Quake Predictor API 🌋

Probabilistic seismic risk API — F1=18.3% backtest accuracy on USGS catalog 2020–2026.

• 🌋 32 global seismic regions monitored using USGS catalog (2020–2026, 50,689 events)
• 📊 Model: logistic regression ensemble, 14-day horizon, M≥6.0, 600km radius
• 🧠 Backtest precision: 17.3% · F1 score: 18.3% (k=7, Aug 2024–2026)
• ⚠️ Disclaimer: probabilistic advisory — not deterministic prediction
• 💰 Free tier: 1 call/day per IP
• 💵 Paid: 0.05 USDC/call via EIP-3009 (Base L2)
• 🔗 Payment address: 0x6dDCd5CC6f0614A291954daf2fF1B41DA44363DE
• 🔗 Endpoint: http://forex2026.mooo.com:5040

Quick Start

# Free call (1/day per IP)
curl "http://forex2026.mooo.com:5040/predict?top=5&mag=6.0"

# Paid call — see EIP-3009 integration in docs

Performance

Metric	Value	Period
F1 score	18.3%	Aug 2024–Jun 2026
Precision	17.3%	Aug 2024–Jun 2026
Recall	variable by region	Aug 2024–Jun 2026
Catalog events	50,689	2020–2026
Regions	32	Global

Backtest methodology: rolling forward validation, k=7, horizon=14d, min_mag=6.0, distance≤600km. USGS catalog only.

API Reference

GET /predict

Returns ranked seismic risk predictions.

Parameter	Type	Default	Description
top	int	5	Number of top-risk regions (max 10)
mag	float	6.0	Minimum magnitude
horizon	int	14	Forecast window in days

Response fields:

• calibrated_prob_14d: calibrated probability of M≥mag event within horizon
• raw_score: uncalibrated model score
• events_30d / events_365d: recent seismic activity count
• days_since_m6: days since last M6+ in region
• magnitude_estimate: expected magnitude range

GET /regions

Returns list of all 32 monitored seismic regions with coordinates.

GET /health

Returns catalog status and server health.

Tech Stack

Python 3 · Flask · Gunicorn · nginx · x402 protocol · Base L2 (Chain ID: 8453) · EIP-3009 · USGS FDSN

Accuracy Note

F1=18.3% means that when the model predicts an event, it occurs approximately 1 in 5.5 times. This is consistent with the theoretical limits of seismic forecasting — USGS and USGS帕德雷克 experiment report similar F1 scores for M6+ 30-day forecasts. The model is useful as a risk prioritization tool, not a deterministic alarm.

License

MIT — github.com/tronnew/quake-predictor