Six lessons, each printed with the numbers it stands on. They carry no picks.
Every draw is independent
Independence means the next draw starts from zero. The machine does not know what it produced yesterday, the balls carry no memory, and no combination is owed anything. A Numbers straight bet — one specific three-digit combination — has probability 1 in 1,000 on tonight's draw, on tomorrow's, and on every draw that follows, whether it appeared last night or has not appeared in a decade.
This single fact decides most questions readers bring to a numbers section. It kills "due" numbers, it kills hot streaks as a method, and it caps what any tracking record can ever show. What tracking can do is keep an honest count of the past — which is what this section publishes — and grade any claim, including its own, against the base rate the independence fact sets.
The practical reading: choose numbers you like, for reasons you like. The record gives no combination an advantage over any other, and it never will, because the advantage would have to come from the one thing the draws do not have — a memory.
The 50-cent dollar: the house-edge math
New York's numbers games are fixed-prize games, and their prize tables price every bet at about half its fair value. A Numbers straight pays $500 on a 1-in-1,000 chance: fair would be $1,000. Expected return per $1 is exactly $0.50. The box plays pay slightly less than half — $0.48 per $1 on a 6-way box — and Win 4 repeats the pattern at 1-in-10,000 scale: $5,000 for the straight ($0.50 back per $1), down to $0.479 per $1 on the 4-way box, which pays the officially printed $1,198, not $1,200.
Take 5 arrives at the same place by a different route. Half of each draw's sales form the prize pool, so the cash expected return is $0.50 per $1 by construction, before counting the free play a match-2 wins — and a free play is itself a ticket in a 50-cent game.
No selection method touches these numbers. Bet type moves the expected return between $0.479 and $0.50 per $1 in these games. Number choice moves it not at all. Anyone selling a system for these games is selling a way to choose which half of the dollar to keep — and the answer is that the house keeps the other half regardless.
Hot and cold: we tested it on 46 years of New York draws
Before this section published a single frequency table, the lab pre-registered a test of the strategies those tables invite — and froze the test design before computing a single outcome. Following the most frequent combination over trailing windows of 60, 180 and 730 draws; following the coldest, most-absent combination the same way; and following the hottest front pair — one $1 straight bet per draw, over every New York draw from 2015 through November 2026. That is 8,652 bets in each of the 13 strategy cells, on the official history going back to 1980.
The result: nothing. No cell came near statistical significance (the smallest p-value in the table is 0.22, against a pre-registered bar of 0.01). Pooled across the twelve straight-bet cells, the strategies recorded 51 winning bets against 51.9 expected from chance alone — almost exactly the independence prediction — and returned about $0.51 per $1, the house edge wearing a strategy costume.
The one positive cell, Win 4 hot over a 730-draw window, rests on 2 hits against 0.87 expected. Two tickets are not a method. The frequency tables on this site are published as the descriptive record they are, and this test is why they will never be dressed up as more.
Why "due" cannot be true
The gambler's fallacy says a combination that has not appeared for a long time is "due" — closer to appearing because of its absence. For that to be true, the draw would have to keep score. It does not. Each draw is a fresh physical event: ten balls, three or four tubes, no ledger. A combination absent for 2,000 draws faces the same 1 in 1,000 (or 1 in 10,000) on draw 2,001 as the combination drawn last night.
The fallacy feels true because long absences are rare for any named combination — but with 1,000 combinations, some combination is always deep into a long absence. In the fetched New York history, the longest-absent Numbers combinations have been missing for thousands of draws, and the overdue table on the Frequency page always has entries exactly like them. That is not a pattern building pressure; it is what a fair game looks like when you watch a thousand combinations at once.
Our backtest bought the most overdue combination every draw for over a decade of draws. It won at the base rate, like everything else. "Due" is a story the absence tells. The arithmetic does not back it.
What tracking actually proves: reading a ledger against a base rate
A tracking record proves something only against the rate chance predicts. Flip a fair coin 100 times and record 54 heads: nothing learned. Follow a 1-in-1,000 bet for 8,652 draws and record 10 wins against 8.65 expected: also nothing — that gap is smaller than ordinary fluctuation, and it is roughly what our hot-strategy cells actually produced.
Three habits separate a record from a story. First, the expectation is computed before the results are admired: for a straight bet it is bets divided by 1,000 (or 10,000), full stop. Second, sample size is part of the claim — a surplus of two wins in ten thousand bets is noise wearing a number. Third, the ledger never resets: a method that is graded only from its good month is not being graded at all.
That is the discipline this section's picks ledger is built on. Every pick the section ever publishes is entered before its draw, settled against the official result, and printed beside the base-rate line — what the same money on random combinations would expect. If a future series cannot beat its own base-rate line over a long run, the ledger will say so, in public, with both numbers side by side.
The one honest nuance: unpopular numbers and shared prizes
There is exactly one place where number choice changes anything, and it is not the chance of winning. In games where a prize is shared or paid from a pool — Take 5 in this section — a win on an unpopular combination is split among fewer tickets. Combinations like 1-2-3-4-5, birthdays (all numbers 31 and under), and patterns drawn on the ticket grid are played by many people; when one of them hits, each winner's share shrinks. A combination nobody else chose pays its winner the whole tier amount.
Read the limits carefully. This does not make any combination more likely. It does not apply at all to Numbers and Win 4, whose prizes are fixed — a straight pays $500 or $5,000 no matter how many tickets share the combination. And in Take 5 the effect changes the expected split of a prize, not the probability of holding it; the pool still returns its 50 cents per $1 to the winners as a group.
So the honest version of the advice is narrow: in a shared-prize game, an unpopular combination wins the same amount of luck and keeps more of it. That is a statement about other players, not about the draw — and it is the only one of its kind the record supports.