How our solvers and trainers work.

Ranges, bet sizes, betting caps and card abstractions all affect the result. Even with matching inputs, nearly equal-EV actions can have different mixes. We compare values and deviation tests as well as percentages. Similar self-play EV alone does not establish equilibrium: an opponent may still be able to exploit the strategy.

The selected node has less than 1% reach in the current model. Reach is a modeled frequency, not a count of users or training samples and not a confidence score. These lines remain available to explore, with trainer scoring paused. A zero-reach line uses exploration assumptions. A larger local betting gap receives a separate Strategy still settling notice, even on a common line.

No. Each current tool solves an abstracted version of its game. Cards are grouped into strategic classes or strength buckets, and future draws use deterministic samples or continuation estimates. Convergence is not guaranteed by running the tool: check the best-response gap within the abstraction. This does not establish full-card game accuracy.

It depends on the game and selected model. In 2-7 Single Draw, CFR+ trains pre-draw betting using sampled one-draw terminal equity, then trains post-draw betting separately; draw declarations remain modeled. The classic fixed-limit baselines solve path-conditioned betting rounds with sampled or heuristic draw continuations. The earlier 40 BB 2-7 study combines learned betting with a low-core draw policy. The separate cached full-hand 2-7 study evaluates the same frozen betting and draw policy through the remaining hand, with heuristic fallback tracked separately. It is not a newly converged solve at every opened node, and neither 2-7 study qualifies as full-game GTO.

Frequency is the current model’s action mix, not a legality rule. Frequencies below 0.01% display as 0%. The action path hides betting options below that threshold; the engine and detailed analysis retain the full action set. Draw counts remain available to explore. A truly zero-probability action has no uniquely defined conditional range. For a forced draw, the tool retains the preceding range as an exploration assumption and shows zero path reach.

EV starts at the current decision. Chips already invested are sunk costs, so folding has zero additional cost. Every action uses the same reference: an old hand-relative Fold EV of -0.50 BB and Limp EV of -0.41 BB become 0.00 BB and +0.09 BB. Action rankings and EV gaps stay the same. If continuing instead has negative EV from this decision, folding is better. Changing this reference does not validate the opening model’s future-betting assumptions.

No. The A-5 and legacy round-local 2-7 baselines value a called opening round using next-draw equity, omitting later betting costs and jointly optimized snowing. This can make a cheap completion look too valuable even for weak hands. That limitation must not be generalized to every current 2-7 model: the separate cached full-hand study includes future betting under its frozen policy, but still has sampling, training and fallback limitations. None of these results validates playing every starting hand. Starting-hand priors are not forced playing frequencies, and a different table format or raise-or-fold tree is not a like-for-like comparison.

The private 2-7 Triple Draw explorer offers cached full-hand studies at 10, 20 and 40 BB. Each uses fixed BTN Top 50% / BB Top 70% starting-hand priors, not solved opening or forced continuation frequencies. Changing the stack loads that stack’s saved policy; it does not retrain the hand or force different percentages. The earlier 40 BB hybrid study is a separate selectable model. The A-5 and Badugi betas offer 10, 20 and 40 BB round-local studies with game-specific rankings and draw policies. Their opening strategies can match across stacks because the opening model uses next-draw equity; later betting limits and all-in boundaries use the selected stack. These studies are not certified full-game equilibria.

No. All current trainers are ungraded practice: you can review model strategies and EV estimates, but no points or verified EV-loss grades are awarded. Cached 2-7 policy practice uses the selected solver’s frozen betting and draw policy. Range-level solver EVs are not values for the player’s exact private hand. Historical scores remain in saved history and do not qualify the current model. A model preference or a narrow sampling interval does not establish GTO accuracy.

Deeper stacks allow larger re-raises and different continuations; opening ranges do not have to widen as stacks increase. The no-limit presets use the same starting-hand abstraction and draw-equity inputs, but train separate betting trees for each stack. The current 40 BB model folds more than its 10 and 20 BB counterparts. This is an outcome of that abstract pre-draw game, whose called pots terminate on sampled draw equity. Post-draw betting is solved separately, so these exact opening frequencies are not a validated full-game benchmark.

Yes when comparing the same abstract betting tree, ranges, bucket count, and sizing configuration. The value measures improvement available to exact best responses inside that supported betting tree; it is not a whole-game guarantee for the modeled pre-draw and draw layers.

No. High Precision is planned to use 40 or more hand buckets together with additional CFR+ iterations. That can reduce abstraction error and expose finer strategic differences, but it cannot guarantee a more accurate full-game solution. A larger model also needs enough training; otherwise rare nodes may converge more slowly or look noisier.

Within the represented Triple Draw ranges, betting uses a joint distribution of compatible hands and discard histories. A hand cannot share a live or discarded card with its opponent. Action EV and equity condition the opponent’s range on the player’s hand bucket; range tables and draw frequencies use the same compatible histories. The final-street matrix scores the represented exact hands. Ranges are still sampled and grouped, so this does not make blocker effects exact for every possible hand or establish full-game accuracy.

The shared principles will: legal game trees, range filtering, hand abstraction, deterministic results, and clear limitations. Each new mixed game will still need its own evaluator, betting structure, draw policy, and validation before release.