Win Rate Predictor

The Win Rate Predictor models the probability of an MMBot configuration hitting stop loss, and the probability of profiting if it does not. Trained on over 205,000 real executed trades across 182 pairs and 11 exchanges over a 90-day window, the model is built on real fills, real slippage, and real outcomes — not backtests or simulations.

⚠ Caution

This is a probabilistic tool. It does not guarantee outcomes.

Reliability is influenced by sample size and market conditions. Configurations with strong historical matches in similar market conditions will produce more dependable results. Low sample sizes, unfamiliar configurations, or atypical market events — such as liquidation cascades, deepegs, or exchange outages — may reduce output reliability.

Use this as one input in your decision-making process, not a deciding factor.

Sample Size

Sample size reflects the number of historical trades that match the configuration entered. A larger sample gives the model more real-world evidence to draw from, improving result reliability.

Configurations with fewer than 3 matching samples will not appear in the Best Configurations suggestions.

Volatility

Volatility is calculated in real time by fetching the last hour of 1-minute candles at the moment the tool is run. The model uses this to compute realized volatility and classify current market stability as calm, steady, choppy, or volatile.

This is why the same configuration may return different results depending on when it is run.

Find Best Configuration

The tool generates 10,800 configuration combinations across bot type, spread, leverage, stop loss, take profit, duration, and directional bias. Each combination is scored against the user's selected pair, exchange, and live volatility using a composite formula.

Stop loss avoidance accounts for 60% of the score. Profit potential — conditional on not hitting stop loss — accounts for the remaining 40%. The top 5 results are filtered for diversity to avoid surfacing variations of the same configuration.

Retraining Schedule

The model is retrained manually every 2–4 weeks using the most recent 90 days of trade data. This cadence captures shifting market dynamics and newly added pairs or exchanges without overfitting to short-term noise. Real-time volatility inputs keep the tool responsive between retraining cycles.

*Updates*

• Grid Rest Threshold + TP Reset — Users can now configure grid reset threshold and TP reset threshold for grid and RGrid bot and consider it as a factor when running the win rate predictor. Find best config will now also provide the best grid reset threshold and TP reset threshold for that exchangeXpair.

• Pair Search Bar — In Bot Configuration, the Pair field now supports text search. Type to find a pair directly instead of scrolling through the full list.

• SL Probability Display — Results with a stop loss probability below 1% will now display as >1% instead of 0% to better reflect actual model output.