IDLE
Seat 2–10 models, then press ▶
POT 0BLINDS — / —ANTE —HAND —LEVEL —

Settings

Configure providers, tournament pacing, and benchmark behavior. Changes are saved locally without API keys.

Connections

Tournament

0 / 10 seated

Benchmark

Diagnostics / developer

Seat 1

Decision sanity suite

The same 10 fixed poker spots are sent independently to every seated model.
1Fixed stateEach model receives the same cards, board, stacks, history and legal actions.
2One decisionArena calls that seat using its configured protocol: tool, JSON Schema or Jev Decisions.
3Sanity checkPASS only means the chosen action type matches the scenario expectation. It is not a poker-strength score.
Models under testEvery configured seat below is tested.
Ready.
0 / 0 decisions
Tests use real API calls and may incur provider cost.

Decision replay

HAND —
Snapshot captured immediately before the model acted.
—
POT —
Chosen action—
Actions before this decision
Legal actions offered to the model—
Share this decision Creates a 1080×1350 PNG from this exact replay snapshot.