Is the draw actually random? We test it.

Chi-square uniformity on every digit position, runs tests and serial correlation on every operator's full draw history — recomputed nightly, p-values published. No other Malaysian results site runs this audit.

Something is worth a second look — see the highlighted rows. 8 operators, 80 statistical tests. The flag threshold is p < 6.3e-4 (Bonferroni-adjusted) — strict on purpose, because running this many tests guarantees a few small p-values by pure chance.

What a flag means — and doesn't. A flagged row says the STORED draw history fails a uniformity test. That has two candidate explanations: an artifact in how the data was collected, or the operator's draw process itself — and the honest order of investigation is data first. Flagged operators get a cross-source verification pass before we draw any conclusion; we do not accuse anyone based on a p-value. Notably, the most-regulated operators pass every test through the exact same data pipeline.

Operator	Draws	Digit positions 1–4 (p)				Runs (p)	Serial (p)	Last 365d (min p)
Sandakan 4D⚠️	1,520	3.2e-41	0.181	5.0e-10	1.4e-9	0.004	0.086	7.3e-7
Grand Dragon⚠️	2,239	1.8e-33	9.0e-12	1.4e-20	1.7e-18	0.447	0.411	2.8e-4
Sabah 88 4D	1,384	0.129	0.001	0.295	0.956	0.270	0.575	0.020
Da Ma Cai 1+3D	5,975	0.134	0.437	0.467	0.469	0.100	0.191	0.013
SportsToto 4D	5,673	0.015	0.582	0.367	0.925	0.193	0.296	0.304
Magnum 4D	6,804	0.439	0.790	0.556	0.120	0.035	0.394	0.126
Singapore 4D	1,467	0.850	0.448	0.772	0.777	0.601	0.727	0.140
Special CashSweep	1,525	0.281	0.828	0.445	0.760	0.539	0.595	0.180

Position columns: chi-square uniformity of digits 0–9 at each of the four positions over all top-3 prizes. Runs: Wald–Wolfowitz on the 1st-prize sequence. Serial: lag-1 correlation of consecutive 1st prizes. Sorted worst-p first.

How to read the p-values

A p-value answers: if the draw were perfectly fair, how often would we see a pattern at least this extreme by luck alone? Small p = surprising under fairness. It is NOT the probability the draw is rigged.

We run 80 tests, so some small p-values are guaranteed by chance — about 1 in 20 tests lands under 0.05 even on perfect data. That's why the flag line sits at p < 6.3e-4, not 0.05: it corrects for how many chances we gave luck.

What would failure look like? A digit position with a persistently tiny p across recomputes, or a runs/serial failure holding up in the recent window. One-off borderline values are what randomness normally produces — and 'everything passes' is the expected, publishable result.

Fairness report →Method leaderboard