๐ก๏ธ 14-Model Ensemble Forecasts
Daily high temperature predictions from weighted model ensemble
๐ Model Agreement
โฐ Hourly Forecast (HRRR Model)
High Resolution Rapid Refresh โข Updates hourly โข Best for 0-18hr predictions
๐ก Data Sources
โ YES = bucket underpriced โข โ NO = bucket overpriced โข Today's resolved/risky bets filtered out
๐ Bucket Agreement Overview
๐งฎ Best Bets Philosophy & Calculation Details
๐ Model Agreement (14 Models)
We query 13 weather models: ECMWF, GFS, ICON, GEM, JMA, HRRR, NWS, Tomorrow.io, OpenWeatherMap, WeatherAPI, PirateWeather, VisualCrossing, and Open-Meteo blend. Each model predicts the daily high temperature.
Agreement % = (Models predicting same bucket) / (Total models) ร 100
๐ฏ Edge Calculation Per Model
For each model prediction, we calculate probability using a normal distribution (ฯ = 2.0ยฐF typical forecast error):
Model Prob = P(low โค actual โค high) using Z-scores
YES Edge = Model Prob โ Market Price
NO Edge = (100 โ Model Prob) โ (100 โ Market Price)
Worst-Case Edge = MIN(all agreeing model edges) โ conservative metric
Average Edge = MEAN(all agreeing model edges) โ used for NEAR-SAFE tier
๐ Confidence Tiers
| ๐ LOCK | 100% unanimous + end-band bucket + worst edge โฅ15% + 5+ models |
| ๐ช STRONG | (100% unanimous OR 90%+ with 6+ models) + worst edge โฅ10% |
| ๐ก๏ธ SAFE | 80%+ agreement + worst edge โฅ10% |
| ๐ NEAR-SAFE | 60%+ agreement + average edge โฅ5% |
โ YES vs โ NO Bets
YES bets: Models agree the bucket is underpriced. We look for consensus where most models predict the temperature will fall IN the bucket.
NO bets: Models agree the bucket is overpriced. 60%+ of models predict the temperature will fall OUTSIDE the bucket.
โก Current Day Filtering
For today's bets, we fetch the observed high from NWS (updates hourly) and filter out:
| โ LOST | YES on range where observed > high bound; NO on "X or higher" where observed โฅ X |
| โ WON | YES on "X or higher" where observed โฅ X; NO on range where observed > high bound |
| โ ๏ธ RISKY | NO bet where observed is IN or within 1ยฐF of the bucket range |
๐ฐ Why End-Bands are Special
End-band buckets ("X or higher" / "X or lower") have infinite range on one side. If all models agree the temp will exceed the threshold, the probability approaches certainty โ making unanimous end-band agreement the strongest signal.
โ ๏ธ Not financial advice. Even LOCK bets can lose due to sudden weather changes, model errors, or measurement issues. Past performance does not guarantee future results. Bet responsibly.
Select a market above
๐ Bucket Analysis
| Bucket | YES % | NO % | Mkt % | YES Edge | YES EV | NO Edge | NO EV | Signal |
|---|---|---|---|---|---|---|---|---|
| Select a market | ||||||||
๐ก Data Sources (14-Model Ensemble)
๐ Model Weights
โญ ECMWF = Most accurate globally โข โก HRRR = Best for same-day
๐งฎ Methodology
14-Model Weighted Ensemble: Blends Tomorrow.io, NWS, GFS, ECMWF, ICON, GEM, JMA, HRRR, OpenWeather, WeatherAPI, PirateWeather, Visual Crossing, and NWS MOS (bias-corrected) with dynamic weighted averaging based on model accuracy, time-of-day, and current weather regime.
YES % (Model) = Probability the temp lands IN this bucket based on ensemble prediction.
NO % (Model) = 100 - YES % = Probability temp lands OUTSIDE this bucket.
YES Edge = YES Model % - Market %. Positive = bucket underpriced, bet YES.
NO Edge = NO Model % - (100 - Market %). Positive = bucket overpriced, bet NO.
EV% = Expected Value as percentage of stake.
โ ๏ธ This is not financial advice. Past performance does not guarantee future results.
๐ Performance Dashboard
Track your betting performance by confidence tier
Performance tracking coming soon. Place bets and verify results to build your record.
๐ฏ Model Accuracy Leaderboard
Dynamic model rankings based on historical accuracy
| # | Model | Avg Error | Bias | Samples | Weight |
|---|---|---|---|---|---|
| No accuracy data yet. Data will populate as markets resolve. | |||||
๐ค๏ธ Weather Regime Detection
Current conditions determine which models perform best
๐ Historical Results
Daily resolution log with model performance
๐ฐ Polymarket Wallet Tracker
Real-time P&L tracking for temperature bets
Data from Polymarket API โข Filters for temperature markets only โข Updates manually
โ๏ธ Model Status Dashboard
Real-time status of 13 weather data sources (6 APIs + 7 individual forecast models)
๐บ๏ธ Model Agreement Heatmap
Visual comparison of model predictions across cities and days
Loading heatmap...