fix(cleanup): remove dead code, refresh comments, add migration test, update docs

assistant/ARCHITECTURE.md

@@ -2094,7 +2094,7 @@ Connected channels are resolved at signal emission time: vellum is always includ
 
 **Audit trail (SQLite):** `notification_events` → `notification_decisions` (with `conversationActions` in validation results) → `notification_deliveries` (with `conversation_id`, `message_id`, `conversation_strategy`, `conversation_action`, `conversation_target_id`, `conversation_fallback_used`)
 
-**Configuration:** `notifications.decisionModelIntent` in `config.json`.
+**Configuration:** `llm.callSites.notificationDecision` (decision engine) and `llm.callSites.preferenceExtraction` (preference extractor) in `config.json`. Both fall back to `llm.default` when unset.
 
 ---
 

assistant/src/__tests__/call-controller.test.ts

@@ -35,10 +35,9 @@ mock.module("../config/loader.js", () => {
       silenceTimeoutSeconds: 30,
       disclosure: { enabled: false, text: "" },
       safety: { denyCategories: [] },
-      model: undefined,
     },
     memory: { enabled: false },
-    notifications: { decisionModelIntent: "latency-optimized" },
+    notifications: {},
     services: {
       tts: {
         mode: "your-own" as const,

assistant/src/__tests__/conversation-process-callsite.test.ts

@@ -1,11 +1,14 @@
 /**
- * Verify that `callSite` threads from `Conversation.processMessage` options
- * all the way down to the per-call provider config, and that user-initiated
- * turns default to `'mainAgent'` when no caller-supplied `callSite` is set.
+ * Verifies that `callSite` threads from `Conversation.processMessage`
+ * options all the way down to the per-call provider config, and that
+ * user-initiated turns default to `'mainAgent'` when no caller-supplied
+ * `callSite` is set.
  *
  * The test mocks `AgentLoop.run()` so it can capture the `callSite` parameter
  * the conversation passes after `processMessage` runs the slash-resolver and
- * runtime-injection pipeline.
+ * runtime-injection pipeline. Adapter callers (heartbeat, filing, scheduler)
+ * pass an explicit `callSite` so `RetryProvider` resolves their per-call
+ * config from `llm.callSites.<id>`.
  */
 import { describe, expect, mock, test } from "bun:test";
 

assistant/src/__tests__/heartbeat-service.test.ts

@@ -9,7 +9,6 @@ let mockConfig = {
   heartbeat: {
     enabled: true,
     intervalMs: 60_000,
-    speed: "standard" as "standard" | "fast",
     activeHoursStart: undefined as number | undefined,
     activeHoursEnd: undefined as number | undefined,
   },
@@ -185,7 +184,6 @@ describe("HeartbeatService", () => {
       heartbeat: {
         enabled: true,
         intervalMs: 60_000,
-        speed: "standard",
         activeHoursStart: undefined,
         activeHoursEnd: undefined,
       },
@@ -547,17 +545,18 @@ describe("HeartbeatService", () => {
     });
   });
 
-  test("callSite is passed regardless of legacy heartbeat.speed value", async () => {
-    // The legacy `heartbeat.speed` field is no longer read by the heartbeat
-    // service — every run unconditionally passes `callSite: 'heartbeatAgent'`
-    // and the resolver maps that to whatever `llm.callSites.heartbeatAgent`
-    // (or the default) configures. PR 19 will remove the schema field.
-    mockConfig.heartbeat.speed = "fast";
+  test("processMessage receives only callSite — no legacy speed knob", async () => {
+    // The heartbeat service unconditionally passes `callSite:
+    // 'heartbeatAgent'` and nothing else. The resolver maps that identifier
+    // to whatever `llm.callSites.heartbeatAgent` (or `llm.default`)
+    // configures.
     const service = createService();
     await service.runOnce();
 
     expect(processMessageCalls).toHaveLength(1);
-    expect(processMessageCalls[0].options?.callSite).toBe("heartbeatAgent");
+    expect(processMessageCalls[0].options).toEqual({
+      callSite: "heartbeatAgent",
+    });
   });
 
   test("end-to-end: llm.callSites.heartbeatAgent.speed resolves to 'fast'", async () => {

assistant/src/__tests__/model-intents.test.ts

@@ -38,8 +38,10 @@ describe("model intents", () => {
   });
 });
 
-// `RetryProvider`'s legacy `modelIntent` normalization path was removed in
-// PR 19 of the unify-llm-callsites plan. The remaining `resolveModelIntent`
-// helper lives in `providers/model-intents.ts` for use by the workspace
-// migration's snapshot table — see `workspace/migrations/038-unify-llm-
-// callsite-configs.ts`.
+// `RetryProvider` normalizes outbound calls through call-site routing
+// (`resolveCallSiteConfig` against `llm.callSites.<id>` / `llm.default`).
+// The `resolveModelIntent` helper exercised above lives in
+// `providers/model-intents.ts` and is used by the unify-llm workspace
+// migration's snapshot table (see
+// `workspace/migrations/038-unify-llm-callsite-configs.ts`) to seed
+// `llm.default` from any pre-existing `modelIntent` setting.