Pilot. Prove. Own.

Every task defines its win as a condition on the simulated world, “the lid ends closed” or “the can ends above the pedestal”, never as a screenshot or a claim.

Each task also defines where objects may start and requires the arm at rest, so a run can't begin half-done. The server enforces this before anything else.

Arrow keys translate the gripper, E and D raise and lower it, Q/W/A/S/Z/X rotate the wrist, Space opens and closes the fingers.

While you pilot, the browser logs your exact control input at every physics step. That log, not a video or a score, is the demonstration you submit.

Before any replay, the claimed initial state must sit inside the task's legal envelope. Claim the box began closed, or hand the arm free momentum, and the run is rejected before the physics even runs.

The simulator is deterministic, so an honest run reproduces exactly. The accepted difference is zero, on every coordinate: one bit off and the run is rejected, with the first divergent step named.

A five-minute episode is thousands of physics steps; the server re-simulates them in about a second before deciding.

The success checker runs on the state the server computed, never on your claim. The verdict is PAY or REJECT: every point is earned, not claimed.

The trajectory is hashed and that hash becomes the Data ID: an onchain account created in your name. One hash, one record, forever.

The program stores the task, points and sequence number in the record itself. No IPFS, no external host, nothing to go offline.

The same transaction credits your points, weighted by difficulty. Points will weight a future token distribution (Phase 3 of the roadmap).

Piloting is free and the operator pays the transaction fees. You connect a wallet because the run is recorded to you, not to an account we hold.