Evidence first.

How rows are scored

• The baseline gives every valid combination the same probability.
• Long, medium and short frequency, gaps and same-month history are estimated with strong statistical shrinkage.
• Noisy effects are pulled back towards zero and uncertain estimates receive an additional penalty.
• Number-pair effects are used only when they repeat across separate parts of the historical record.
• Rows with lower estimated sharing risk are preferred separately from draw probability.
• The portfolio optimiser rewards unused numbers, new pairs and new near-winning higher-order subsets across multi-game selections.
• The selected portfolio is displayed from highest to lowest base model score.

The evidence gate

The signal model learns from earlier draws, then predicts a later untouched period. It must improve the proper Brier probability score with a positive 95 percent confidence bound before any historical draw signal is switched on.

Back-testing samples unseen draws across the available history. It compares match rates and probability scores with the fair-draw baseline.

Sharing and coverage

Player habits such as birthdays, lucky numbers and obvious patterns do not affect which balls are drawn. They can affect how many people share a pooled prize. This is why sharing value remains separate from probability evidence.

Coverage cannot make one valid row more likely to win. Across several rows, it reduces duplicated coverage and increases the number of distinct high-match outcomes represented by the published portfolio.

Rows stay fixed until verified history, lottery rules or the scoring method changes. Weather remains excluded unless exact pre-draw data demonstrates reliable value on unseen draws.