DOCUMENTATION

Protocol Architecture

Nox Aeterna operates as a 4-stage autonomous pipeline. Each stage runs independently with fault tolerance and automatic recovery.

STAGE 01
Data Ingestion
STAGE 02
Prediction
STAGE 03
Edge Detection
STAGE 04
Execution
01

Data Ingestion

The collector runs every 30 minutes, fetching bracket prices from the Polymarket Gamma API for all 10 weather stations. Ensemble weather forecasts are pulled from Open-Meteo every 6 hours.

02

Prediction

A dual LightGBM model predicts both mean temperature and variance for each station. The Gaussian CDF maps these to per-bracket probabilities. Platt scaling calibration corrects systematic biases.

03

Edge Detection

Model probabilities are compared against normalized Polymarket prices. An edge is the difference between model estimate and market price. Only edges above 7% generate trading signals.

Active

Signal Logic

edge = model_prob - market_price. BUY YES if edge > +7%, BUY NO if edge < -7%

#Threshold#7%
Explore →
Active

Kelly Sizing

Half-Kelly criterion with 20% max fraction and 10% minimum probability floor

#Half-Kelly
Explore →
04

Execution

Trades are executed via the Polymarket CLOB API using py-clob-client. Supports both BUY YES and BUY NO positions with automatic fee handling and position tracking.

Active

CLOB Orders

Limit orders placed at best available price via Polymarket CLOB API

#Polygon#CLOB
Explore →
Active

BUY YES + BUY NO

Both sides supported. 2% fee on winning trades. neg_risk=True for weather

#2% Fee
Explore →