Features

14 Statistical Prediction Engines

Each engine applies a distinct statistical methodology to your spin history, giving you a multi-angle view of pattern probability rather than a single number to blindly follow.

• Hot/Cold Number Frequency — Tracks which numbers and sections appear above or below their expected frequency over your chosen lookback window.
• Markov Chain Analysis — Models spin-to-spin transition probabilities. Detects sequential patterns that pure frequency counting misses.
• Bayesian Probability — Updates prior probability estimates with each new spin, weighting recent data more heavily as evidence accumulates.
• Gap/Drought Analysis — Measures how long each number has gone without appearing. Identifies numbers statistically overdue relative to their expected recurrence interval.
• Neighbour Clustering — Analyses the physical wheel layout. Detects whether results cluster around specific wheel sectors — a classic indicator of mechanical bias.
• Chi-Square Goodness of Fit — Tests whether observed spin distribution deviates significantly from a perfectly fair wheel (uniform distribution).
• Standard Deviation Scoring — Quantifies how far any number or section has drifted from its expected mean across multiple time windows.
• Trend Momentum — Short-window exponential weighting to detect emerging streaks before they fully develop.
• Entropy Analysis — Measures randomness of the spin sequence. Low entropy suggests non-random patterns worth investigating.
• Z-Score Ranking — Normalises deviation scores across all 37 or 38 numbers so engines are directly comparable regardless of sample size.
• Sector Bias Detection — Groups numbers into wheel thirds and halves. Detects lopsided results at the sector level before individual number analysis.
• Cycle Detection — Looks for repeating sub-sequences in spin history using autocorrelation.
• Regression to Mean — Flags numbers that have recently deviated sharply and may revert toward expected frequency.
• Composite Score — A weighted aggregate of all other engines, calibrated to your session length and table sample size.

6 AI & Machine Learning Models

Machine learning models run entirely on your device using TensorFlow Lite. No data ever leaves your phone.

• XGBoost (Gradient Boosting) — Gradient-boosted decision tree model. Excels at finding non-linear interactions between features extracted from spin history.
• Transformer-LSTM Hybrid — Combines the sequential memory of LSTM with the attention mechanism of Transformers. Best suited for detecting long-range dependencies in extended sessions.
• Convolutional Neural Network (CNN) — Treats the spin history as a 1D signal. Detects local pattern motifs (short repeating runs) across the sequence.
• Random Forest Ensemble — Averages predictions across hundreds of decision trees. Robust to overfitting on small samples.
• LSTM (Long Short-Term Memory) — Pure recurrent model. Learns which historical patterns to retain and which to discard based on your session data.
• Ensemble Meta-Model — A second-layer model trained on the outputs of the other five. Produces a calibrated final prediction by learning which base models perform best on your specific data profile.

Session & Casino Management

• ✓Track multiple simultaneous sessions across different tables
• ✓Bankroll tracking and net gain/loss per session
• ✓Per-table history with full spin logs
• ✓Casino Finder — locate nearby casinos using Google Places
• ✓Export session data as JSON for your own analysis
• ✓Delete all data at any time — one tap

Privacy by Design

All spin data, session history, and AI inference runs locally on your device using an encrypted SQLite database. No account is required. No data is ever transmitted to our servers or any third party (except anonymous crash reports via Firebase Crashlytics, which contain no spin or casino data).