Pattern test
Pattern tester
Enter a pattern such as BBBP. The tool counts it across as many random shoes as you like: how often it appeared, how often banker came next, how far that sits from theory, and — the part that matters — whether the gap survives correction.
Method
Method
Shoes are dealt in your browser with eight decks and the full drawing rules; ties are skipped and do not enter the pattern. The count then goes through a one-proportion test, corrected by Bonferroni across all 2^k patterns of the same length. That last step is the whole point: finding one p below 0.05 among thousands of candidates is guaranteed to happen, and is not a discovery.
How to read it
How "significant" gets manufactured
The problem is not what you found, it is how many times you looked
p < 0.05 means that if a pattern has no predictive power, chance alone still makes it look this striking 5% of the time. Test one pattern and that threshold is reasonable. Nobody tests one. There are 64 patterns of length six; sweep all of them and you should expect 3.2 of them to come back "significant" — in data that is purely random. The chance of at least one is 96.2%.
This is where road patterns come from
Read the table backwards: anyone who sifts enough hand histories will certainly find some pattern after which banker shows up unusually often. Finding it requires no underlying regularity, only enough candidates. This is not a flaw in statistics — it is statistics working correctly, telling you that the discovery should not surprise you. What would be surprising is if it survived correction.
What the Bonferroni correction does
It divides the threshold by the number of candidates. For patterns of length six that makes it 0.00078. This is severe, severe enough to miss small real effects — and here that is the right trade, because we know in advance that no information exists in the pattern: the shoes are randomly generated. Anything that fails the correction should be discarded. The tool shows you the verdict before and after correction so you can watch the difference appear.
| Pattern length | Candidates | Expected "significant" | P(at least one) | Bonferroni threshold |
|---|---|---|---|---|
| 3 (e.g. BBP) | 8 | 0.40 | 33.7% | 0.0063 |
| 4 (e.g. BBPB) | 16 | 0.80 | 56.0% | 0.0031 |
| 5 | 32 | 1.60 | 80.6% | 0.0016 |
| 6 | 64 | 3.20 | 96.2% | 0.00078 |
| 7 | 128 | 6.40 | 99.9% | 0.00039 |
| 8 | 256 | 12.80 | > 99.99% | 0.00020 |
Single-test threshold α = 0.05. Expected count = 0.05 × 2^k; at least one = 1 − 0.95^(2^k). There is no predictable structure in the data — every one of these is a false positive.
Frequently asked
- Do road patterns really carry no predictive power at all?
- In a properly shuffled shoe, none. The reason is mechanical rather than statistical: the probability of each hand depends only on the composition of the remaining cards, and whether the previous hand went to banker or player barely changes that composition. The one real memory effect comes from dealing without replacement, and it sits in the fourth decimal place — orders of magnitude below the house edge.
- What if a result stays significant after correction?
- Rerun it with a different number of shoes. The corrected threshold for length six is 0.00078, which random data will still trip occasionally — that is what a threshold means, as opposed to a guarantee. If it fires every single time, the generator is broken. If it fires once in ten runs, you met the once.
- Why are ties skipped?
- Because that is how the road is recorded: on the big road a tie is drawn as a slash and does not break the banker/player sequence. If the object is to test road patterns, the counting has to follow the road’s own rules. Ties occur on 9.52% of hands, and including them would make pattern length mean something different from what a player actually sees.
- Will this tool help me find a pattern that works?
- No — it is built to do the opposite. It walks you through sweep, discovery, correction and collapse so you can watch a "rule" being manufactured out of random data. If you use it on real hand histories and find something, this table is the reminder to ask first how many candidates you looked at.