Are these numbers right?

Every test in this project checks its code against its own code. That catches a thing that broke and cannot catch a thing that was wrong from the start. These compare against work by other people: eval7, a separate hand evaluator written in C, and the published solutions the preflop charts claim to be transcribed from.

6 passed, 0 failed, 1 not run · generated 2026-08-27 19:59 UTC at 46ad114.

vs eval7 Hand ranking pass

does evaluate() order two hands the way eval7 does?

20,000 random pairs on a full board, every one ordered identically.

vs eval7's own sampler The reference itself pass

is the enumeration every row above is checked against right?

Worst gap 0.146 points against a 0.335-point sampling band at 200,000 draws.

spot	ours	theirs	diff
As Ks vs Qh Jd on 7c 2d 9s	77.475	77.34	0.135
Ac Ad vs Kc Kd on 2h 7s Ts	91.616	91.615	0.001
Jh Th vs Ad Kc on 9h 8c 2h	69.697	69.843	-0.146
7c 7d vs As Kh on Kd 4s 2c 9h	4.545	4.613	-0.068
Qs Jc vs Ah Ts on Kd Qd 3h 8s	86.364	86.272	0.092
5h 5c vs Ac Qd on Kh 9s 4d 2c 7h	100.0	100.0	0.0

vs eval7's evaluator, enumerated here Hand versus hand pass

is exact enumeration exact?

Every spot agrees to the last place a float has.

spot	ours	theirs
As Ks vs Qh Jd on 7c 2d 9s	77.4747	77.4747
Ac Ad vs Kc Kd on 2h 7s Ts	91.6162	91.6162
Jh Th vs Ad Kc on 9h 8c 2h	69.697	69.697
7c 7d vs As Kh on Kd 4s 2c 9h	4.5455	4.5455
Qs Jc vs Ah Ts on Kd Qd 3h 8s	86.3636	86.3636
5h 5c vs Ac Qd on Kh 9s 4d 2c 7h	100.0	100.0

vs eval7's evaluator, enumerated here Hand versus range pass

does the sampler agree with an exact solve, inside its own error bar?

Worst gap 1.45 standard errors, over 8,000 samples a spot.

spot	ours	theirs	diff	sigma
As Ks vs QQ+,AKs,AQs on 7c 2d 9s	35.631	35.536	0.095	0.17
Jh Th vs TT+,AJs+,KQs on 9h 8c 2h	56.212	55.985	0.227	0.41
7c 7d vs 22+,A2s+,KTs+ on Kd 4s 2c 9h	45.131	45.942	-0.81	1.45
Qs Jc vs 88+,ATo+,KJo+ on Kd Qd 3h 8s	46.325	46.333	-0.008	0.01

vs eval7's evaluator, enumerated here Equity per holding pass

is the per-holding pass right, one holding at a time and pooled?

Exact on every street that enumerates, both per combination and pooled.

spot	combos	ours	theirs	diff
As Ks vs QQ+,AKs,AQs on 7c 2d 9s	18	35.5359	35.5359	0.0
Jh Th vs TT+,AJs+,KQs on 9h 8c 2h	39	55.9855	55.9855	0.0
7c 7d vs 22+,A2s+,KTs+ on Kd 4s 2c 9h	112	45.9416	45.9416	-0.0
Qs Jc vs 88+,ATo+,KJo+ on Kd Qd 3h 8s	75	46.3333	46.3333	-0.0

vs published 6-max 100bb solutions Opening ranges pass

does a chart labelled `solver` open as often as the solve it names?

Worst gap 2.4 points of opening frequency, against a 3-point band.

spot	ours	theirs	diff	equal blind	widened by
UTG	17.9	15.5	2.4	19.2	1.2
HJ	21.6	19.5	2.1	23.1	1.5
CO	26.1	26.5	-0.4	30.0	3.9
BTN	43.6	44.0	-0.4	50.5	6.9

vs TexasSolver Postflop strategy not run

do the postflop numbers match a real solve?

No solver binary. Build TexasSolver and point TEXASSOLVER_BIN at its console_solver, or put that on the path. Until then nothing here is checked against a postflop solve, and `rollout.py`'s numbers are `model` - exact against these five bots, and not a claim about equilibrium.

What is still not checked

The postflop numbers. Nothing on this page tests them against a solver, because a postflop comparison needs a binary this repository does not ship. That is why every postflop number in a review is labelled model and not solver: it is exact against these five bots, whose strategies are known functions, and it is not a claim about equilibrium. The bet-sizing curve stops at showdown on the street you are on, so it understates a small bet that sets up a bigger one on the next card.

Back to the table