Coverage, tested in the regime that matters

Run 2026-09-19. Code: gate/coverage_human.py. Output: gate/coverage_human.json. Uses the 1,650 cells already in gate/prios.json plus the electrode coordinates published with the deposit. No new data was read.

Why this run exists

gate/RESULT-COVERAGE.md tested whether spatial extent carries the wake-versus-anesthesia contrast, using the mouse deposit, and could not reach the relevant geometry: thirty electrodes spread over an entire cortical surface cannot be subset into the dense local patch a human subdural grid occupies. The conclusion recorded there was deliberately narrow — unsupported over the range testable, and the testable range excludes the case in question.

The human deposit is the case in question. PRIOS publishes electrode coordinates for every patient, the seven implanted grids differ substantially in extent, and every per-site contrast has already been computed.

Test 1 — between patients

Does a patient's contrast track that patient's array extent?

Spearman rho(extent, ratio) = −0.179. Rho(channels, ratio) = −0.143.

Both are the wrong sign for the hypothesis and neither is distinguishable from noise at n = 7. The patient with the widest array (PRIOS02, 135.7) has nearly the weakest contrast; the strongest contrast (PRIOS06, 3.69) comes from one of the smaller arrays and only eight sites.

Seven points cannot carry a conclusion and this table is not asked to.

Test 2 — within patients

A stimulation site near the middle of the array is surrounded by recording electrodes on every side. A site at the edge is not. If the spatial relationship between perturbation and recording matters, central sites should separate the two states better than peripheral ones — and this test has 248 sites rather than seven patients.

Pooled across 248 sites: rho = +0.009. Central sites median 1.201, peripheral 1.206. Difference −0.005, permutation p = 0.926 over 20,000 permutations.

That is as null as a result gets. Not a weak effect in the predicted direction — no effect, in either direction, on the largest sample available.

Verdict

The coverage hypothesis is dead. It failed on the mouse deposit over the range that deposit could test, and it has now failed in the human deposit, in the regime it was specifically invented to explain, on both a between-patient and a within-patient test.

Three documents in this repository asserted it as the leading explanation for the weak human contrast. All three were wrong, and the assertion was made three times before anybody tested it.

What stands in its place

Comolatti, Hassan, Mikulan et al., Brain Stimulation 18(5):1444–1454 (2025), already published the explanation. In one framework, on the same measure:

The intracranial paradigm compresses the contrast, published, before any question of geometry arises. Their absolute intracranial wakefulness value also brackets this register's computed 32.26 (NS-0012, NS-0017).

Limitations, stated

• Test 2 varies the position of the stimulation site within a fixed recording array. It does not vary the extent of the recording array itself, which is the literal form of the hypothesis. It is a weaker instrument than the mouse channel-subsetting was, and it is reported as one.
• A direct within-patient test — shrinking each patient's recording array and recomputing — requires re-reading the recordings and was not run. Given a pooled rho of +0.009 and p = 0.926, the expected value of that run is low, and it is recorded as available rather than pending.
• PRIOS03 contributes four sites and PRIOS06 eight. Neither constrains anything.

Consequence for the register

NS-0012, NS-0013 and NS-0015 carry a caveat naming local coverage as the former leading explanation. That caveat is updated to record that the hypothesis has now been tested twice, in both deposits, and has failed in both.