Understanding Ball Colours

Every number in this app is colour-coded to give you an instant read on its recent behaviour. Here is exactly what each colour means:

Hot and cold thresholds are recalculated dynamically every time the app loads, based on the last 200 draws. They shift week by week as new results come in. A number labelled hot today may return to gold next month if its frequency normalises.

Data Source

2,071 real verified Saturday Lotto draws (Draw 533, 30 Aug 1986 → Draw 4673, 2 May 2026) are permanently embedded in this app. Source: australia.national-lottery.com, cross-verified against Lotterywest official government API and au.lottonumbers.com. Zero synthetic or generated data is used anywhere in analysis or prediction. Tap Sync to pick up each week's new draw — no other downloads needed.

Prediction Method Overview

Predictions use a weighted ensemble of three statistical signals derived from the full historical dataset, combined with empirical balance constraints. The model scores each of the 45 possible balls, then finds the highest-scoring combination that passes all balance filters.

Signal 1 — Frequency Deviation (40%)

Mathematical basis: Law of Large Numbers + observed deviation from expected frequency.

Over 2,071 draws, each number is expected to appear 276.1 times (2071 × 6 ÷ 45). Numbers appearing significantly above this baseline receive a higher score. For example, ball 1 has appeared 316 times (+14.4% above expected), while ball 2 has appeared only 242 times (−12.4%).

The chi-square statistic across all 45 balls is 48.10, below the 95% confidence critical value of 60.48 — meaning the overall draw distribution is consistent with randomness. However, individual ball deviations are used as a weak directional signal.

Honest caveat: Individual deviations at this sample size may be pure noise. A statistically conclusive mechanical bias would require ~10,000+ draws.

Signal 2 — Gap Ratio (30%)

Mathematical basis: Personal return cycle — current gap vs historical average gap per number.

Each ball has its own average gap — the mean number of draws between consecutive appearances. Ball 1 averages a gap of 6.6 draws; ball 38 averages 7.3 draws. If ball 1 has not appeared in 13 draws, its gap ratio is 13 ÷ 6.6 = 1.97, meaning it is nearly twice as overdue relative to its own historical rhythm. This ratio is capped at 3× to prevent extreme outliers from dominating.

This is distinct from the Gambler's Fallacy. Rather than claiming a ball is "due," it measures relative deviation from each ball's own observed pattern.

Honest caveat: Each draw is an independent event. A ball 3× overdue has exactly the same probability of appearing as one drawn last week. The gap ratio is observational, not causal.

Signal 3 — Pair Affinity (10%)

Mathematical basis: Co-occurrence frequency vs baseline expectation.

With 45 balls choosing 6, any specific pair has a baseline appearance probability of approximately 14.3% per draw. Pairs that appear together significantly more often than this baseline across 2,071 draws receive a small score boost. The most common pair in the dataset is 11+32, appearing 53 times vs ~43 expected.

Honest caveat: With 990 possible pairs, some will appear elevated purely by chance (multiple comparisons problem). This is the weakest signal in the model and is weighted accordingly at 10%.

Balance Constraints (enforced on every combination)

Mathematical basis: Empirical distribution matching from 2,071 historical draws.

Every generated combination must pass three filters derived from the actual historical record:

Odd/even split: 2–4 odd numbers required. The most common split across 2,071 draws is 3 odd / 3 even (32.4%), followed by 4 odd / 2 even (26.0%) and 2 odd / 4 even (22.8%). Combinations with all odd or all even are extremely rare (1.0–1.2%) and are excluded.

Low/high split: 2–4 low numbers (1–22) required. The distribution mirrors odd/even: 3 low / 3 high (32.6%), 2 low / 4 high (24.6%), 4 low / 2 high (23.2%). All-low or all-high combinations each occur in less than 1.2% of draws and are excluded.

Sum range: The six main balls must sum between 115 and 160. The historical mean sum is 137.2, and this range covers 53.6% of all draws. Extreme sums (very low or very high) are statistically unlikely.

12-Combination Coverage Strategy

Mathematical basis: Combinatorial coverage maximisation.

When generating multiple combinations for a single ticket, the engine enforces a diversity constraint: no two combinations in the set may share more than 2 numbers (for tickets of 12 or fewer games). This maximises the total number space covered across your ticket, reducing the chance that all 12 combinations cluster around the same numbers.

The combinations are ranked by total score and selected greedily — the highest-scoring valid combination that satisfies the diversity constraint is added first, then the next-highest, and so on. For larger tickets (18–25 games), the maximum shared constraint is relaxed to 3–4 to allow the count to be filled.

What This Model Cannot Do

Saturday Lotto draws are designed to be independent, identically distributed random events. Formally, the probability of any specific ball appearing in the next draw is always 6/45 = 13.3%, regardless of all previous draws. No statistical model can change this fundamental property.

What this model does instead: it produces combinations that are statistically consistent with historical draw patterns, prioritising numbers with above-expected frequency and above-average gap ratios. This does not increase the probability of winning Division 1, but it does produce combinations that are less likely to be chosen by other players (avoiding birthdate-heavy numbers like 1–31), which would improve your share of a prize if you did win.

Install as App

On iOS: tap Share → Add to Home Screen.
On Android: tap ⋮ → Add to Home Screen or Install App.