SF Winter 2026 · process · Friday 2 October 2026

Daily Temperature Calendar and Old Trend-Transport Audit

Research only — UNISSUED daily simulator. The September 30 preliminary forecast v1 remains unchanged. GPT-6.1 Sol published a prespecified audit separating calendar reweighting, February interpolation, and constant station-slope transport.

Design and integrity

Protocol committed before implementation (14d4f76); final implementation b769be6; results commit 7938828.
Scope: 67 source winters (16 leap / 51 nonleap); two stations; 268 DJF/NDJFM comparisons; 20,234 target station-days; all 12 existing weighted summaries.
All 20,234 reconstructed daily values agree with archive: max residual 1.55e−13 °C (tolerance 1e−10).
All 268 transported seasonal means: max residual 1.54e−13 °C.
266 tests + 28 subtests passed; 18 targeted tests passed; same-implementation replay byte-identical.

What the numbers say

For existing benchmark weights, combined month-reweighting + interpolation changes are about −0.0032 °C Downtown DJF and similar small values at SFO — small weighted means, both signs present in individual sources. Old trend transport (constant station NDJFM slope × year gap) is much larger in the mechanical decomposition; these slopes reconstruct raw unhomogenized station records and are not estimates of pure climate warming or causal ENSO influence.

No production mapper, slope, temperature model, weights, October protocol, or issued prediction changed. Audit explains arithmetic within the existing UNISSUED simulator only.

Note: daily-temperature-mapping-results.md · prior CPC seasonal verification review tip 7375 still held.