Study 01 · Fairness
Every era of Florida Fantasy 5 tested for number bias — 13,254 drawings, four matrix eras, one chi-squared test each.
A lottery is supposed to be a fair machine: over enough drawings, every number should appear about as often as every other. "About" is doing a lot of work in that sentence, and it is where most lottery analysis goes wrong. In 10,397 drawings of a 5-from-36 game, some numbers will lead and others will trail by hundreds of appearances purely by chance. The question is whether the gap is bigger than chance can explain.
We counted how often every number appeared in every era of the game, then ran a chi-squared goodness-of-fit test against the flat distribution a fair machine produces. A low p-value means the pattern is too lopsided to be luck. A high p-value means what you see is ordinary noise. Each era is tested separately, because the game's matrix changed three times and mixing eras would invent a bias that isn't there.
| Era | Dates | Draws | Chi-squared | df | p-value |
|---|---|---|---|---|---|
| Era 1 · 5/39 | 1989-05-02 – 1993-04-28 | 558 | 27.3 | 38 | 0.90 |
| Era 2 · 5/26 | 1993-04-30 – 2001-07-15 | 2,299 | 16.3 | 25 | 0.91 |
| Era 3 · 5/36, Evening | 2001-07-16 – 2023-03-19 | 7,917 | 31.6 | 35 | 0.63 |
| Era 4 · 5/36, Midday + Evening | 2023-03-20 – 2026-08-10 | 2,480 | 23.2 | 35 | 0.94 |
Every era passes comfortably. The lowest p-value is 0.63, far above the 0.05 threshold where a statistician starts to suspect something real. In the current 5/36 era the most-drawn number appeared in 14.5% of drawings and the least-drawn in 13.1%, against 13.9% expected — a gap that looks meaningful in a table and disappears the moment you test it.
Florida added a midday drawing in 2019, which gives a natural experiment: two separate machines, two separate draw crews, same rules. Comparing 1,240 midday drawings with 9,157 evening drawings number by number, the count that differ at the 99% level is 1 (number 13). Testing 36 numbers at that level, roughly one false positive is exactly what you expect to see. In other words: the two drawings behave identically.
A chi-squared test compares what happened with what a fair machine would produce, and converts the gap into a single number. The p-value is the plain-English part: it is the chance of seeing a gap at least this big if the machine is fair. A p-value of 0.63 in the current era means a fair machine would produce a spread at least this uneven about 63 times in 100. Nothing to explain, in other words. Researchers usually start investigating below 0.05; no era here comes close.
In the current 5/36 era the top and bottom of the table look like this:
| Number | Times drawn | Share of drawings |
|---|---|---|
| 35 | 1,498 | 14.4% |
| 27 | 1,497 | 14.4% |
| 16 | 1,493 | 14.4% |
| 7 | 1,491 | 14.3% |
| 1 | 1,490 | 14.3% |
| … | ||
| 8 | 1,389 | 13.4% |
| 13 | 1,394 | 13.4% |
| 23 | 1,399 | 13.5% |
| 30 | 1,402 | 13.5% |
| 26 | 1,405 | 13.5% |
The gap between first and last is 109 appearances over 10,397 drawings — the kind of spread a fair machine produces routinely. Any site can rank these numbers and call the top ones "hot". Whether that ranking predicts anything is a separate question, and we tested it.
Most "number analysis" is built on the assumption that a game is quietly biased and that counting past results can find the bias. For this game, over this record, the assumption is false. That is not a disappointing result — it is the foundation for everything else we do. Once you accept that the numbers are fair, the only honest questions left are about combinations: which shapes of ticket actually occur, and how people's picking habits compare. That is the subject of the shape study.
Florida Lottery official Fantasy 5 results, 13,254 draws from 1989-05-02 to 2026-08-10, validated for completeness and tagged by matrix era and draw type. The figures on this page are produced by a script that runs against that record, so they can be recomputed from the same source at any time. Last updated September 22, 2026.
What this does not mean. Nothing on this page improves anyone's chance of winning. Every draw is independent, and a fair game cannot be beaten by studying it. We publish these results because they are true, including the parts that work against our own product.