Strategy Description

⚠️ Important Disclaimer: This is a theoretical research prototype, not a complete trading strategy

The current code is only a theoretical/research prototype for probability modeling. It is not a validated, production-ready trading strategy and must not be treated as such. Do not use it with real funds before completing sufficient independent validation.

---

1. What is this?

This is an experiment in independent terminal-probability modeling for Polymarket’s 15-minute BTC Up/Down binary market.

The core idea is to:

• Use Binance BTC/USDT tick-trade flow together with Polymarket RTDS data, including Chainlink and TWAP-60;
• Use the official benchmark price-to-beat, the current BTC reference price, remaining time, and forecast volatility to calculate a terminal direction probability that is independent of the Polymarket order book;
• Compare the probability with the order-book Ask price using EV, defined as the absolute expected profit per token, to identify potential mispricing;
• Use multiple entry gates, including a Kelly fraction, EV threshold, two-sided order-book consistency, and round-trip loss;
• During the holding period, exit based on a predictive EV shock combined with a fixed price stop-loss or take-profit.

2. Current status and known limitations

Dimension	Current status
Probability model	Terminal probability is based on a lognormal approximation (terminal_probability). Volatility is estimated using multiple empirical windows. The model has not been calibrated through historical backtesting, and many parameters are still prior assumptions.
43-feature Logistic model	The model exists, but legacy43Trading=false, so it does not currently participate in the real-time trading probability. It is only stored for research purposes.
Online calibration	ONLINE_CALIBRATION=false. The previous 600-second snapshot calibration has been disabled, while the new stage-based sampling mechanism is still accumulating data and has not yet become active.
Evolution engine	EVOLUTION_ENABLED=false. The Champion/Challenger significance-testing framework is not active.
Parameter calibration	Many parameters, including volatility floors and caps, drift shrinkage, EV thresholds, and stop-loss distances, are still magic constants without sufficient statistical calibration.
Live-trading constraints	IsVirtual() is enforced to reject backtesting. Redemption, fill, and slippage logic have not yet been sufficiently validated under real execution conditions.
Out-of-sample validation	There are currently no systematic out-of-sample or forward-testing results that support any expectation of profitability.

3. Explicit risks and disclaimer

• This is a methodological exploration designed to test the hypothesis that independent probability estimates can consistently identify positive EV relative to market pricing. The hypothesis itself has not been proven.
• The core parameters of the forecasting model have not been fitted to sufficient data. Live performance is highly uncertain and may result in substantial losses.
• Tail risks related to binary-contract settlement, disappearing liquidity, order-book manipulation, and feed latency have not been fully modeled.
• The large number of TODO-style default-off switches indicates that the system is still an incomplete prototype.

---

Conclusion

This project should be treated as a research-framework skeleton that still requires validation.

Before its real trading value can be assessed, the following work remains necessary:

• Historical backtesting;
• Parameter calibration;
• Out-of-sample validation;
• Forward testing;
• Small-size live stress testing.

Only after these steps can the system’s potential value as a real trading strategy be evaluated.