Methodology

PLO bomb pots: what the audit establishes.

Cucaroochoo covers a specific four-card PLO bomb pot game: everyone antes, then each player enters for a fixed wager equal to the initial ante pot or folds on the flop. There is no further betting.

A strategy file, a numerical audit and the trainer's feedback are distinct evidence. Each must be evaluated on its own terms.

Verification status

2-handed

Certified

Certified by an exact best-response check of every one of the 1,761 flops, on the 210 stored hands per flop. No player can gain more than 0.0000091 antes per hand by deviating; the bar was 0.001. An exact-arithmetic replay confirmed it. In November a second, independently written exact checker recomputed all 1,761 flops against the shipped strategy and matched the certificate on every seat.

3-handed

Certified

Certified. Every one of the 1,761 flops was checked exactly for the strategy the app plays: no seat on any flop can gain more than 0.000903 antes per dealt hand by deviating. The limit was 0.001 ante, under the 210-hand finite-pool sequential-dealing model.

In November a second, independently written exact checker recomputed all 1,761 flops against the strategy the app plays and agreed with the certificate on every seat.

How it was checked

The Exact certificate checks the current hybrid strategy in one uniform run, bound to its played-strategy export. It was independently reviewed and accepted. This is a strategy deviation-gain bound, not a per-hand grading error bound.

The original audit computed each seat's best reply against the other two seats' fixed strategy. The earlier averaged-v1 check evaluated all 47 flops whose probabilities change under suit averaging. Its uniform-flop global bound below 0.00000625 and the following links are retained as historical calculation provenance, not the current hybrid policy certificate. The current conditional bound retains the reviewed exact arithmetic and conservative rounding allowances.

The historical October hybrid per-flop bound combined the certified raw policy on the two repaired boards with the reviewed suit-averaged bounds on the remaining records. The public proof-input bundle binds the original receipts and source profile. Physical-class IID selection weights are split over duplicate records; this does not certify unrestricted physical four-card hands.

The independent arithmetic review states the error, overflow, policy-rounding and model assumptions. The independent exact aggregation records the corrected seat bounds. The earlier population intake is calculation provenance, not that proof review.

This establishes a bound for the stated finite-pool policy. It does not certify unrestricted physical four-card hands, every trainer grade, or website and phone delivery.

4-handed

Certified

Certified by an exact check of every one of the 1,761 flops, on the 210 stored hands per flop. No seat can gain more than 0.0006 antes per hand by deviating; the bar was 0.001. A second engine and an independent machine cross-checked it.

In November a second, independently written exact checker recomputed all 1,761 flops against the shipped strategy and agreed with the certificate on every seat.

On average, a seat could gain about 0.0002 antes per hand. Every flop is under the bar.

5-handed

Certified

Certified by an exact check of every one of the 1,761 flops, on the 210 stored hands per flop. Every seat’s expected gain from deviating is less than 0.001 ante per dealt hand. Two independent computations agreed on every flop, and three flops certified earlier were recomputed from scratch as controls and matched exactly.

On average, a seat could gain about 0.0002 antes per hand. Every flop is under the bar.

A second reviewer independently rechecked every flop's bound from the retained records, all 8,805 seat bounds, and accepted them.

Library strategy independently checked, EVs estimated: the trainer shows five-handed EVs as simulation estimates.

6-handed

Certified

Certified by an exact check of every one of the 1,761 flops, on the 210 stored hands per flop. Every seat’s expected gain from deviating is less than 0.001 ante per dealt hand. Two independent computations agreed on every flop, and a separately written exact engine recomputed 70 flops from scratch, including every flop near the bar, and matched on all 420 seats.

A second reviewer independently rechecked every flop's bound from the retained records, all 10,452 seat bounds, and accepted them.

Library strategy independently checked, EVs estimated: the trainer shows six-handed EVs as simulation estimates.

Seven-handed

Estimated practice tier

Seven-handed: estimated practice (not certified). Paid only. Seven-handed uses a stored practice strategy (1,761 flops, 126 action histories each, 210 stored hands per flop) with practice states from a held-out evaluation. A stored hand with no opponent adjustment is graded from that practice strategy; any other hand, or any opponent adjustment, is graded by a 20,000-runout estimate with a fixed seed. After a one-time explainer, the app shows each seven-handed EV figure as an estimate with its margin of error. A grade is called only when the estimate is clear of zero by more than that margin, at 3.5 to 4 standard errors and at least 0.10 antes from zero; otherwise the trainer says Too close. Match recap and Ledger mark seven-handed totals as estimates.

That margin covers runout sampling only. It does not cover how far the seven-handed strategy itself sits from a true equilibrium: that strategy is not certified, that distance is unmeasured, and no claim of a certificate is made for it.

Library audit status: earlier recorded snapshot

This is an earlier evidence snapshot, kept exactly as recorded. It predates the six-handed certificate, the seven-handed practice tier and the November second checks shown above, so it is historical, not current status.

Loading the dated audit record. If it cannot be verified, no acceptance status will be shown.

The accepted two-player model

The audit covers 1,761 stored flop records, each with an ordered pool of 210 hands and the actual six-bit strategy probabilities supplied to the trainer. The first player draws uniformly from that pool; the button draws uniformly from the remaining compatible hands. All 820 legal turn–river combinations are included for each compatible pair.

A best response chooses the higher-EV action for each hand against the opponent's fixed strategy. The audit measures how much that response improves on the stored strategy, using exact arithmetic. The acceptance threshold is 0.001 ante per player per dealt hand. Each player is assessed separately, including hands on which that player has no decision.

The accepted two-player strategy passes this threshold on every stored flop. This does not certify optimality across all physical four-card hands, zero exploitability, or the correctness of every trainer grade. It does not establish acceptance of any other table size.

Reading the trainer's numbers

Expected value
The average value of an action under the stated model. Enter EV is measured relative to folding: Fold is zero, and the ante already paid is sunk. An entry wager equals the initial ante pot; no wager is collected from a sole entrant. Two-, three- and four-handed spots against the stored strategy are marked Exact and calculated against the stated finite opponent pool and chance model; other spots use a simulation estimate. Five- and six-handed EVs are estimates. Seven-handed EVs are estimates shown with a margin of error. Exact feedback is not unrestricted physical-hand equilibrium certification.
Equity
A hand's expected share at showdown, including ties, against specified opponent hands and runouts. Equity alone does not account for uncontested wins or every possible pot size.
Strategy frequency
How often the stored policy enters with a particular pool hand at that history. It is not the probability that the hand wins.
EV loss
The estimated cost of the chosen action relative to the better available action against the modeled opponents. It is not money lost in a particular hand, a forecast of winnings, or proof that the stored strategy is optimal. When no stored mix applies and the exact Enter and Fold values are less than 0.01 antes apart, the trainer calls it Too close to call: either action is fine. A stored mixed hand is Correct either way within 0.05 antes.
Realized result
The outcome of the cards and actions in a particular hand. A profitable decision can lose; a costly decision can win. Decision quality and short-run results should be assessed separately.

What remains separate

Trainer grading, arbitrary-card analysis, opponent adjustments, saved-hand migration and browser delivery require their own checks. A matching file hash confirms the supplied bytes; it does not repeat the numerical audit or validate the surrounding interface.

The certified two- to six-player strategies are available in the trainer. Their verification scopes are described separately above. Seven-handed: estimated practice (not certified). Paid only.

20 free hands a day. No credit card. Create a free account