Simulation only. No human participants have been recruited or run. Every figure on this site is computed from synthetic observers with known ground truth, and says so.

What the last trial does to this one

When people judge the tilt of a line, their answer drifts toward the tilt they saw a moment ago. TimeMind is a browser experiment and an analysis pipeline for asking whether that drift is perceptual — or whether it is response habit, regression to the mean, or an artefact of how the trials were ordered.

Nothing here is a finding about people. No one has taken part. What the project does have is a complete path from a seeded trial sequence to a stored, versioned result, and a set of synthetic observers whose true history dependence is known. Every number below is the pipeline’s answer about one of those observers, which is the only honest way to test an estimator before it meets a human being.

Fig. 1 Serial dependence in a synthetic reference cohort

SIMULATED DATA

−202−90−60−300306090previous − current stimulus orientation (°)response error (°)
Mean response error as a function of how far the previous stimulus was from the current one, 30 synthetic observers, 17,450 trials. Points are observer-averaged bins (± SEM across observers); the line is the group derivative-of-Gaussian fit. The adjusted model’s peak amplitude is +2.56° (95% CI +2.09 to +2.99), where the generating population mean was 2.5°.Source: results/serial_dependence.parquet · run serial_dependence-280346c-1
Table view (193 rows)
stratumΔ orientation (°)mean error (°)SEM (°)observers
all−82.5+0.2840.26430
all−67.5−1.0990.32030
all−52.5−1.4170.22730
all−37.5−2.5840.34130
all−22.5−3.0150.44130
all−7.5−1.8490.25730
all+7.5+0.9680.31930
all+22.5+2.9470.35230
all+37.5+2.9190.36330
all+52.5+1.0230.28030
all+67.5+0.0390.21630
all+82.5−0.1190.21030
fitted_group_dog−90.0−0.006—30
fitted_group_dog−89.0−0.007—30
fitted_group_dog−88.0−0.008—30
fitted_group_dog−87.0−0.009—30
fitted_group_dog−86.0−0.011—30
fitted_group_dog−85.0−0.012—30
fitted_group_dog−84.0−0.015—30
fitted_group_dog−83.0−0.017—30
fitted_group_dog−82.0−0.020—30
fitted_group_dog−81.0−0.023—30
fitted_group_dog−80.0−0.027—30
fitted_group_dog−79.0−0.031—30
fitted_group_dog−78.0−0.035—30
fitted_group_dog−77.0−0.041—30
fitted_group_dog−76.0−0.047—30
fitted_group_dog−75.0−0.054—30
fitted_group_dog−74.0−0.061—30
fitted_group_dog−73.0−0.070—30
fitted_group_dog−72.0−0.080—30
fitted_group_dog−71.0−0.091—30
fitted_group_dog−70.0−0.103—30
fitted_group_dog−69.0−0.116—30
fitted_group_dog−68.0−0.131—30
fitted_group_dog−67.0−0.148—30
fitted_group_dog−66.0−0.166—30
fitted_group_dog−65.0−0.186—30
fitted_group_dog−64.0−0.208—30
fitted_group_dog−63.0−0.232—30
fitted_group_dog−62.0−0.259—30
fitted_group_dog−61.0−0.288—30
fitted_group_dog−60.0−0.319—30
fitted_group_dog−59.0−0.353—30
fitted_group_dog−58.0−0.389—30
fitted_group_dog−57.0−0.429—30
fitted_group_dog−56.0−0.471—30
fitted_group_dog−55.0−0.516—30
fitted_group_dog−54.0−0.565—30
fitted_group_dog−53.0−0.616—30
fitted_group_dog−52.0−0.671—30
fitted_group_dog−51.0−0.729—30
fitted_group_dog−50.0−0.789—30
fitted_group_dog−49.0−0.853—30
fitted_group_dog−48.0−0.920—30
fitted_group_dog−47.0−0.990—30
fitted_group_dog−46.0−1.062—30
fitted_group_dog−45.0−1.137—30
fitted_group_dog−44.0−1.214—30
fitted_group_dog−43.0−1.293—30
fitted_group_dog−42.0−1.374—30
fitted_group_dog−41.0−1.456—30
fitted_group_dog−40.0−1.539—30
fitted_group_dog−39.0−1.623—30
fitted_group_dog−38.0−1.706—30
fitted_group_dog−37.0−1.790—30
fitted_group_dog−36.0−1.872—30
fitted_group_dog−35.0−1.952—30
fitted_group_dog−34.0−2.030—30
fitted_group_dog−33.0−2.106—30
fitted_group_dog−32.0−2.178—30
fitted_group_dog−31.0−2.245—30
fitted_group_dog−30.0−2.308—30
fitted_group_dog−29.0−2.366—30
fitted_group_dog−28.0−2.417—30
fitted_group_dog−27.0−2.461—30
fitted_group_dog−26.0−2.497—30
fitted_group_dog−25.0−2.525—30
fitted_group_dog−24.0−2.545—30
fitted_group_dog−23.0−2.555—30
fitted_group_dog−22.0−2.555—30
fitted_group_dog−21.0−2.545—30
fitted_group_dog−20.0−2.524—30
fitted_group_dog−19.0−2.493—30
fitted_group_dog−18.0−2.449—30
fitted_group_dog−17.0−2.395—30
fitted_group_dog−16.0−2.329—30
fitted_group_dog−15.0−2.251—30
fitted_group_dog−14.0−2.162—30
fitted_group_dog−13.0−2.062—30
fitted_group_dog−12.0−1.951—30
fitted_group_dog−11.0−1.830—30
fitted_group_dog−10.0−1.699—30
fitted_group_dog−9.0−1.558—30
fitted_group_dog−8.0−1.408—30
fitted_group_dog−7.0−1.251—30
fitted_group_dog−6.0−1.086—30
fitted_group_dog−5.0−0.915—30
fitted_group_dog−4.0−0.738—30
fitted_group_dog−3.0−0.558—30
fitted_group_dog−2.0−0.374—30
fitted_group_dog−1.0−0.187—30
fitted_group_dog+0.0+0.000—30
fitted_group_dog+1.0+0.187—30
fitted_group_dog+2.0+0.374—30
fitted_group_dog+3.0+0.558—30
fitted_group_dog+4.0+0.738—30
fitted_group_dog+5.0+0.915—30
fitted_group_dog+6.0+1.086—30
fitted_group_dog+7.0+1.251—30
fitted_group_dog+8.0+1.408—30
fitted_group_dog+9.0+1.558—30
fitted_group_dog+10.0+1.699—30
fitted_group_dog+11.0+1.830—30
fitted_group_dog+12.0+1.951—30
fitted_group_dog+13.0+2.062—30
fitted_group_dog+14.0+2.162—30
fitted_group_dog+15.0+2.251—30
fitted_group_dog+16.0+2.329—30
fitted_group_dog+17.0+2.395—30
fitted_group_dog+18.0+2.449—30
fitted_group_dog+19.0+2.493—30
fitted_group_dog+20.0+2.524—30
fitted_group_dog+21.0+2.545—30
fitted_group_dog+22.0+2.555—30
fitted_group_dog+23.0+2.555—30
fitted_group_dog+24.0+2.545—30
fitted_group_dog+25.0+2.525—30
fitted_group_dog+26.0+2.497—30
fitted_group_dog+27.0+2.461—30
fitted_group_dog+28.0+2.417—30
fitted_group_dog+29.0+2.366—30
fitted_group_dog+30.0+2.308—30
fitted_group_dog+31.0+2.245—30
fitted_group_dog+32.0+2.178—30
fitted_group_dog+33.0+2.106—30
fitted_group_dog+34.0+2.030—30
fitted_group_dog+35.0+1.952—30
fitted_group_dog+36.0+1.872—30
fitted_group_dog+37.0+1.790—30
fitted_group_dog+38.0+1.706—30
fitted_group_dog+39.0+1.623—30
fitted_group_dog+40.0+1.539—30
fitted_group_dog+41.0+1.456—30
fitted_group_dog+42.0+1.374—30
fitted_group_dog+43.0+1.293—30
fitted_group_dog+44.0+1.214—30
fitted_group_dog+45.0+1.137—30
fitted_group_dog+46.0+1.062—30
fitted_group_dog+47.0+0.990—30
fitted_group_dog+48.0+0.920—30
fitted_group_dog+49.0+0.853—30
fitted_group_dog+50.0+0.789—30
fitted_group_dog+51.0+0.729—30
fitted_group_dog+52.0+0.671—30
fitted_group_dog+53.0+0.616—30
fitted_group_dog+54.0+0.565—30
fitted_group_dog+55.0+0.516—30
fitted_group_dog+56.0+0.471—30
fitted_group_dog+57.0+0.429—30
fitted_group_dog+58.0+0.389—30
fitted_group_dog+59.0+0.353—30
fitted_group_dog+60.0+0.319—30
fitted_group_dog+61.0+0.288—30
fitted_group_dog+62.0+0.259—30
fitted_group_dog+63.0+0.232—30
fitted_group_dog+64.0+0.208—30
fitted_group_dog+65.0+0.186—30
fitted_group_dog+66.0+0.166—30
fitted_group_dog+67.0+0.148—30
fitted_group_dog+68.0+0.131—30
fitted_group_dog+69.0+0.116—30
fitted_group_dog+70.0+0.103—30
fitted_group_dog+71.0+0.091—30
fitted_group_dog+72.0+0.080—30
fitted_group_dog+73.0+0.070—30
fitted_group_dog+74.0+0.061—30
fitted_group_dog+75.0+0.054—30
fitted_group_dog+76.0+0.047—30
fitted_group_dog+77.0+0.041—30
fitted_group_dog+78.0+0.035—30
fitted_group_dog+79.0+0.031—30
fitted_group_dog+80.0+0.027—30
fitted_group_dog+81.0+0.023—30
fitted_group_dog+82.0+0.020—30
fitted_group_dog+83.0+0.017—30
fitted_group_dog+84.0+0.015—30
fitted_group_dog+85.0+0.012—30
fitted_group_dog+86.0+0.011—30
fitted_group_dog+87.0+0.009—30
fitted_group_dog+88.0+0.008—30
fitted_group_dog+89.0+0.007—30
fitted_group_dog+90.0+0.006—30

The problem it is built around

An observer who simply regresses toward the average of the last few orientations — with no memory of the previous trial as such — produces exactly the attractive curve in Figure 1. So does an observer who tends to leave the dial near where it last was. A naive analysis cannot tell these apart. In the red-team study, the standard lag-1 model called a central-tendency observer serially dependent in 100% of 24-observer experiments; the adjusted model TimeMind uses did so in 5.5% (nominal 5%). The red team page shows every route to a false positive that was tried, including two that the adjusted model does not fully close.

Separating a perceptual from a decisional account needs more than a better model; it needs a design that makes the two predictions differ. Asking for the orthogonal orientation on some trials — a manipulation introduced by Cicchini, Mikellidou and Burr (2017) — does that. Without it the recovery errors of the two amplitudes correlate at −0.85; with half the trials orthogonal they correlate at −0.44 (design optimisation).

What exists

A working orientation-estimation task with measured stimulus timing, session recovery and a frozen exclusion policy; ten computational models from a no-history baseline to a Bayesian observer; parameter recovery, model discrimination, power and failure analyses; a preregistration-ready protocol; and a manuscript framed, as it must be, as methods work. The primary effect is fixed in advance: the adjusted model’s lag-1 stimulus-history amplitude, tested two-sided by an observer-level bootstrap (here p < 0.001 for the synthetic cohort).

What does not exist is a single human trial. The human results page says so, and will say so until study_status.json changes for a real reason.