Fig. 5. Task sensitivity in the context-dependent experiments.

Shown for the a TI and b TAE experiments is the average across observers of the task sensitivity (d′, Eq. (3)) divided by the orientation difference (Δθ = 2θ) as a function of RT (four bins). The d′ was measured between the −θ° and +θ° target orientations, and averaged across the two contexts (averaging is possible because of the symmetry of the +20° and −20° context orientations). θ is the step size between adjacent target orientations (θ = 1°, 2°, or 3°, see the Methods; for the TI experiments having both θ = 1° and θ = 3° versions, there was no qualitative difference in d′/Δθ between 1° and 3°, permitting an average over observers from both experimental versions). Results showed that sensitivity is mostly stable across RTs, possibly exhibiting a small improvement at the fastest RTs, and a small deterioration at the slowest RTs, clearly unlike the bias dynamics (Fig. 4, Supplementary Fig. 5). We note that due to the strong decision bias at shorter times and lack of sufficient trials, the d′ at shorter times may be under-estimated (see Supplementary Fig. 4). For individual data see Supplementary Fig. 7. Error bars are ± 1SEM.