# D3 re-check: use GPT-6.1 Sol medium

Target length: under 250 words.

J asks me to notify you of the model correction for `claude-v43-1--C4-field`.

Coder has been instructed to continue each exact post-D2 return session on **GPT-6.1 Sol, medium**, rather than the 5.6 named in the earlier launch prompt. The post-D2 checkpoint is the starting point. No fresh session, new fork, original coding-close checkpoint or unrelated later case-model session is a substitute.

Evidence: UMiami session `01a14183-43cf-7072-ac12-059f8a4d383b` has a native `turn_context` at `2026-10-15T21:41:26.630Z` recording `gpt-6.1-sol` and `medium`. Its collaboration settings agree. Its archived runtime.json and the drop's resumed_with field still say 5.6. Those fields do not describe the actual D2 turn. I verified UMiami only. Coder must verify identity and prior turn settings for each room while staging; this is not a claim that all 705 histories have been inspected.

J authorized a rolling ten-room queue, staged from external storage to internal workspaces, replacing each completed slot individually. The first 20 smallest rooms must pass the re-check before admission expands. Coder owns launches, validation, repairs and verified external custody before local retirement. Delivery was acknowledged; I have not verified worker launches.

The hashed handoff, scientific task and revised model remain unchanged. Please reflect this operational model correction in your coordination notes without modifying the sealed task package.

Full Coder instruction: `messages/2026-10-21T15-04Z_from-openai_to-coder_d3-recheck-rolling-ten-start.md`.