Dashboard
Most recent draw
Does tonight’s date mean anything?
A ball looks hot here because it won a 52-way contest for the title, not because the date favours it. Pick the winner after seeing the results and almost any slice of history will look remarkable.
Most frequent · vs expected
Least frequent · vs expected
Hottest (last 20 draws)
Coldest (longest current absence)
Analyze
Analyze
Every ball against expected · all 52 at once
Longest unseen (“overdue”)
Draws are independent events — past frequency has no effect on future draws, and nothing is ever “due”. Shown for interest only.
Number frequency engine (1–50)
Click any column header to sort. Gaps and streaks are measured in number of draws.
History
Sets for the next draws
Scored from the calendar crosses: each date's weekday, day of month and numerological root, against every ball's record under those same conditions. Opens standalone: forecast.html
Or draw at random
Chart 1 — Generated numbers: historical appearances
This generator is a fair uniform draw. Showing historical frequency of generated numbers is purely informational and does not make any combination more or less likely to win. Over 400 replayed draws the scored sets above did not beat it.
Visualizations
Chart 2 — All numbers frequency distribution (1–50)
Chart 3 — Hot vs Cold numbers
Red = above-median frequency in window (hot), Blue = below-median (cold).
Chart 4 — Number trend over time
Pattern Detection
Odd / Even ratio
Low (1–25) / High (26–50)
Consecutive numbers
Sum analysis
Pattern Explorer
Every observed pattern — pairs, triplets, single balls, digit-roots, date numerology, draw days — counted for whatever date range you pick, with era-correct expectations. Opens standalone: pattern-explorer.html · significance tests: pattern-tests.html
Combination Analysis
Most common pairs
Most common triplets
Co-occurrence heatmap
Darker = the two numbers appeared together more often. Hover a cell for the count.
AI Research Engine
A transparent multi-factor scoring model: frequency 45%, recency 20%, gap/overdue 20%, trend 15%. The weights are applied to standardised components — each divided by its own spread — because a weight only equals influence when the things it multiplies vary by the same amount, and these do not. The measured shares are printed under the table. Three illustrative strategies:
Composite score breakdown (top 15)
These sets are produced by weighting historical statistics. Because draws are independent and random, no weighting scheme can beat chance. Treat the output as a study of the historical data, not advice.
Machine Learning & Backtesting (Research Scaffolding)
Heavyweight models (Random Forest, XGBoost, LSTM, Transformer) live in the optional Python backend. In this browser app we run transparent statistical baselines under proper walk-forward validation so you can see the central result clearly: no strategy meaningfully beats random selection.
AI Chat Assistant
Ask about the historical database. Try: hottest numbers · most overdue · top 20 pairs · analyze number 19 · generate a balanced recommendation