}pp)` };
+    }
+    return { decision: 'continue', reason: `diff ${(diff * 100).toFixed(1)}pp within noise band` };
+}
+/**
+ * The routing policy registry - manages candidate and active policies.
+ */
+export class RoutingPolicyRegistry {
+    policies = new Map();
+    activePolicyId = 'default';
+    register(policy) {
+        if (this.policies.has(policy.id))
+            return { ok: false, reason: `policy ${policy.id} already registered` };
+        this.policies.set(policy.id, policy);
+        return { ok: true };
+    }
+    get(id) {
+        return this.policies.get(id);
+    }
+    list() {
+        return [...this.policies.values()];
+    }
+    /** Get the currently active policy */
+    getActive() {
+        return this.policies.get(this.activePolicyId);
+    }
+    /** Promote a policy to active (rollback = set back to 'default') */
+    promote(policyId) {
+        if (!this.policies.has(policyId))
+            return false;
+        this.activePolicyId = policyId;
+        return true;
+    }
+    /** Rollback to default policy */
+    rollback() {
+        this.activePolicyId = 'default';
+    }
+    get activeId() {
+        return this.activePolicyId;
+    }
+}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@hmharness/evolution",
-  "version": "0.14.19",
+  "version": "0.14.20",
   "description": "hmharness evolution subsystem: persistent memory, insight capture, skill library, and the bench that gives evolution its fitness signal. First-class kernel citizen, not a plugin.",
   "type": "module",
   "main": "dist/index.js",