T3 overview tier — mid-day checkpoints + cross-day lesson

Hot path (T1/T2) stays mistral + qwen2.5. The new T3 tier runs a
thinking model SPARINGLY — after every misplacement, every N-th event
(default N=3), and once post-scenario for the cross-day lesson.

- agent.ts: generateCloud() for Ollama Cloud (gpt-oss:120b etc). Uses
  the same /api/generate shape; thinking field is discarded.
- scenario.ts: runOverviewCheckpoint + runCrossDayLesson. Outputs land
  in checkpoints.jsonl and lesson.md. Lesson also seeds playbook_memory
  under operation "cross-day-lesson-{date}" — future runs pick it up
  through the existing similarity boost.
- Env knobs: LH_OVERVIEW_CLOUD=1 routes T3 to cloud, LH_OVERVIEW_MODEL
  overrides (default gpt-oss:20b local, gpt-oss:120b cloud),
  LH_T3_CHECKPOINT_EVERY controls cadence, LH_T3_DISABLE=1 turns it off.

Why this shape: prior feedback_phase19_seed_text.md warned that verbose
seeds dilute the embedding and silently kill the boost. T3's rich prose
goes to lesson.md; the embedded "approach" + "context" stay terse.

Verified end-to-end: local 20b checkpoint 10.9s, lesson 4.0s; cloud
120b lesson 3.7s. Cloud output is both faster AND more specific than
local (sequenced, tactical, logging advice included).

tests/multi-agent/agent.ts

@@ -9,6 +9,12 @@
 export const GATEWAY = "http://localhost:3100";
 export const SIDECAR = "http://localhost:3200";
 
+// Ollama Cloud — used for the T3 overview tier when LH_OVERVIEW_CLOUD=1.
+// Same /api/generate surface as local Ollama; just needs the bearer key.
+// Default base and key are read from env so secrets never land in git.
+export const OLLAMA_CLOUD_URL = process.env.OLLAMA_CLOUD_URL ?? "https://ollama.com";
+export const OLLAMA_CLOUD_KEY = process.env.OLLAMA_CLOUD_KEY ?? "";
+
 // --- Shared types ---
 
 export type Role = "executor" | "reviewer";
@@ -111,6 +117,48 @@ export async function generate(model: string, prompt: string, opts: {
   return text;
 }
 
+// Cloud generate — hits Ollama Cloud directly with the bearer key. Same
+// /api/generate shape as local Ollama; `thinking` field (for gpt-oss:Nb)
+// is discarded, only `response` is returned. Caller should budget
+// num_predict ≥ 400 so thinking-model reasoning has room before the
+// visible response starts.
+export async function generateCloud(model: string, prompt: string, opts: {
+  max_tokens?: number;
+  temperature?: number;
+  system?: string;
+} = {}): Promise<string> {
+  if (!OLLAMA_CLOUD_KEY) {
+    throw new Error("OLLAMA_CLOUD_KEY not set; cannot reach Ollama Cloud");
+  }
+  const body: Record<string, any> = {
+    model,
+    prompt,
+    stream: false,
+    options: {
+      temperature: opts.temperature ?? 0.3,
+      num_predict: Math.max(opts.max_tokens ?? 800, 400),
+    },
+  };
+  if (opts.system) body.system = opts.system;
+  const resp = await fetch(`${OLLAMA_CLOUD_URL}/api/generate`, {
+    method: "POST",
+    headers: {
+      "Authorization": `Bearer ${OLLAMA_CLOUD_KEY}`,
+      "Content-Type": "application/json",
+    },
+    body: JSON.stringify(body),
+  });
+  if (!resp.ok) {
+    throw new Error(`Ollama Cloud ${resp.status}: ${await resp.text().catch(() => "?")}`);
+  }
+  const data: any = await resp.json();
+  const text = data.response ?? "";
+  if (!text) {
+    throw new Error(`Ollama Cloud returned empty response for ${model}: ${JSON.stringify(data).slice(0, 200)}`);
+  }
+  return text;
+}
+
 // --- Prompt construction ---
 
 const TOOL_CATALOG = `