# Phone benchmark 1

This is a separate diagnostic, not a replacement for River Poker Lab. Open `phone-benchmark/` on an HTTPS host (or localhost). A raw GitHub file view does not run the page. The standalone `river-phone-benchmark.zip` can be hosted with its files at the site root. No hosting has been performed by creating this package.

Tap **Run comparison**. Start with river/turn, then enable flop and three repeats. Keep Safari foregrounded. Use **Save results** or **Copy results**, and share the JSON in the chat. Reports persist in local browser storage where available. No results are uploaded automatically. Offline status covers the test runtime; source downloads are not precached. Stop terminates the active worker and discards its unfinished solve.

The test uses the same artifact fingerprints as `benchmarks/ENGINE_SELECTION.md`; `manifest.json` lists hashes. Binaries are public deployed builds, not locally rebuilt versions. Matching these deployed builds exactly to source commits has not been established. The supplied source snapshots are the inspected repository revisions, not a claim of independently reproducible identical binary builds.

## Sources and notices

- [WASM Postflop](https://github.com/b-inary/wasm-postflop), inspected revision `97360db7644329b1c23a7adf06e9aa59406e4d4b`. [Engine](https://github.com/b-inary/postflop-solver), inspected revision `9d1509fe5077d019825f833eed04b16d342dfda1`. Copyright 2022 Wataru Inariba; GNU AGPL version 3 or later. Original minified single-thread binding and WASM binary are unchanged. [License](source/AGPL-3.0.txt), [source snapshot](source/binary-source.zip).
- [ucsandman/postflop](https://github.com/ucsandman/postflop), inspected revision `5fc7ee3d92b823b6c58e4f58cbee7d50d5e9e6de`. Copyright 2026 Wes Sander; MIT license. Original browser bindings and WASM binary are unchanged. [License](source/MIT.txt), [source snapshot](source/mit-source.zip).
- River Lab's standalone test adapter, controls and packaging were added on 13 November 2026. This test harness is provided under AGPL-3.0-or-later; the MIT component retains its original license. No warranty.

Accuracy is the sum of unilateral improvements (NashConv), not the upstream engines' differently normalized labels. The known newer-engine raked diagnostic failure is retained for comparison; a negative value is flagged as failure. Memory figures are not total Safari RAM. None of these small tests establishes capacity for unrestricted full-hand poker.