SPK timing
Single-unit omission onset latency. 187 units resolve out of 4,130 eligible units.
A1 · Census funnel
| Step | n | of eligible | meaning |
|---|---|---|---|
| 1. Eligible units | 4,130 | 100.0% | every spike-sorted unit with an analysable omission arm |
| 2. Omission response detected | 637 | 15.4% | a change from baseline exists (this is DETECTED, not resolved) |
| 3. Omission latency resolved | 187 | 4.5% | independent estimators agree to within the spread criterion |
| 4. Stimulus reference defined | 139 | 3.4% | the unit also has an analysable stimulus arm, so ΔT exists |
| 5. |ΔT| > 50 ms | 96 | 2.3% | descriptive subset; not an inferential threshold |
Naming hazard, stated explicitly: in the underlying receipts the
column named resolved counts detected responses
(637), not resolved latencies
(187). The public tables rename these to
n_omission_detected and n_latency_resolved so the two cannot be
confused downstream.
A2 · Per-unit resolved latency
One point per resolved unit, sorted by latency, with Tom ± 1.96 × bootstrap SD. Subject is encoded by colour and response direction by marker shape, independently — area is available on hover rather than as a visual grouping, so the display cannot suggest an area hierarchy the design cannot support.
Median Tom = 190 ms (IQR 120–250 ms; 86 increases, 101 decreases). All onsets are one-sided by construction: the estimator search window begins at the omitted event, so no unit can report an onset before it. A pre-zero deviation lands on the bound or returns undefined — it is never negative.
Source table: analysis6a_spk_resolved_public.csv (187 rows).