ML model
a model retrained nightly on 90 nights of real boards — published so it can fail in public (so far it learns that nothing predicts)
If you had backed every pick
RM1 iBox Big on each pick, on each Malaysian board that night, priced with the real dividend table. An iBox bet covers every ordering of the number, so exact and permutation hits both pay.
A fair draw returns 65.9% on these same numbers, so this stream is running at 0.50× its own chance line. (Its own, not a shared one: a number with a repeated digit has fewer orderings and a lower expected return, so each stream is measured against the numbers it actually picked.)
Every locked pick
Newest first. Each number was written to the database before its draw and never changed.
Picks are locked before the draw (first write wins) and scored at read time against the real boards, so a corrected result re-scores this whole page. Box hits pay on the scoreboard but never move any clock — the number drawn was a different number.