The tissue number, and it is much smaller than the constructed one

Run: gate/decimation.py → gate/decimation.json Date: 2026-09-21. The measurement METHOD §3.5 was written without.

Question or issue resolved

Two runs this day bracketed a question without answering it. gate/RESULT-REPLACEMENT.md measured, on a constructed system, that an exact functional replacement of a system's units moves every screened measure to between 0.53× and 1.79× of its value. gate/RESULT-PITCH.md established that recording pitch across this register's own human supply varies by 3.1×. §9.4.1 forbids converting the first into a claim about tissue, so §3.5 was written with exact equality of pitch as its comparison rule — conservative, decidable, and unsupported by any measurement on a brain.

This is the measurement. The same recordings are read at successively coarser subsets of their own contacts. Nothing about the brain, the stimulus, the admitted trials or the parameter set changes; only which contacts are admitted. Three arms, because "fewer channels" and "wider spacing" are different manipulations and conflating them is how this would have produced a confident wrong number:

At matched channel count, spread against cluster is the pitch effect with count held constant — the same construction as the duplication null that made gate/RESULT-REPLACEMENT.md's claim narrow enough to trust.

Two substrates: mouse surface EEG array (dandi-000458, behind NS-0006 and NS-0011), 16 state-cells across 6 subjects; and human subdural grid (ds004370, behind NS-0012), one subject, computed under NS-0012's own issued parameter set including its declared 15 ms departure.

The controls

The first is the one that matters. This is not a similar pipeline; at every contact admitted it returns the register's own numbers exactly, so what changes below is the decimation and nothing else.

Finding: on tissue, pitch moves the number far less than the constructed run implied

Mouse, pitch effect at matched channel count, 16 cells per row:

There is a graded effect and its direction is the interesting part: widening the spacing while holding the extent raises PCIst. That is mechanically sensible — adjacent contacts on a dense array see overlapping signal, and spreading the same number of contacts over the same area samples more independent sources, which a complexity measure reads as more complexity. It is not an artifact to be removed; it is what the measure does.

The magnitude is the finding. A 1.26× pitch difference moves the median reading by 8%; 1.66× by 17%; 2.26× by 32%. The constructed system moved by up to 79% under a change that altered nothing at all. Tissue is far less sensitive than the constructed test bed suggested, and §9.4.1 — which forbade carrying the constructed number across — was right to exist.

In the human grid the effect is not detectable at all: across pitch ratios from 1.77× to 4.17× the PCIst ratio runs 0.82 to 1.10 with no trend, median 1.03. That is one subject, seven subsets, and it is a weak basis for a negative. It is reported as what it is rather than as agreement with the mouse.

Finding: extent matters more than pitch, which vindicates an existing caveat

Holding pitch and shrinking the recorded extent to 0.67× moves mouse PCIst to a median 0.761, IQR 0.60–0.99, with individual cells reaching 0.18. Against that, widening pitch 1.49× at full extent moves it to 1.014.

NS-0012 already carries a caveat saying that local subdural grid coverage "may be unable to observe long-range integration," and gate/RESULT-COVERAGE.md asked the extent question directly. That caveat now has a number behind it, and the number is larger than the one grain contributes. The register spent the day discovering that it had no rule for resolution, and the measurement says the rule it already had — about extent — was the more consequential one.

What this corrects in gate/RESULT-PITCH.md, written two hours earlier

That document said the register's single cross-cohort agreement, NS-0026 at 35.83 against NS-0012 at 32.26, "was taken at grain differing by roughly 1.3×" and called it a cross-grain comparison nobody declared. That claim leaned harder than the evidence supports, and on the wrong quantity.

It rested on measured nearest-neighbour distance. §3.5 was revised in the same hour to compare nominal pitch, for the independent reason that a measured quantity carries estimation error — and on the nominal figure both deposits are clinical grids at 10 mm, which makes the comparison within-grain rather than across it. On top of that, the effect at 1.3× measures 8% in mouse and is undetectable in human, against a difference between the two readings of 11%. The comparison is not compromised.

What survives from that document is narrower and still holds: pitch across the register's human supply genuinely varies by 3.1×; ds004696 at 3.50 mm and ds004774 at 9.43 mm are treated as interchangeable supply and are not; and ds004370 declares no coordinate units and its grid contacts measure 21–41% below nominal, which fails §3.5's consistency check and is a real defect in the deposit behind three issued readings.

The correction is recorded here rather than by editing the earlier file's claim away.

Conclusion

§3.5's comparison rule is revised a second time, and this time from a measurement on tissue rather than from caution. Exact equality of nominal pitch is replaced by a tolerance of 1.3×, because at that ratio the measured effect is 8% in mouse and undetectable in one human subject — below the smallest state contrast this register publishes, wakefulness against propofol at 1.21×. Beyond 1.3× the §7 caveat is required, and the caveat now states a measured sensitivity instead of a worry.

Three versions of one section in one day, each replacing the last for a stated reason: exact equality because nothing was known; nominal rather than measured because the first rule made nearly every reading unissuable; and a measured tolerance because the effect was finally measured. The section is better for it, and the sequence is on the record rather than compressed into a clean-looking rule that arrived fully formed.

What remains unmeasured is the human case. One subject is not a cohort, and the mouse and human results do not agree. The extension is cheap — six more PRIOS subjects are public, already listed, and the loader now exists.