A compressed, buildable specification for a PVT-derived sustained-attention task that fits an under-2-minute phone battery. Companion to Lesson 4. Prints on one page.
| Purpose | Detect sustained-attention decline (fatigue signature: lapses + variability) |
| Duration | ~60 s (target); validate against a 3-min PVT-B before trusting |
| Stimulus | One salient target appears on screen; user taps as fast as possible |
| Inter-stimulus interval | 1–4 s, uniform random — never fixed or patterned |
| Trials | ~15–20 (as many as fit in the window) |
| Response | Single tap; record time from stimulus onset to touch |
| Per-trial feedback | Optional: show RT (e.g., "312 ms"). No cumulative "score to beat" |
| Device note | Subtract/measure touchscreen input latency; validate on real target devices |
| Response speed ★ | mean(1/RT) — reciprocal RT; most sensitive, most normal-distributed |
| Variability ★ | SD(RT) or CoV = SD/mean — the wake-state-instability readout |
| Lapses | count(RT ≥ 500 ms) |
| Minor lapses | count(RT ≥ 355 ms) — keeps signal when trials are few |
| False starts | count(RT < 100 ms OR pre-stimulus) — flag rhythm-tapping/guessing |
In a 60-s task, lean on response speed + CoV (per-trial signal). Lapse counts need many trials to stabilize.
Safe to add: theme/skin, per-trial RT feedback, gentle progress bar, sound.
Never add: predictable rhythm, combo/streak mechanics that reward strategy, a difficulty ceiling that prevents failure, anything learnable across sessions. Each reintroduces practice or ceiling effects and breaks validity.
Feed mean(1/RT) and CoV into the per-user baseline; flag impairment via the Reliable Change Index against that baseline's own variability. Exclude flagged-impaired sessions from the baseline; let the baseline mature past the learning curve first (see Lesson 3).