Zebrafish
Neural

Recovered from game 16, ply 33. Later AWS-only history is unavailable. Recovery record ↗

ZEBRAFISH NEURAL / SHARED EXPERIMENT

Watch the Fish Play Chess

One game for everyone. Watch the controller compare legal moves, then inspect the inputs and motor margins behind each recorded choice.

CONNECTIONConnectingWaiting for the shared game.
Shared game—
Controller plays—
Phase—
Completed games—
Controller W / D / L— / — / —

Shared board

Opponent awaiting confirmation

Waiting for a position from the server.

The game runs independently of viewers.No move received

Current comparison

WAITING
Left input channel—
Right input channel—
No active comparisonMotor L − R: —

Left and right identify candidate input channels, not directions on the chessboard.

Feature input

32 × 16

Handcrafted move features, not a board screenshot.

Motor output

LeftRight

No comparison samples received.

The fish plays —

Waiting for the controller’s first move.

Loading the zebrafish body.

DRAG TO ROTATE

Body motion illustrates received Spinal drive and Motor L/R output during move comparison.

Neural controller

Eight population states · model rates

LeftRightILLUSTRATIVE GEOMETRY
DRAG TO ROTATE
WAITING FOR A SAMPLEModel time —

Original controller · frozen gains

The model is not chess-trained. This adapter turns defined move features into synthetic input, compares motor output, and selects from legal moves.

Move record

0 PLIES

The board above stays on the shared game.

Moves will appear as the server commits them.

PGN ↗Game JSON ↗

Move inspector

SELECT A MOVE

Select a controller move to inspect its recorded candidate features and comparison margins. Inspecting a move does not change the shared game.

What this experiment shows

METHOD & SCOPE

The server owns the game and runs both players. The controller compares legal candidates using predefined features. The opponent is identified above; game outcomes count only completed records.

The brain and retina show actual samples from the selection process. Between comparisons, the last received sample remains labeled. A connection interruption freezes these values instead of advancing a local substitute.

Move records expose inputs, comparison margins, seed and parameters. This is a software experiment with a frozen controller, not a chess-trained biological brain. The Wikipedia and learning experiments remain separate.