Each trial fires two stimuli in rapid succession: S1 at t = 0, S2 at t = SOA. Every task passes through three serial stages — perception (stimulus identification), response selection (deciding which response to make), and motor execution (producing the response):
RT = t(perception) + t(response selection) + t(motor)
The response-selection bottleneck hypothesis (Pashler, 1994; Welford, 1952) holds that this middle stage is a single-channel, central resource: it can process only one task's response selection at a time. Task 2's perception can proceed in parallel with Task 1's, but if Task 1 is still occupying response selection when Task 2 is ready, Task 2 must queue:
RS2_start = max(perception2_end, RS1_end)
slack = RS2_start − perception2_end (queueing delay, ≥ 0)
RT2 = RS2_start − SOA + RS2 + motor2
This produces the classic Psychological Refractory Period (PRP) effect: as SOA shrinks, RT1 stays essentially flat, but RT2 grows — the two curves trace an "underadditive" slowdown that is the empirical signature of a structural bottleneck rather than a general slowing of both tasks.
- SOA slider — gap between the two stimuli. Small SOA → more queueing → bigger RT2 cost.
- Task difficulty sliders — lengthen a task's perception stage, shifting when it becomes "ready" for response selection.
- Bottleneck toggle — turn the central limit off to see the counterfactual: with parallel response selection, RT2 barely depends on SOA at all.
- The purple pillar in the scene is the shared response-selection gate — only one glowing token may occupy it at a time; a queued token pulses red while it waits.
Real-world relevance: this bottleneck is why texting while driving, or answering a phone call mid-task, costs measurable extra time on the second action — attention's central decision stage cannot truly multitask.