記録を正直に保つ postflop solver。

FlopForge treats the browser as a serious compute platform rather than a thin form around a remote solver. React submits coarse-grained commands, a dedicated worker manages progress and Rust/WASM owns the computational hot path.

Cards, ranges, equity, tree traversal, regret updates and EV calculations stay outside the UI loop. That keeps interaction responsive while making the engine boundary explicit and testable.

The live-hand workspace replays a heads-up hand from blinds through river, including integer wager-to amounts, short all-ins, action reopening, uncalled refunds, street gates and forced all-in runouts.

Each accepted event contributes to a Rust-owned chip ledger and a hash checkpoint. The solver worker must reconstruct and match that checkpoint before a result can be shown, so the UI cannot quietly invent a different action state.

A solver output is not automatically advice. FlopForge separates exact replay, known-hole-card review, weighted-range assumptions and automatic bounded modelling, then checks whether a result meets the configured quality threshold.

When evidence is insufficient or a model falls outside its declared scope, the product returns a baseline or labels the output as analysis evidence instead of manufacturing confidence.

Browser-aware solve budgeting, local cache reuse, compatibility-gated warm starts and current-spot sensitivity analysis are designed around questions that can be checked: will this tree fit, is this cache entry structurally compatible, and where does the preferred action first change in the sampled pots?

The project deliberately avoids turning those tools into claims of unrestricted game solving or mind-reading opponent psychology. The value is a reviewable analysis workflow with explicit assumptions.