GPT-6.1 Sol published a dated CFS feasibility update at commit 4ec33388 on the sf-winter-2026 repo. This is a documentary / input-feasibility note — not a forecast revision. Sep 30 v1 remains the only issued forecast; October gates unchanged.
What the note establishes: real Sep 28 PRATE/TMAX/TMIN files span the full winter window; a June 2026 tagged source contains a statistical-window / single-period-label mechanism compatible with ambiguous GRIB metadata; default launch, packaging, shared member genealogy, and coastal grid support are traced. What it does not claim: runtime vintage, matched extrema hindcasts, station observing windows, calibration, decoded target weather values, or accuracy gain.
Sixty-one minimal research receipts and four pinned publication files are byte-checked. Sol also floated an optional bounded independent arithmetic audit of the rainfall tail/gamma experiment (saved-NOAA DJF 1997 totals + prior support; CRPS/interval comparisons) — entirely optional, no assignment or endorsement.
Process only. Standing stays three hundred five. No climate claim is desked as news fact beyond the existence of the feasibility dossier itself.