Every frequency, range and equity on this site is either decoded from a solved game tree in our solver database or computed by a method described on this page. Nothing is estimated, and no number is written by a language model. Each data page prints the tree it came from and the date it was decoded or computed; this page collects the assumptions behind them in one place.

Preflop charts

Push/fold ranges

The 5bb and 8bb charts are decoded from the solver database: tournament, chip EV, eight-handed. The 10bb to 20bb chip-EV charts and the ICM charts (final table with eight paid, and bubble with seven of eight paid) are Nash jam-or-fold equilibria computed by BeyondGTO:

Eight seats (UTG, UTG+1, LJ, HJ, CO, Button, SB, BB) with equal stacks, blinds 0.5/1 and a big-blind ante of 1.
The pusher is first in. Every seat behind may call; the first caller closes the action, so there are no overcalls.
Hand-vs-hand equities come from a Monte Carlo table of 200,000 trials per matchup across all 169 by 169 hand classes, with card removal applied to the opponent's combinations.
Chip EV maximises final chips. ICM prices every outcome with the Malmuth-Harville model against the payouts shown on each chart.
Ranges are solved by fictitious play: alternating best responses for the pusher and every caller until neither side can gain. Marginal hands come out as mixed frequencies rather than a hard line.

Each chart says which of the two it is. Decoded charts carry the node they were read from; computed charts say "Nash equilibrium" in their copy.

Opening ranges and facing a raise

First-in opening ranges are decoded for tournaments by stack depth, for 6-max cash at 100bb with an NL500 rake structure, and for three-handed Spin & Go. The facing-a-raise pages use the same 6-max cash tree: the hero's first decision against a single 2.5bb open (3bb when the small blind opens), with every other seat folded. The starting-hand pages are the same data inverted, so a hand's page shows what every chart says about that hand.

Postflop c-bet pages

All c-bet pages come from one 6-max cash tree at 100bb. In single-raised pots the opener raises to 2.5bb, the big blind calls and checks the flop, and the page shows the opener's decision. In 3-bet pots the 3-bettor is out of position and acts first. The tree offers two bet sizes, printed on each page, so the grids show three outcomes per hand: the small bet, the larger bet and a check. Per-hand figures are the solver's mix for that hand class, normalised so each hand's row sums to 100%. Hand classes ignore suits, which is why flush-draw boards are described at the class level and never combo by combo.

Equity tables

Hand-vs-hand equities are Monte Carlo with 200,000 trials per matchup, computed once for all 169 by 169 hand classes and reused everywhere a matchup appears. Multiway equities are 250,000 trials per hand and player count, from heads-up to six-way against random hands. The equity calculator computes in your browser on the exact cards and ranges you enter rather than looking anything up.

Data studies on the blog

Each study ends with a method note stating the tree, the seats, the stack depth, the bet sizes available and the date the data was decoded. Range-level percentages are weighted by combinations and by how often each hand reaches the decision; per-hand percentages are the solver's mix for that hand. Percentages are rounded, so rows may not total exactly 100. Where a study builds on data decoded on different dates, the note gives both.

Competitor figures

Prices and plan details for other products appear only on the comparison pages. Every figure there is stored with a source URL and the date it was checked, and the pages print that date. Nothing about a competitor is stated from memory.

Dates and updates

A page's "updated" date changes only when its data is regenerated or its copy materially changes. Data is regenerated when the solver database gains a format or stack depth, when a computation method changes, or when an error is found. Rounding differences across pages, and solver outputs that are not monotonic from one hand to the next, are left as the solver produced them rather than smoothed.

What is not done

No frequency, range or equity is invented, estimated by hand, or generated by a language model.
No chart is adjusted to look tidier than the solve. If a row looks odd, it is reported as decoded.
No date is bumped without a change behind it.
No competitor is named outside the comparison pages, and never without a dated source.

Corrections

If a number on this site looks wrong, write to [email protected] with the page URL. Errors are corrected on the page and noted in its updated date.

Methodology