How the Viper GTO solver is validated
A numerical solver needs checks with known answers and repeatable inputs. We publish the scope and limits of our validation rather than treating one accuracy number as a guarantee for every spot.
Start free ↗An analytic game with a known value
Our small river test uses board AsKsQsJs2d, pot 100, effective stack 100, OOP range 2h2c and IP range ThTd,4h3h. Only pot-sized bets are allowed, with no raises or rake. The target IP game value is 75 and OOP is 25.
The value 0.75P is not a 75%-pot bet size. The test runs 2,000 iterations and compares best-response values with tolerances around the analytic values. Download the scenario and recorded run below.
Historical comparison with TexasSolver
The August 28, 2026 record aligns 580 decision nodes for Ks7h2d5c, pot 50, stack 2,000, 50%-pot bets and no raises. Our independent best response measured the exported TexasSolver strategy at 0.310% of the pot; our own strategy was measured at 0.004% in that case.
This is a specific turn-tree comparison, not a claim that every Viper solution has 0.31% error. Different mixed equilibria and export rounding can produce frequency differences. The original full comparison inputs are not included in the small analytic example below.
Convergence and reproducibility
The source validation on September 5, 2026 passed 63 tests: 48 unit tests, 13 equilibrium tests and two reproducibility tests. They cover more than a single strategy screenshot, including suit isomorphism and repeated solves.
These checks apply to the tested source. Web and native releases can use different engine versions; comparing platforms requires matching the version, game parameters and accuracy.
Limits of the evidence
The solver studies a configured two-player game with chosen ranges and bet sizes. It does not infer a uniquely correct opponent range or guarantee profit.
The historical cross-check encountered problems with the external macOS build on flop and river-only inputs, so it does not establish a full cross-solver validation of every street. New claims require new experiments.
Clairvoyance experiment · 5 September 2026
- Theoretical IP game value
- 75 / 100 = 0.75P
- Measured exploitability after 2,000 iterations
- 0.0127 chips · 0.013% pot
The CLI run measures exploitability, not the theoretical value directly. Percentages use the initial 100-chip pot. This small river test does not establish accuracy on all game trees.
Download inputs and measured result (JSON) ↓