We checked the dream book against 41 years of results.

This is the check that had to come first, and it changed the whole test. We counted all 32,000 entries across the 5 dictionaries and charts and looked at which numbers they cover.

The shape-gap column is the biggest difference, in percentage points, between the dictionary's mix of digit shapes (8888, 8887, 1212, 1123, 1234) and the mix across all 10,000 numbers. A dictionary that loved repeated digits would show up here, and comparing its hit rate against a flat baseline would then be misleading. A complete table reads exactly 0.00 by construction — we measured it rather than assuming it.

Every prize slot published by every Malaysian operator since 1985 — 1st, 2nd, 3rd, special and consolation — comes to 612,168. Spread over 10,000 numbers that is 61.22 wins per number. We do not use that flat figure as the yardstick, though: we use each number's own digit shape.

Why bother: if a dream mapped mostly to repeated-digit numbers, and repeated-digit numbers happened to come up more often, a flat baseline would credit the dream with an edge that belongs to the digits. So each dream's expectation is the sum of its own numbers' shape rates. On this archive the correction turns out to be tiny — the furthest shape from flat is Quad, 3.57% away from 61.22 wins per number — but that is a measurement, not something we assumed before looking.

This is the actual test. Each of the 67 dream symbols on this site pulls its own set of numbers out of the dictionaries — 4,418 distinct numbers between them — and every set is scored against 612,168 prize slots using the shape-matched yardstick above.

Read the p column carefully. 2 of the 67 dreams came in under a naive p < 0.05, and pure chance predicts about 3.4. That is the whole story of this table: the "luckiest" dream in it is the one luck was always going to hand us.

The last column is how big an edge a dream's numbers would have needed for this test to flag it, given how many numbers it has. The median across the table is 5.5%, so a genuine 10% advantage on a typical dream would not have gone unnoticed. Dreams with only a handful of numbers can rule out much less — read those nulls loosely.

Each dream gets a two-sided binomial test with a continuity correction, against the sum of its own numbers' shape-matched rates. Dreams overlap (a coffin is also a funeral), so the goodness-of-fit above uses a strict partition instead: each dream's exclusive numbers, one cell for numbers two or more dreams claim, one for numbers no dream on this site claims. Prize slots on one board are drawn without replacement, which makes the true variance very slightly smaller than assumed — the conservative direction.

The 大伯公千字图 · 观音千字图 charts name three-digit numbers, and a three-digit number is played as the last three digits of a 4D result — which is how the dream lookup on this site already reads them. So they are scored against the 1,000-number tail space, where every published prize slot still counts and each tail is worth 612.2 wins. We never pad a chart's number out to four digits: 372 does not become 0372, because the chart never said that.

The 3D charts get no per-dream tests. With 1,000 entries a dream maps to a handful of numbers, and adding a hundred-odd low-power tests would have tripled the batch — making the threshold stricter for every real test — while being able to detect nothing. That is a deliberate choice, stated rather than quietly applied.