previous_run_not_resumable: a Format run with nothing to continue

400 previous_run_not_resumable means the run you named in previous_run_id left nothing to continue: it has no thread_id, and it neither completed nor produced any artifacts. Start a fresh run instead; the error's details show previous_run_status, has_thread and artifact_count so you can see why.

The refusals are listed in Runs and results and Errors and spend, read on 2026-10-02.

What are all four refusals?

A continuation is a new run that replays what an earlier run produced, so it can redo one part. Four checks guard it.

How do I check before I call?

You can read the answer off the earlier run's receipt. It is continuable when thread_id is not null, and it either completed or has a non-empty artifacts[]. A failed run that left work behind can be continued; one that left nothing cannot.

That asymmetry is useful. A run that failed after rendering three of five scenes is a good candidate. A run rejected before any generation, for example by an unattended block, usually has no artifacts, and the docs tell you to start fresh.

What does a continuation look like?

Name the earlier run, say what to redo, and bind the same schema again: output_schema is per run and is not inherited. Use a new Idempotency-Key, because the old one is bound to the receipt you already have.

curl -sS -X POST "https://api.sume.com/v1/formats/acme/live-commerce/runs" \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: order-8823-retry-sc7" \
  -d '{
    "previous_run_id": "arun_e43e6c5cb2b74052",
    "instruction": "Retry scene 7 only. Keep the rest unchanged.",
    "input": { "scene_id": "sc_7" },
    "generation_spend_cap_usd": 8
  }'

What stays the same and what is new?

A continuation gets a new run id, a new receipt, its own spend cap and its own single webhook. The original run never changes. Both share thread_id, which is read-only. Never send a thread id yourself; that is 400 unknown_parameter.

artifacts[] on a continued run lists everything the whole conversation generated, while usage stays per run, so you pay for the new turn's generation only.

When should I not continue?

When the earlier run produced nothing, when it was on a different Format, or when you want a different brief entirely. In those cases start fresh with a new key. Continuing is a cost tool for partial work, not a way to force a run that never started.

A small guard in your client

The guard costs one read. A refused continuation is a 400 and costs nothing, but a blind fresh run on a run that did leave work behind spends on clips you already own.

• Read the earlier run's receipt first and check thread_id is set.
• Continue only if the run completed or artifacts[] is non-empty.
• Otherwise start a fresh run with a new Idempotency-Key and the full brief.
• Compare usage.billable_amount_usd_micros with the cap before raising a cap on the retry.

What about the spend cap on a continuation?

Each continuation has its own cap. Send generation_spend_cap_usd on the new run; omitting it inherits the Format's cap, which is $400 when the Format never set one. A retry of a single scene needs a fraction of a full show's cap, so set a small one on purpose.

A run that wants to spend past its cap fails with format_run_failed, so compare usage.billable_amount_usd_micros with usage.generation_spend_cap_usd_micros on the receipt before raising the number.

Can I continue a run that is still going?

No. 409 previous_run_not_terminal says the run has not finished. Poll status_url until the status is terminal, or take the webhook, then call again. That error is retryable once the earlier run settles; the not-resumable error is not.

If the earlier run is stuck, remember the force-finalize bounds: 90 minutes from creation, or sooner when the run is older than 25 minutes and silent for 10. After that it is failed and may be continuable if it left artifacts.