Seasons, decades and what follows what.

Every published prize slot in the archive — 718,743 of them across 9,134 draw dates and 31,260 operator boards — sorted by when it was drawn. The test is whether the MIX of digits and digit shapes differs by period, never how many draws a period holds.

That distinction matters: the government approves extra special draws around the festivals, so festival weeks genuinely hold more draws. Counting draws would "find" a CNY effect that is only the draw calendar. Each test below conditions on the period's own slot count, so only the shape of the numbers is on trial.

Month of year × digit

below expectedabove expectedEvery cell prints its own value — the colour is only a shortcut.

Festival windows

Hari Raya moves about 11 days earlier every Gregorian year, so it cannot be approximated by a fixed month — that would silently be testing "September". The 43 anchor dates here are 1 Syawal from the tabular Islamic calendar, which is arithmetic and therefore reproducible. Malaysia's official date is declared by moon sighting and can land a day or two either side of it; the ±7-day window is wide enough to cover that slack, and wide enough to cover the whole holiday period.

Festivals we did NOT test, and why: Deepavali (Hindu lunisolar — needs an ephemeris this site does not carry) · Thaipusam (same reason) · Hari Raya Aidiladha (same machinery, but it would add two more tests to the batch for a window the folklore does not name)

Start, middle and end of the month

All four slices use the same universe and the same two tests, so they are directly comparable. Nothing in this section is a prediction: even a real seasonal effect would tell you about the past, not about tonight.

Every published prize slot in a decade, pooled across all operators and all prize tiers, ranked. This section is entertainment — it is a scoreboard of what already happened, and it carries no information about what comes next.

Archive depth is NOT the same across decades and the counts are not comparable between them: the 1980s hold only Magnum from April 1985, the 2010s added the East Malaysia operators, and the 2020s added Grand Dragon's nightly draw. Each block prints its own board count, slot count and operator list so you can see exactly what it is built on.

31,167 consecutive pairs of 1st-prize numbers across 10 operators. Each row is the leading digit of one draw, each column the leading digit of the very next draw by that same operator. If the machine had any memory at all, some cell in this grid would be too warm.

below expectedabove expectedEvery cell prints its own value — the colour is only a shortcut.

Chains never cross operators, and never cross a hole in the archive: two draws more than 14 days apart are not consecutive draws, so the pair is dropped instead of chained — 21 pairs went that way. 2 draw dates carrying more than one 1st prize were dropped too; that is a filing problem, not a night with two winners. Operators with fewer than 500 usable pairs are left out entirely, because a 10×10 table needs about five transitions per cell before its chi-square means anything.

The flatness IS the finding. A fair machine has no memory, so the honest expectation was a grid with nothing in it — and a 10×10 grid gives 100 chances for noise to look like a pattern, which is exactly why the chi-square is run on the whole table at once instead of on the warmest cell.