APH Weather Predictor

SCANNING GLOBAL STATIONS

Markets

Stations Online

Live METAR Observations

Engine Activity

How APH Weather Predictor Works

This app finds mispriced Polymarket temperature markets by comparing real-time airport weather observations to current market prices. It reads the exact same data source that Polymarket uses to resolve markets — ASOS stations at major airports, reported via METAR and archived on Weather Underground.

Four background engines run continuously without any manual intervention:

Engine	Source	Frequency	What It Does
METAR Observer	aviationweather.gov	Every 2 min	Pulls live temperature from every resolution airport station
NWS Forecaster	api.weather.gov	Every 10 min	Refreshes hourly forecast data for all US cities
Market Scanner	Gamma API	Every 15 min	Discovers new temperature events across all cities and dates
Price Tracker	Gamma API	Every 60 sec	Refreshes YES/NO prices for all known markets

After every METAR or price update, the EV calculator re-runs across all markets. The UI refreshes every 10 seconds.

Understanding the Edge

There are three layers of edge, in order of strength:

1. Observation Edge (strongest)

As the day progresses, METAR observations show the actual temperature right now at the resolution station. By early afternoon, the daily high is often effectively locked in — but the market still prices uncertainty. The pace-adjusted projection incorporates this live data. When you see running hot or already hit in the Pace Adj column, the model is highly confident.

The observation edge is strongest between 12pm-4pm local time for High markets, and 4am-8am local time for Low markets. These are the hours where the resolution temperature is being set but the market hasn't fully repriced yet.

2. Forecast Divergence Edge

NWS forecasts are updated hourly but Polymarket prices lag behind, especially for tomorrow's markets. When the NWS shifts a forecast by 3-4 degrees, the market often takes 30-60 minutes to adjust. The system catches these shifts automatically.

3. Model Run Edge

The GFS model updates at 00, 06, 12, and 18 UTC. ECMWF updates at 00 and 12 UTC. When a new model run drops with a significant forecast shift, you have a window of 15-45 minutes before the market reprices.

When to Place Bets

Best Windows for Today's Markets (same-day)

Time (Local)	Market Type	Why This Window
6:00-8:00 AM	Low Temp	Morning low is being set right now. METAR shows real-time low. Market still pricing yesterday's forecast.
10:00 AM-12:00 PM	High Temp	Temperature is ramping. Pacing data tells you if it's running ahead or behind. Market hasn't repriced for the latest observations.
1:00-3:00 PM	High Temp	Peak window. The high is nearly locked in at most stations. Observation edge is at maximum. If you see "running hot" or "already hit", bet with high confidence.

After 4pm local for High markets, the daily high is usually final. You can still find edge if the market hasn't fully resolved to the winning bucket yet, but the window is closing.

Best Windows for Tomorrow's Markets

Time (UTC)	What Happens	Action
03:00-05:00 UTC	GFS 00z run completes	Check for forecast shifts vs. current market prices
09:00-11:00 UTC	GFS 06z + ECMWF 00z available	Best time to bet tomorrow's markets — freshest model data, market is sleepiest
15:00-17:00 UTC	GFS 12z + ECMWF 12z	Afternoon update. Good for refining positions or catching late shifts.

Most Profitable Cities (from Backtest)

10-day backtest results ranked by profitability:

Rank	City	Win Rate	Optimal Sigma	Why It Works
1	Miami	100%	1.0°F	Extremely predictable coastal temps. Market consistently overestimates uncertainty. The tightest sigma of any city.
2	Los Angeles	100%	1.0°F	Marine layer makes LAX temps highly regular. Same tight sigma as Miami. Market treats it like an inland city.
3	Atlanta	91%	1.0°F	Strong performance. Occasional misses from surprise storms, but very predictable most days.
4	Chicago	82%	1.0°F	Lake Michigan adds volatility, but the model handles it. Higher miss rate but still very profitable.
5	Houston	100%	2.0°F	Gulf Coast consistency. Slightly wider sigma than coastal cities but still well-calibrated.
6	Denver	100%	3.5°F	Widest sigma due to Front Range weather swings. Fewer bets but high conviction when they trigger.

Focus your largest positions on Miami and LA. These cities have the tightest sigma (1.0°F), meaning the model is most confident and mispricing is most exploitable.

Bet Sizing Strategy

The Half-Kelly column shows the recommended bet size as a percentage of your bankroll. This is derived from the Kelly Criterion, the mathematically optimal sizing formula that maximizes long-term growth while managing risk.

Rules of Thumb

Half-Kelly %	Meaning	Suggested Action
> 20%	Massive edge detected	Verify the data manually before betting this large. Could be a data anomaly. Cap at 10% of bankroll per bet.
10-20%	Strong edge	Bet comfortably. These are your bread-and-butter plays.
5-10%	Solid edge	Standard bet. Good risk/reward.
2-5%	Marginal edge	Small bet or skip. Only take these if you have high conviction from the pace data.

Never put more than 10% of your total bankroll on a single bet, even if the Kelly formula suggests higher. Temperature markets can have surprises (sudden storms, station equipment issues, data corrections).

Understanding Pace Status Tags

Tag	Meaning	Confidence Level
running hot	Current temp already exceeds the NWS forecast. Projection adjusted upward.	Very high for HIGH markets — the actual high will likely exceed forecast.
already hit	The observed high today has already reached the forecast. Projection uses actual observed high.	Extremely high — the daily high is essentially known. Bet aggressively on any mispriced buckets.
pacing behind	Current temp is below the hourly forecast curve. Projection adjusted slightly downward.	Moderate — the high may still come in lower than forecast, but temps could also catch up.
low already set	For LOW markets: the observed low is already below forecast. Uses actual observed low as projection.	Very high — the daily low is effectively locked in once morning warming begins.
no obs	No METAR observation available for this station yet. Falling back to NWS forecast only.	Lower — you're operating on forecast alone without observation confirmation.

Tips and Tricks

The 2pm Rule: For same-day HIGH markets, the highest-edge bets happen between 1-3pm local time. By then, the station has usually recorded a temperature within 1-2 degrees of the final high. The model's "running hot" or "already hit" signals are most reliable here.

The Sunrise Rule: For LOW markets, the lowest temperature nearly always occurs within an hour of sunrise. If the METAR shows the low has been set and morning temps are climbing, the "low already set" signal is close to guaranteed.

Bet the NO side on adjacent buckets: When the model clearly projects into one bucket (e.g., 86-87°F), the highest EV is often not YES on that bucket — it's NO on the neighboring buckets that the market is still overpricing. Look for 2-3 adjacent NO bets that all benefit from the same directional view.

Stack Miami and LA first: These have the tightest calibrated sigma (1.0°F), which means the model's edge is largest and most consistent. Build your bankroll here before expanding to more volatile cities like Denver or Chicago.

Watch for model run drops: GFS updates hit at 00, 06, 12, 18 UTC. If you see a 3°F+ shift in the NWS forecast after a model run, check if the market has repriced yet. If not, you have a 15-45 minute window to trade.

Beware of very high EV (>30%): Sometimes this signals an illiquid market or a stale price, not a real edge. Always check that the current temp and forecast look reasonable before placing a large bet on any >30% EV signal.

Denver is a trap for the impatient: Denver has the widest sigma (3.5°F) because the Front Range creates genuinely unpredictable temperature swings. The model correctly bets less frequently here. Don't force bets in Denver — let the model find them.

Quick Reference: Market Resolution

Every Polymarket temperature market resolves based on the daily weather history page on Weather Underground for a specific airport ASOS station. The station is listed in each market's rules. Resolution typically happens within hours of midnight local time, after the full day's data is finalized.

Temperature is measured to whole degrees (Fahrenheit for US, Celsius for international). The reading comes from automated sensors that update every 60 seconds, with formal METAR reports every hour and special reports every 5 minutes when conditions change rapidly.