diff --git a/packages/producer/src/services/distributed/plan.test.ts b/packages/producer/src/services/distributed/plan.test.ts
index 966f34b9a..2106a90b3 100644
--- a/packages/producer/src/services/distributed/plan.test.ts
+++ b/packages/producer/src/services/distributed/plan.test.ts
@@ -164,9 +164,12 @@ describe("plan() — golden planDir + planHash determinism", () => {
     async () => {
       const planDir = join(runRoot, "plan-layout");
       mkdirSync(planDir, { recursive: true });
+      // Pin chunkSize=240 so this fixture exercises the single-chunk path
+      // (totalFrames=30 → ceil(30/240)=1 chunk). The auto-sized variant
+      // (chunkSize=undefined) is exercised by the dedicated test below.
       const result = await plan(
         projectDir,
-        { fps: 30, width: 320, height: 240, format: "mp4" },
+        { fps: 30, width: 320, height: 240, format: "mp4", chunkSize: 240 },
         planDir,
       );
 
@@ -185,7 +188,7 @@ describe("plan() — golden planDir + planHash determinism", () => {
       // ── PlanResult contract ─────────────────────────────────────────────
       expect(result.planDir).toBe(planDir);
       expect(result.planHash).toMatch(/^[0-9a-f]{64}$/);
-      expect(result.chunkCount).toBeGreaterThanOrEqual(1);
+      expect(result.chunkCount).toBe(1);
       expect(result.totalFrames).toBe(30); // 1s @ 30fps
       expect(result.width).toBe(320);
       expect(result.height).toBe(240);
@@ -218,6 +221,37 @@ describe("plan() — golden planDir + planHash determinism", () => {
     TIMEOUT_MS,
   );
 
+  it(
+    "auto-sizes chunkSize end-to-end when caller omits it",
+    async () => {
+      // Integration check that the auto-sizer wired through plan() actually
+      // produces multi-chunk output for the same fixture that single-chunks
+      // when chunkSize is pinned. With totalFrames=30 and the default
+      // maxParallelChunks=16, the auto-sizer picks
+      // max(MIN_CHUNK_SIZE=10, ceil(30/16)=2) = 10 → ceil(30/10) = 3 chunks.
+      const planDir = join(runRoot, "plan-autosized");
+      mkdirSync(planDir, { recursive: true });
+      const result = await plan(
+        projectDir,
+        { fps: 30, width: 320, height: 240, format: "mp4" },
+        planDir,
+      );
+      expect(result.chunkCount).toBe(3);
+      const chunks = JSON.parse(
+        readFileSync(join(planDir, "meta", "chunks.json"), "utf-8"),
+      ) as Array<{ index: number; startFrame: number; endFrame: number }>;
+      expect(chunks).toHaveLength(3);
+      // Encoder gopSize must follow the auto-sized chunk so chunk-boundary
+      // IDR keyframes still land at frame 0 of each chunk.
+      const encoder = JSON.parse(
+        readFileSync(join(planDir, "meta", "encoder.json"), "utf-8"),
+      ) as Record<string, unknown>;
+      expect(encoder.gopSize).toBe(10);
+      expect(encoder.chunkSize).toBe(10);
+    },
+    TIMEOUT_MS,
+  );
+
   it(
     "produces a byte-identical planHash on a second invocation",
     async () => {
diff --git a/packages/producer/src/services/distributed/plan.ts b/packages/producer/src/services/distributed/plan.ts
index e8cbcea62..821bc27b2 100644
--- a/packages/producer/src/services/distributed/plan.ts
+++ b/packages/producer/src/services/distributed/plan.ts
@@ -109,6 +109,13 @@ export interface DistributedRenderConfig {
    * ceil(totalFrames / maxParallelChunks))`. The auto-size floor
    * (`MIN_CHUNK_SIZE = 10`) keeps per-chunk fixed overhead from
    * swamping the parallelism gain on tiny renders.
+   *
+   * `effectiveChunkSize` also drives `LockedRenderConfig.gopSize` — every
+   * chunk's first frame is an IDR keyframe, so smaller chunks mean a
+   * tighter GOP and larger encoded files. Callers who optimize for
+   * output bytes (rather than wall-clock parallelism) should pass an
+   * explicit `chunkSize` matching their target GOP — e.g. `240` for the
+   * old 8-second-GOP behavior.
    */
   chunkSize?: number;
   /** Default `16`. Caps long renders to fewer-but-longer chunks for operational fairness. */
@@ -350,22 +357,22 @@ export function measurePlanDirBytes(planDir: string): number {
 
 /**
  * Compute `(chunkCount, effectiveChunkSize)` from total frames and the
- * caller's chunking knobs:
+ * caller's chunking knobs. The operative chunk size is
+ * `resolvedChunkSize` — equal to `configChunkSize` when the caller
+ * passes one, otherwise auto-sized from `maxParallelChunks`:
  *
- *     chunkCount = min(maxParallelChunks, ceil(totalFrames / chunkSize))
- *     effectiveChunkSize = max(configChunkSize, ceil(totalFrames / maxParallelChunks))
+ *     resolvedChunkSize   = configChunkSize ?? max(MIN_CHUNK_SIZE, ceil(totalFrames / maxParallelChunks))
+ *     chunkCount          = min(maxParallelChunks, ceil(totalFrames / resolvedChunkSize))
+ *     effectiveChunkSize  = max(resolvedChunkSize, ceil(totalFrames / chunkCount))
  *
  * Long renders auto-rescale to fewer-but-longer chunks rather than
  * fragmenting infinitely. Returned `chunkCount >= 1` (`totalFrames === 0`
- * is rejected upstream); `effectiveChunkSize >= configChunkSize`.
+ * is rejected upstream); `effectiveChunkSize >= resolvedChunkSize`.
  *
- * When `configChunkSize` is `undefined`, the input is auto-sized from
- * `maxParallelChunks`: `max(MIN_CHUNK_SIZE, ceil(totalFrames /
- * maxParallelChunks))`. This honors the caller's fan-out intent — passing
- * `maxParallelChunks=16` without `chunkSize` now produces 16 chunks
- * (subject to the `MIN_CHUNK_SIZE` floor on tiny renders) instead of
- * silently clamping to a 240-frame default. Explicit numbers, including
- * `240`, take precedence over the auto-sizer.
+ * The auto-sizer (triggered when `configChunkSize` is `undefined`) honors
+ * the caller's fan-out intent: passing `maxParallelChunks=16` without
+ * `chunkSize` produces 16 chunks (subject to the `MIN_CHUNK_SIZE` floor
+ * on tiny renders). Explicit numbers, including `240`, take precedence.
  */
 export function resolveChunkPlan(
   totalFrames: number,