Where the strategies come from
Every strategy in the solver is decoded from MonkerSolver's own solution files, using Monker's exact compact-row hand indexing. Nothing is rebuilt from equity buckets and nothing is interpolated between boards. Where a solve stores a mixed strategy, frequencies come through at half-percent resolution: if the solver bet a hand 62.5% of the time, the tool says 62.5. Where a solve stores a pure strategy instead, you get that unsmoothed, and the mixing you see across a hand class is the honest blend of its combos rather than a smoothed average.
The PLO library is nine solves: 6-max from 12 to 200 big blinds, plus heads-up at 100, each solved preflop through river. The NLHE library has nine solves: eight 6-max depths from 20 to 200 big blinds, plus heads-up at 100. PLO5 adds two solves, 6-max and heads-up at 100bb. Rake is stated next to the stack selector on every depth, because ranges move with rake and a range quoted without its rake assumption is not a fact.
Exact where possible, simulated where not
Postflop equity is exact enumeration - every runout counted, not sampled. Preflop equity is exact where the combination count allows and Monte Carlo where it does not, and every page that uses a simulated figure states the trial count and the standard error that follows from it.
This is why some tables report a tier rather than a rank. The PLO5 rankings run 200,000 trials per class, which gives roughly 0.11 percentage points of standard error. The top 142 classes sit inside a 0.5-point band. Numbering them 1 to 142 would be presenting noise as precision, so the page reports the band and says why.
The combinatorics we publish
Counting arguments are stated in full rather than asserted, because they are the numbers most often repeated wrongly. PLO4 has 16,432 canonical starting-hand classes over 270,725 combinations - the figure is frequently miscited as 16,384, which is 214 and not the answer to any question about poker. Flop texture frequencies are enumerated across all 22,100 distinct flops rather than sampled.
Dates and drift
Every dataset page carries the date its underlying tables were generated, and that date is what the page's structured data reports as its publication date. When a solve is regenerated the derived pages are regenerated with it. A page that has not been recomputed does not get a fresh date just because it was edited.
What is reproducible, and what is not
Equity against a random hand is reproducible by anyone with a correct hand evaluator and enough compute. That is exactly why those datasets are published as CSVs under CC BY 4.0 - the claim is checkable, so giving it away costs nothing and lets anyone verify us. The solver strategies are a different matter: reproducing them needs a MonkerSolver licence, the same trees and the same rake model, so we describe the inputs precisely instead of pretending the output is trivially checkable.
Stated limits
The solver library covers 4-card PLO, NLHE (6-max, plus heads-up at 100bb) and PLO5 at 100bb, cash only. It does not cover PLO6, does not solve multiway postflop, and has no ICM and no nodelocking. The one solve that runs on demand is a custom river with your own sizes; a full sim of your own is outside the plan. The equity calculators, the reference tables and the hand tracker do cover PLO5, PLO6, Big O, Omaha hi-lo and Short Deck. Where a tool can't answer something exactly, it shows what it assumed: the hand tracker's decision report, for example, sets your first-in play beside a named solve with the sample count, the bucketed stack depth and the unchecked rake and sizes printed next to the numbers, and it doesn't call a difference a mistake.
Corrections
If a number here is wrong, tell us and we will correct it and say what changed. We would rather publish a correction than carry a figure we cannot defend.