Lottery Randomness Auditor
A reproducible statistical audit of published UK Lotto winning numbers, testing for detectable deviations from fairness era by era.
Details
- Started
- 21 Oct 2026
- Data
- 3,256 published draws, 1994 to 2026, pinned and checksummed
- Battery
- 9 test areas with Monte Carlo references
- Output
- HTML and PDF report, marimo notebook
- Reproducible
- Same seed, byte-identical report
- Language
- Jupyter Notebook
- Last commit
- 23 Oct 2026
ConjecturePublished UK Lotto results are indistinguishable from fair random draws when each era is tested against its own rules.
FindingStill open.
The write-up
The auditor tests the published winning numbers only. A pass means the battery found no convincing deviation in the snapshot after multiple-comparison control. It is not a guarantee about machines, ball sets, procedures or future draws.
What the battery checks
| Area | Question |
|---|---|
| Frequency | Are any balls drawn too often, or not often enough? |
| Homogeneity | Do same-night rounds come from the same process? |
| Pairs | Are any ball pairs unusually sticky or avoidant? |
| Runs | Is the sequence too streaky or too regular? |
| Serial dependence | Does one draw leak information into later draws? |
| Gaps | Do waiting times between appearances match the fair geometric pattern? |
| Order statistics | Do the minimum, maximum and range follow fair reference curves? |
| Spectral | Is there a hidden cycle in ball appearances? |
| Bayesian uniformity | How far are estimated ball probabilities from the fair share? |
Eras are never pooled
The game’s rules define what fair looks like, and they have changed over time: 49 balls until 2015, 59 balls after, and two rounds per draw from mid-2026. Each era is tested against its own reference.
The report explains every chart in three parts: what the plot shows, what fair looks like, and what would be worrying.