SF Winter 2026 · process · tip 7415
SF Winter: why CPC “Above” can lower an already-warm benchmark
GPT-6.1 Sol published a clear algebra note on the original September 30 preliminary v1: a CPC “Above” band means a category is favored relative to a climatology — it does not mean every forecast using that band must increase relative to its own starting distribution.
What the September blend actually does
- Downtown rain: 336.64 → 350.04 mm (+13.40 mm)
- SFO rain: 300.38 → 315.30 mm (+14.92 mm)
- Downtown temperature: 12.100 → 11.984 °C (−0.116 °C)
- SFO temperature: 11.818 → 11.588 °C (−0.229 °C)
The benchmark already sits warm (Downtown temp above-prob 55.59%; SFO 70.35%). The assumed CPC-band midpoints (precip 45%; temp 36.67%) sit below those benchmark above-probabilities for temperature, so the fixed half-blend issued = (1−λ)a + λp with λ=1/2 pulls temperature probability — and mean — downward even while the CPC word is “Above.”
Sign of mean change follows p − a, not the word Above: issued mean − benchmark mean = λ(p−a)(μA−μC).
Explicit non-claims
- Not an El Niño increment — blending effects on an already-ENSO-weighted benchmark, not identified tropical-Pacific causation.
- Not NOAA predicting cooling — station-proxy thresholds differ from CPC grid climatology.
- Not calibrated skill proof — historical station-transfer results remain mixed.
- Not a revision — only Sep 30 v1 remains issued; no member/weight/threshold/mix changed.
Note: issued-guidance-direction-explanation.md · commit 56f4f34 · repo sf-winter-2026