Fit to the random model Lotto
Do the results deviate from the uniform and independent model?
What would you like to check?
Enter or load your numbers in the analyzer. The latest draw report describes the actual draw, not your selection.
Explore all statistics — guided topics
These pages describe the history of this game. They do not predict future numbers. Use the selection tool on a page to compare your own numbers.
Numbers and gaps
Start with frequency and gaps: how often each number appeared in the selected historical sample.
Draw structure
Compare sum, parity, endings and range. On a page with a selection comparison, open it to check your numbers.
Relationships and random model
Explore pairs, triples and partners, then the random model. Historical relationships do not improve future odds.
Draws and tools
Read the latest draw report for the actual drawn numbers. Use the analyzer and matching draws for your own selection.
- Game
- Lotto
- Number pool
- 49
- Selected range
- 500 most recent
- Draws in the sample
- 500
- From
- 2023-07-25T20:00:00Z
- To
- 2026-10-01T20:00:00Z
- Latest draw in the sample
- 2026-10-01T20:00:00Z
- Calculated at
- 2026-10-01T23:56:26+02:00
- Source
- Service draw archive
- Freshness
- Checking…
Agreement of frequencies with the random model
The uniform, independent model assumes that each of the 49 numbers in every Lotto draw had the same 6/49 chance of being drawn. It also assumes that the result of one draw did not depend on the previous ones. This page checks the first condition with a frequency test; independence is checked by separate analyses (autocorrelation, gaps).
Frequency statistic T (N = 500 draws, source: history)
65,3
expected under the model: df = 48
How many deviations from expected: (T − df) / √(2·df)
1,77
source: χ²(df) model; a positive value — frequencies more spread out than expected, negative — more even
P-value (asymptotic, indicative)
0,049
percentile calibrated by simulation: —
Simultaneous band — did any number fall past the threshold
Largest deviation of a single number: max |z| = 2,76 (N = 500, source: history). 95% band threshold: —, 99%: —.
Simulation calibration of the band thresholds is in preparation — without it we do not judge whether max |z| is unusual.
All windows side by side
Each row is a different window of recent draws. Windows overlap (the last 50 includes the last 25), so they are not independent samples.
| range | N | T | df | (T−df)/√(2df) | p | max |z| |
|---|---|---|---|---|---|---|
| last 1 | 1 | 48,0 | 48 | 0,00 | — | 2,68 |
| last 7 | 7 | 44,3 | 48 | -0,38 | — | 3,62 |
| last 14 | 14 | 39,1 | 48 | -0,91 | — | 1,86 |
| last 25 | 25 | 48,4 | 48 | 0,04 | 0,455 | 1,87 |
| last 50 | 50 | 40,0 | 48 | -0,82 | 0,788 | 2,10 |
| last 100 | 100 | 36,7 | 48 | -1,15 | 0,882 | 2,21 |
| last 250 | 250 | 52,3 | 48 | 0,44 | 0,311 | 2,43 |
| last 500 | 500 | 65,3 | 48 | 1,77 | 0,049 | 2,76 |
| all | 6,547 | 59,5 | 48 | 1,18 | 0,123 | 3,12 |
Windows without a row in the table have no data — the reason is shown by the range selector.
How we calculate this and what it doesn't tell you
T is the sum over all 49 numbers of the squared deviation (O − E)² divided by the variance N·p·(1−p), with the (m−1)/m correction, where O is the number of occurrences in N draws, E = N·p, p = 6/49, m = 49. Under the uniform model T has approximately a χ² distribution with df = 48 degrees of freedom, so the expected value is df and the deviation √(2·df). The p value from that distribution is asymptotic — for short windows the approximation is weak, hence the percentile calibrated by simulation (in preparation), computed as the share of simulations with T strictly smaller than the observed one.
A result consistent with the model does not prove randomness — the test may miss departures of a kind other than uneven frequencies. Watching many overlapping windows raises the risk of an accidental alarm: at a threshold of 0.05 a false signal in one of the windows is to be expected. Do not assume the windows are independent. None of these numbers shows what will come up next time.
Four model tests on this site: frequency (T, this page), the simultaneous band (max |z| against the c95 threshold, this page), gaps between occurrences of a number (geometric distribution, a separate analysis) and autocorrelation of draws (a separate analysis).
Interactive tools
Explore the same data actively. The tools describe history and the random model; they do not predict the next result.
My selection — check its properties against history
This describes the entered selection and its position in the historical sample. It does not increase its chance.
History player — move through past draws
Choose a year, use the slider or start automatic playback. This is a presentation of recorded results, not a forecast.
Open the module to load the available years.
Random-model simulator — see ordinary random variation
Generate independent draws with the rules of this game and compare observed counts with the common expected value. Every run is different.
What do these numbers show?
- In 500 draws the frequency statistic T came to 65,3 with 48 degrees of freedom; the uniform model gives an expected 48 and a deviation of 9,8, so T lay 1,77 deviations from the expectation (source: history and model).
- The asymptotic p value came to 0,049 — below the conventional threshold of 0.05. This is an indicative result; it proves neither randomness nor its absence.
- A p value below 0.05 showed up in 1 of 6 windows with data. The windows overlap, so they are not independent — with that many comparisons a single alarm is nothing unusual.
These are historical data. Each draw is independent of previous draws, so these numbers do not tell us what will be drawn next.
How do I use this page?
- Choose a number of recent draws or a range of years.
- Read T and how many deviations separate it from the expected value df — that is the scale of the divergence between the frequencies and the model.
- Treat the p value as indicative; below 0.05 is a conventional boundary, not proof.
- In the window table, check whether the divergence holds across ranges, remembering that the windows overlap.
What this page does not do: it does not predict future results, does not point to numbers to play and does not improve your chance of winning.
How to read statistical measures
Random model
Expected value
Deviation
Percentile
Typicality
Related analyses
Frequency test
Does the distribution of hits for all numbers match the model?
Simultaneous band
Did any number fall outside the band calculated for all numbers at once?
Gaps vs. model
Do gap lengths look like a geometric distribution?
Sum autocorrelation
Does a draw's sum depend on the sums of previous draws?