How our Polymarket works
End-to-end pipeline for Catalysm bots on 15-minute crypto direction markets.
Catalysm bots trade crypto direction markets on Polymarket: binary contracts that resolve to “up” or “down” over a fixed interval. This page is the map. Each stage has its own doc linked below.
What the bot trades
The flagship market is 15-minute Bitcoin up/down. Every 15 minutes a new market opens. The bot decides whether BTC will finish the interval higher or lower than its opening price.
| Property | Value |
|---|---|
| Asset | BTC |
| Interval | 15 minutes |
| Market type | Up / Down |
| Platform | Polymarket |
Two live instances run in parallel with different parameter sets so you can compare strategies on the Bots page.
Pipeline at a glance
Each window runs four stages inside the bot. A fifth stage, the weekly agentic loop, tunes parameters between windows.
flowchart LR
A[Strategy] --> B[Historian]
B --> C[Judge]
C --> D[Execution]
D --> E[Agentic loop]- Strategy: Pull BTC candles and Polymarket book data. Run analysis modules and indicators. Each signal returns a bias and a confidence. Weighted averaging produces a directional score.
- Historian: Read the bet log and compute rolling win rate, pattern stats, and calibration. At runtime this is a snapshot for logging; the deep review runs weekly.
- Judge: Apply gates (spread, dead zone, confidence, expected value, pattern rules). Pass every gate, or the bot skips the window with
NO_TRADE. - Execution: Place a fill-or-kill order on Polymarket when the judge approves a side. Log the decision either way.
- Agentic loop: Weekly agents observe live performance, propose changes to most bot parameters (weights, gates, sizing, risk limits), and redeploy approved updates.
The bot fires between 5 minutes and 30 seconds before each 15-minute window so it can read the book and enter before the interval starts.
One window in plain terms
- Gather 15-minute and 1-hour BTC candles plus the active Polymarket order book.
- Compute indicators and run analysis modules (patterns, trend, volume, mean reversion, microstructure).
- Read historian stats from the bet log for context.
- Combine every signal into a weighted score, then run judge gate checks.
- Size the bet from setup grade and edge, then submit the order or log
NO_TRADE. - After resolution, update outcomes in the bet log for stats and the weekly review.
Only trades that actually filled on-chain count toward performance stats.
Where to go next
- Strategy: Indicators, weights, and how the score is built
- Judge: Gates, expected value, and position sizing
- Historian: Bet-log analytics and calibration
- Execution: Polymarket orders, paper vs live, guards
- Agentic loop: Weekly observe → optimize → deploy cycle