diff --git a/packages/coding-agent/src/auto-thinking/classifier.ts b/packages/coding-agent/src/auto-thinking/classifier.ts
index 4755ad074..62e922e47 100644
--- a/packages/coding-agent/src/auto-thinking/classifier.ts
+++ b/packages/coding-agent/src/auto-thinking/classifier.ts
@@ -15,7 +15,6 @@
  * the caller falls back to a concrete level and continues the turn.
  */
 import { type AssistantMessage, completeSimple, Effort, type Model } from "@oh-my-pi/pi-ai";
-import { THINKING_EFFORTS } from "@oh-my-pi/pi-catalog/effort";
 import { getSupportedEfforts } from "@oh-my-pi/pi-catalog/model-thinking";
 import { prompt } from "@oh-my-pi/pi-utils";
 
@@ -96,10 +95,9 @@ export async function classifyDifficulty(
 		backend === ONLINE_AUTO_THINKING_MODEL_KEY
 			? await classifyOnline(input, deps, ceiling)
 			: await classifyLocal(input, backend, deps);
-	// Policy clamp before the model clamp: a hallucinated `max` must not cross a
-	// ceiling the user did not opt into, even on a model that supports the tier.
-	const capped = THINKING_EFFORTS.indexOf(effort) > THINKING_EFFORTS.indexOf(ceiling) ? ceiling : effort;
-	return clampAutoThinkingEffort(deps.model, capped);
+	// The ceiling goes into the clamp itself: capping the request alone is not
+	// enough, because a sparse ladder snaps an excluded request back up.
+	return clampAutoThinkingEffort(deps.model, effort, ceiling);
 }
 
 async function classifyOnline(input: string, deps: ClassifyDifficultyDeps, ceiling: Effort): Promise<Effort> {
diff --git a/packages/coding-agent/src/thinking.ts b/packages/coding-agent/src/thinking.ts
index 85739955d..9ef31c328 100644
--- a/packages/coding-agent/src/thinking.ts
+++ b/packages/coding-agent/src/thinking.ts
@@ -197,6 +197,11 @@ export function parseCliThinkingLevel(value: string | null | undefined): Configu
  * above Low (falling back to the full supported set only when the model maxes
  * out below Low). Within that pool the request snaps to the highest level not
  * exceeding it, or the pool minimum when the request is below the pool.
+ * `ceiling` bounds the pool from above, so a policy ceiling survives the model
+ * clamp: a sparse ladder such as `["max"]` must not snap an `xhigh` request up
+ * to `max`. When no supported tier sits at or below the ceiling there is
+ * nothing legal to pick and the result is `undefined` — auto leaves the current
+ * level alone rather than billing a tier the caller excluded.
  *
  * Returns `undefined` for reasoning-capable models without a controllable
  * effort surface (`thinking.efforts` empty — e.g. devin-agent models, where
@@ -205,12 +210,19 @@ export function parseCliThinkingLevel(value: string | null | undefined): Configu
  * forward a concrete effort that would then trip {@link requireSupportedEffort}
  * downstream.
  */
-export function clampAutoThinkingEffort(model: Model | undefined, effort: Effort): Effort | undefined {
+export function clampAutoThinkingEffort(
+	model: Model | undefined,
+	effort: Effort,
+	ceiling: Effort = Effort.Max,
+): Effort | undefined {
 	const supported = model ? getSupportedEfforts(model) : THINKING_EFFORTS;
 	if (supported.length === 0) return undefined;
 	const lowIndex = THINKING_EFFORTS.indexOf(Effort.Low);
-	const eligible = supported.filter(level => THINKING_EFFORTS.indexOf(level) >= lowIndex);
-	const pool = eligible.length > 0 ? eligible : supported;
+	const ceilingIndex = THINKING_EFFORTS.indexOf(ceiling);
+	const withinCeiling = supported.filter(level => THINKING_EFFORTS.indexOf(level) <= ceilingIndex);
+	if (withinCeiling.length === 0) return undefined;
+	const eligible = withinCeiling.filter(level => THINKING_EFFORTS.indexOf(level) >= lowIndex);
+	const pool = eligible.length > 0 ? eligible : withinCeiling;
 	const requestedIndex = THINKING_EFFORTS.indexOf(effort);
 	let chosen = pool[0];
 	for (const candidate of pool) {
@@ -253,12 +265,16 @@ export function resolveTaskEffortLevel(model: Model | undefined, effort: TaskEff
  * the model's `defaultLevel`, otherwise High, clamped into the auto range.
  *
  * Deliberately stays below {@link Effort.Max}: the placeholder must not bill the
- * top tier for a turn nobody classified, so a `defaultLevel` of `max` is capped
- * at XHigh before clamping. Classification itself may still resolve Max on
- * models that expose the tier. Returns `undefined` for non-reasoning models.
+ * top tier for a turn nobody classified, so XHigh is passed as a hard ceiling
+ * rather than only capping the preferred level — otherwise a sparse `["max"]`
+ * ladder would snap straight back up. A model whose ladder offers nothing at or
+ * below XHigh therefore has no provisional level, and `auto` leaves the current
+ * one in place. Classification itself may still resolve Max on models that
+ * expose the tier when the user opts in. Returns `undefined` for non-reasoning
+ * models.
  */
 export function resolveProvisionalAutoLevel(model: Model | undefined): Effort | undefined {
 	if (!model?.reasoning) return undefined;
 	const preferred = model.thinking?.defaultLevel ?? Effort.High;
-	return clampAutoThinkingEffort(model, preferred === Effort.Max ? Effort.XHigh : preferred);
+	return clampAutoThinkingEffort(model, preferred === Effort.Max ? Effort.XHigh : preferred, Effort.XHigh);
 }
diff --git a/packages/coding-agent/test/auto-thinking-classifier.test.ts b/packages/coding-agent/test/auto-thinking-classifier.test.ts
index 7636bedab..71cae11ec 100644
--- a/packages/coding-agent/test/auto-thinking-classifier.test.ts
+++ b/packages/coding-agent/test/auto-thinking-classifier.test.ts
@@ -325,6 +325,33 @@ describe("auto thinking classifier helpers", () => {
 		expect(await classifyDifficulty("untangle this cross-service race", fixture.deps)).toBe(Effort.XHigh);
 	});
 
+	it("never bills a max-only ladder without opt-in", async () => {
+		// `["max"]` has nothing at or below the default ceiling, so the model clamp
+		// must not snap the request back up — auto yields nothing and the session
+		// keeps its current level.
+		const defaulted = createOnlineFixture(buildLadderModel("mock-max-only", [Effort.Max]), "xhigh");
+		expect(await classifyDifficulty("cut over the storage layer", defaulted.deps)).toBeUndefined();
+
+		vi.restoreAllMocks();
+
+		const optedIn = createOnlineFixture(buildLadderModel("mock-max-only", [Effort.Max]), "max", "max");
+		expect(await classifyDifficulty("cut over the storage layer", optedIn.deps)).toBe(Effort.Max);
+	});
+
+	it("has no provisional level on a max-only ladder", () => {
+		expect(resolveProvisionalAutoLevel(buildLadderModel("mock-max-only", [Effort.Max]))).toBeUndefined();
+	});
+
+	it("keeps the ceiling out of the pool for sparse ladders", () => {
+		const maxOnly = buildLadderModel("mock-max-only", [Effort.Max]);
+
+		expect(clampAutoThinkingEffort(maxOnly, Effort.XHigh, Effort.XHigh)).toBeUndefined();
+		expect(clampAutoThinkingEffort(maxOnly, Effort.Max)).toBe(Effort.Max);
+		expect(
+			clampAutoThinkingEffort(buildLadderModel("mock-hm", [Effort.High, Effort.Max]), Effort.Max, Effort.XHigh),
+		).toBe(Effort.High);
+	});
+
 	it("keeps the provisional auto level below max even when the model defaults to it", () => {
 		const maxDefaultModel = buildModel({
 			id: "mock-max-default",