A Fair-Value & Verification Desk for the Atmosphere
Fair value for the weather.
Then the receipts.
Every operational model — GFS, the European ensemble, AIFS, ICON, the NWS blend — folds into one calibrated, bias-corrected fair value: tomorrow's high, the odds it clears a line, the bust risk before the models break. Then every call is scored against what actually happened. Fair value you can check — priced, verified, settled.
Just want your weather? Get your local forecast → — today, the week, and how far out to trust it.
Price. One calibrated probability per outcome, blended with diversity-aware weights — correlated models don't get counted twice.
Prove. Every model scored against the settlement truth on one common board — no cherry-picking.
Persist. Every forecast frozen at issue time and settled against the record — a track you can audit, not a claim to trust.
— Yesterday, settled · WT called it —
KLAX · high > 103°F · Oct 16 — WT fair value 0% vs ForecastEx 2% · edge -2 pts · OI 3,666
issue 2026101712 · updated 0 h 20 min ago · verified through 2026101612
9 models · one common truth · 544 runs · 20260101→20261011 · every model's card →
Northern Hemisphere day-5 Z500 skill, run over run — the top three models, scored on one common truth. Nobody else publishes this board. Read the full record →
tomorrow (Oct 18) · issue 2026101712 · consensus of the whole model fleet, bias-corrected, with the odds it beats the official number · all 52 stations →
EPS week-2 mean · N. Hemisphere · init 2026101612 — the ensemble splits into storylines; each cluster is a real branch of the future with a probability attached · the cluster engine → · live maps →
Named storylines, by probability mass: Quiet pattern IV 74% (+44) · Quiet pattern 14% (-12) · Quiet pattern III 10% (-4) · the ticker →
Your place, now & the week
Current conditions from your nearest station, the seven-day high and low, when the rain comes, and any active alerts — with the one thing only this desk adds: how far out it's actually trustworthy.
Get your forecast →The local record book
Where this week's forecast heat and cold land against a century of records at your nearest long-term station: the record for the date, how long since it was this hot or cold, the rank.
Open the almanac →Will it rain — when, how much
The seven-day chance and amount at your exact point from the National Weather Service grid, plus the hour-by-hour timing for today and tomorrow.
See the rain outlook →Today's fair value
The calibrated per-station outlook: every model's daily high side by side, the bias-corrected consensus and its odds of clearing a line, and the Fork — ensemble divergence as a bust-risk tell.
See the fair value PRO →The receipts
Every model scored against one common truth: day-5 skill run over run, who to trust in each flow, the reliability diagram, and the season accuracy record — provable by anyone.
Read the record →How it's built
The calibration, the common-truth board that trains the blend, the diversity-aware weighting, and the tamper-evident provenance — every number on the site traced back to how it was made.
Open the method →Is the next run worth waiting for?
Trained on our own frozen record: the probability the consensus moves ≥3°F by the next model cycle, per station and day — a quantified answer to the call every forecaster makes by gut.
Get the verdict →How many models do we really have?
The AI models train toward ECMWF, so their errors travel together. Neff measures the effective number of independent opinions in the fleet — live, trended, by region.
Read the index →Which scenario is verifying?
The ensemble forked at issue; the atmosphere has been voting since. The posterior on each branch given the verifying analysis — which storyline the evidence now supports, and what moved it.
See the verdict →Do we beat the market?
Our fair value as published against the price on Kalshi, Polymarket and ForecastEx, on every settled contract — proper scores and confidence intervals by settlement day. Honest either way.
See the record against the markets →Named storylines, tracked
Each distinct ensemble storyline gets a name the run it emerges and keeps it while it lives — probability mass flowing between scenarios, run over run, on one ticker board.
Watch the ticker →