Tools for exploring the benchmark
OptoRigBench tests AI agents operating a simulated all-optical neuroscience rig: they image a neuron culture, stimulate cells with light and map how they connect. These two tools run the benchmark's own simulator on this server, so you can drive the culture by hand or watch how a model's episode played out.
Culture bench
Drive a T1 culture dish with the agent's own tools: the camera, LEDs, segmentation, recording and stimulation. Toggle the faults the agents met, edit the wiring, and open the truth drawer the agent never sees. One simulator per visitor; at most a few at once.
Open the benchReplay
Pick any scored episode (the 195 headline episodes, the ablations and the experiments E1–E3 and E5; the E2 and E3 lists also offer their provider-error attempts, labelled and not scored) by model, task, scenario and trial. Its replay movie is rendered on demand and you watch the frames as they are drawn; then jump to its page in the run viewer.
Open the replay tool