The task
Continuous orientation estimation: see a faint striped patch for half a second, then set a line to its tilt — or, when the cue says so, to the tilt at right angles to it.
Orientation is used first because it gives a precise continuous error on a well-understood circular scale, it is easy to render in a browser, and serial dependence was first reported with it (Fischer and Whitney 2014). Every other perceptual dimension waits until this one works end to end.
Fig. 1 One trial
SCHEMATIC — NO DATA
- Fixation300 ms
- Gabor500 ms
- Blank delay200 or 1500 ms
- Cue: SAME / ORTHOGONALuntil response
- Orientation dialself-paced
- Inter-trial interval700 or 1400 ms
Why ask for the orthogonal orientation
If every trial asks for the stimulus orientation, the previous stimulus and the previous response are nearly the same number, and no analysis can say which one pulls the current report. Asking for the orientation at right angles on a random half of trials — the manipulation of Cicchini, Mikellidou and Burr (2017) — shifts the previous response 90° away from the previous stimulus on about half of consecutive pairs. A perceptual account predicts a bias tied to the previous stimulus; a motor or decisional account predicts one tied to the previous response. The design page shows how much this buys.
Controls built into the task
The response line starts at a random angle on every trial, so “leave the dial where it was” cannot masquerade as response history. The stimulus sequence is constructed so that consecutive orientation differences are spread evenly over the full range, which removes sequence autocorrelation by construction. The true orientation is shown only after practice trials. Nothing on the task screen animates.
Timing, measured rather than assumed
Browsers do not offer laboratory timing. The task schedules the stimulus on animation frames, records the frame timestamps at onset and offset, counts frames, and measures the display’s refresh rate before starting. Trials whose stimulus was shown more than 50 ms off the requested 500 ms, and trials during which the window lost focus, are excluded by the frozen policy. Stimuli are not gamma-calibrated, so nominal contrast is exact in screen code values but not in luminance.
Try a 23-trial demo → Demo trials stay in your browser and never enter a dataset.
The full 612-trial protocol also runs, as infrastructure: study status is SIMULATION_ONLY, nothing is transmitted, and the page says so before it starts.