Describe the situation in one honest paragraph. The model reads; the solver decides; the board, the collapse, the mix and the reading all generate from here.
The question
Bargaining quick path — deterministic, no brain
For a two-party price negotiation you only need two true numbers. The solver builds the board itself; nothing is generated.
Buying at your ceiling and selling at their floor are both fine deals — the fight is only over the surplus between them.
And remember L2: the strongest move is rarely inside this box at all. It is finding Candice.
印 = equilibrium · ★ = Pareto-efficient · struck names died by dominance · tap a square for why
The reading — prose with provenance
The middle line is never generated.
It's computed.
Anthropic API key
Stored on this device only (browser storage), sent only to api.anthropic.com. Never put a key on a shared or public computer. No key? The bargaining quick path on the Ask page works fully offline.
The extraction seam
The UI and solver never know which brain read the board. This web build ships the Anthropic brain (Sonnet); the JSON contract is the same as the Mac app's — refusal + missing question when the scenario gives positions instead of values.
The middle line is never generated.
It's computed.