We hold hand histories where every seat's cards are visible, a solver farm built for these rooms, and the sessions to check both against. The blog is where the numbers go.
We priced 36,000 pool preflop decisions against solver charts and luck-adjusted 17,896 of our own hands. The pool leaks less than the rake preflop.
8,232 NL200 ring hands with every seat’s cards visible (35,470 pool seat-hands), plus 17,896 of our own Ignition hands, June–October 2026.
Ignition hides names but exports every hole card. From stack sizes alone we re-identified seats within a table at 98.6% precision. Zone is immune.
A range is a guess at P(hand | action). We score every range we ship against the hands the pool actually held, with a baseline and confidence intervals.
Open-limps are rarer than the forum myth: 5% outside the blinds at NL200, 7% at NL100. But one early-position limp in three is a premium. 13,000 hands.
The Ignition pool folds its big blind 51% to a button open and 61% to an early open. Measured defence ranges from 16,506 visible-card hands.
Same stake, mostly the same players, different behaviour. Zone leaks more preflop, stabs the flop more and bigger, and cannot be read at all.
A plain-English look inside an AI poker bot: how it thinks in ranges, how a personality changes its play, and why its randomness makes it hard to read.
A simple routine for practising No-Limit Hold'em against AI bots: pick one opponent style, play a focused block of hands, then review the ones you lost.
