An ordinary 2×2 identification task, run as an experiment rather than a widget: set it up here, and the task takes over the screen until you finish or stop. Nothing is simulated except the stimulus — the responses are yours, and the fit is the same network the R and Python packages ship.
Fitting a GRT model by numerical optimisation takes seconds, so it can only happen once a session is over. Here it takes a fraction of a millisecond, so the design can make a decision at every block break.
You have a saved session
Stored in this browser only — the model runs on your machine, so there is nowhere for it to be sent. Running a new task replaces it.
Set it up
The four stimuli
What happens
A welcome screen, then the stimuli and keys, then blocks of trials. Between blocks you see the fit so far — with every earlier estimate still on the plot, dimmed, so the convergence is visible rather than asserted. The task ends when the posterior reaches your precision target or the ceiling is hit, whichever comes first, and you can then take the data to the Analyse page.
Trials on which you miss the deadline are discarded, not scored as errors: a missed deadline is not a confusion, and counting it as one would bias every parameter. Esc stops the task at any point.