CWG INC applies statistical modeling, machine learning, and 26 years of draw history to surface non-random structural patterns in lottery outcomes.
Not fortune-telling. A reproducible, peer-reviewed statistical pipeline.
All 1,127 official draw results since 2002 are ingested via the DHLOTTERY public API, normalized, and stored in a versioned PostgreSQL warehouse.
47 statistical features are computed per round: frequency distribution, odd/even ratio, sum range, consecutive pairs, zone coverage, and recency decay.
Random Forest, LSTM time-series, and Gradient Boosting models are trained independently and combined via a weighted stacking meta-learner.
Candidates pass a 200-round back-test filter. Only combinations meeting confidence threshold ≥ 0.70 are surfaced to users as weekly picks.
Every analytical capability used internally, exposed through a clean mobile interface.
Auto-updated every Saturday post-draw. The model retrains on the latest result and surfaces a new pick within 2 hours.
Visualize draw frequency for all 45 numbers across any date range.
Each pick is scored against 4,891 historically significant patterns.
Optimal spread across 1–10, 11–20, 21–30, 31–40, 41–45 zones.
Draws indexed automatically. Model retrained within 2 hours of each result.
Aggregate 51K+ users into a consensus model — collective intelligence as a signal.
Raw signal extracted from 1,127 rounds of official lottery data.
Cumulative draw appearances across 1,127 rounds. Green = above average, Purple = below average.
Rolling sum of 6 winning numbers. Mean ≈ 138.
3-odd/3-even appears in 31.2% of draws.
Color intensity maps to draw frequency. Hover for exact count.
We don't claim lottery is predictable in the absolute sense. We identify statistically significant deviations from perfect randomness — a measurable edge, not magic.
Every model release is validated against the last 200 rounds before deploying to production. Underperforming models are discarded.
We publish hit-rate by prize tier. No cherry-picked samples. Historical model performance is permanently archived.
All data sourced exclusively from DHLOTTERY public API. No third-party, scraped, or manipulated data enters the pipeline.
Models never run stale. Automated CI/CD pipeline triggers retraining within 2 hours of each Saturday draw result.
Free tier includes weekly AI picks, frequency heatmap, and 5 saved combinations. No credit card required.