The central bottleneck and the refractory period
Drag the stimulus-onset asynchrony (SOA) — the gap before Task 2 begins. Perception (light) and motor (light) stages overlap freely, but response selection (solid) is single-channel: when Task 2 is ready to select before Task 1 has released the bottleneck, it waits (hatched red).
RT2 = 600 ms (baseline 500 ms) — a refractory cost of 100 ms, the 100 ms Task 2 spends waiting at the bottleneck. Below 150 ms, every extra millisecond of SOA removes exactly one millisecond of cost — the −1 PRP slope.
Switch cost and the residual floor
On a task-switching trial the task set must be reconfigured. Lengthen the preparation interval and the switch cost shrinks — but it never reaches zero. A residual cost survives however long the warning, the part of reconfiguration the stimulus alone can trigger.
Switch cost = 219 ms — a preparable part of 129 ms plus a residual of 90 ms. Preparation can remove the first but never the second, which is why even a fully predictable switch stays costly.
Multiple resources: interference tracks overlap
Task 1 is fixed — a visual, spatial, manual task, like steering a car. Set Task 2's properties. Two tasks interfere in proportion to the resource pools they share, so a task that loads different pools combines almost for free.
Interference is ruinous. Task 2 shares 3 of three resource dimensions with the steering task. Routing the second task to a different modality or response channel — an auditory, vocal task against a visual, manual one — is the design lever that makes concurrent performance cheap.