🌀 weather-oracle

autonomous weather prediction markets engine.

built by br0br0 — cosmic thinker, 100x engineer. finding edges where meteorology meets markets.

---

what is this

weather-oracle scans Polymarket temperature markets, builds multi-source ensemble forecasts, calibrates against historical station data, and identifies mispriced contracts. it paper-trades using fractional Kelly sizing with per-station bias corrections.

the atmosphere doesn't lie. markets do.

architecture

┌─────────────┐     ┌──────────────┐     ┌─────────────────┐
│  scanner.py  │────▶│ ensemble     │────▶│  paper_trade.py │
│  slug lookup │     │ forecasts    │     │  kelly sizing   │
│  market parse│     │ ICON + GFS   │     │  early exits    │
└─────────────┘     │ + NWS (US)   │     └────────┬────────┘
                    └──────────────┘              │
                                                  ▼
┌─────────────┐     ┌──────────────┐     ┌─────────────────┐
│ calibrate.py │────▶│ station bias │     │  dashboard.py   │
│ 30-day hist  │     │ corrections  │     │  terminal UI    │
└─────────────┘     └──────────────┘     └─────────────────┘
        ▲
        │
┌─────────────┐
│   metar.py   │
│ live NOAA    │
│ observations │
└─────────────┘

stations

13 global stations calibrated. METAR observations from NOAA every 5 minutes.

station	city	MAE	bucket accuracy	status
LFPG	Paris	0.42°C	67%	🟢 primary
KMIA	Miami	0.60°C	53%	🟢 best US
WSSS	Singapore	0.70°C	40%	🟢
KSEA	Seattle	0.77°C	43%	🟢
VHHH	Hong Kong	0.83°C	27%	🟢
KATL	Atlanta	0.90°C	33%	🟢
SBGR	São Paulo	0.91°C	27%	🟢
RJTT	Tokyo	1.16°C	37%	🟢 bias-corrected
KORD	Chicago	1.18°C	20%	🟡 warm bias
KDAL	Dallas	1.19°C	20%	🟡 warm bias
CYYZ	Toronto	1.51°C	30%	🟡 caution
KLGA	NYC	2.05°C	7%	🔴 no trade
RKSI	Seoul	3.33°C	10%	🔴 no trade

scripts

• scanner.py — scans Polymarket via slug-based lookup, parses °C and °F markets, builds ensemble probabilities, identifies edges
• paper_trade.py — executes paper trades with fractional Kelly, tracks P&L, supports early exits when market reaches target
• calibrate.py — 30-day rolling calibration per station, measures bias/MAE/RMSE, sets tradeable flags
• metar.py — real-time METAR observations from NOAA ADDS, tracks intraday max temps
• dashboard.py — rich terminal dashboard, Bloomberg-terminal aesthetic

setup

# clone
git clone https://github.com/br0br0/weather-oracle.git
cd weather-oracle

# python 3.10+
pip install rich  # optional, for terminal dashboard

# create data directory (not tracked)
mkdir -p data

# run calibration first
python3 scripts/calibrate.py

# scan markets
python3 scripts/scanner.py

# paper trade
python3 scripts/paper_trade.py

# terminal dashboard
python3 scripts/dashboard.py

data sources

• Open-Meteo — ICON + GFS ensemble forecasts (free, no key)
• NOAA ADDS — METAR observations (free)
• NWS API — US gridpoint forecasts (free)
• Polymarket — temperature prediction markets

risk management

• fractional Kelly sizing (quarter-Kelly default)
• per-station bias corrections from 30-day calibration
• max 15% bankroll per trade, 30% total deployed
• station tradeability gates based on MAE thresholds
• early exit strategy at configurable price targets

dashboard

live dashboard at br0br0.xyz/oracle — mission control for weather prediction markets. equity curves, station accuracy heatmaps, active positions.

philosophy

markets are information systems. weather is entropy made visible. the edge lives in the gap between ensemble consensus and crowd psychology. calibrate everything. trust nothing. let the data breathe.

---

built in public · br0br0.xyz · @br0br0ing