PulseGTO
Blueprints & databases
Hand abstraction
A blueprint uses a configurable abstraction level for its postflop strategy. This keeps whole-game training practical on a desktop, but its postflop result has less detail than a per-combo re-solve of one board.
Strength
The Strength setting (small, medium or high) selects the abstraction level. A higher level generally uses more RAM and can represent postflop play in more detail. See Abstraction strength and open ranges for one measured comparison.
An abstraction-based blueprint can give different exact hands the same postflop strategy. Its accuracy number measures convergence at the selected abstraction level, not the difference from a full-detail solve of every hand and board.
The "!" badge: classes that share a strategy
The Strategy tab marks nodes where the abstraction gives several preflop hand classes the same postflop strategy with a "!" badge. You can add more detail in two ways:
- Repair ! classes (Preflop Settings): continue training with more detail at the marked nodes and save the result in the same
.pgb. - Recalculate Node: a re-solve of that node without hand abstraction, with every combo separate.
Compatibility note
Some older checkpoints cannot be resumed after an abstraction-version change. Older blueprints may still open for preflop browsing. Check the file's compatibility message in the app before relying on it.
When to use a blueprint and when to re-solve
A blueprint is useful for navigating a whole game and finding the ranges that reach a line. A selected postflop re-solve is useful when you want individual combination decisions on one board. These are complementary jobs: the local re-solve receives a starting range and does not reconstruct a different preflop equilibrium by itself.
More training reduces residual learning error within the chosen representation. It cannot turn a shared-hand strategy into separate decisions for every combination. A more detailed abstraction can change both memory cost and the represented game, so compare at a stated budget and do not assume that a larger setting is automatically more converged.
Save the chosen node’s per-combo result and keep track of the Re-solved badge. Other nodes still show the blueprint until separately refined. Accuracy can be low inside an abstract model while a no-abstraction solve prefers different actions in the same public spot.