The stimulation protocol moves the measure more than the states do

Run 2026-09-20. Code: gate/chocs.py, gate/analyze_chocs.py. Output: gate/chocs.json (276 rows). Deposit: OpenNeuro ds003848, tasks SPESclin and CHOCS1, subjects RESP0749 and RESP0800.

Every value here is the declared-departure response window, [15, 300] ms — the column every issued reading uses, emitted as pcist_departure. The documented [0, 300] ms window is computed alongside for disclosure and runs materially higher because it still contains the amplifier saturation; it is now emitted as pcist_documented rather than pcist, after an analysis elsewhere in this session read the plain name and took the contaminated column for the real one (gate/RESULT-STREEF.md, defect 4).

Question or issue resolved

Whether the stimulation protocol — the shape and rate of the pulse, as distinct from its intensity — changes the measure, and if so whether it changes it by more or less than consciousness does.

METHOD 2.1 makes stimulation intensity a required field because intensity varies the quantity 5.4-fold. It says nothing about waveform or repetition rate, which means two laboratories could follow the standard exactly and still produce incomparable numbers. Until now nothing in the register tested that gap, because no deposit carried two protocols on the same patients.

ds003848 does. It holds an undocumented second task alongside the usual clinical one, on the same implanted contacts, in the same session, which makes the protocol the only thing that differs and the question answerable.

The design

CHOCS1 is an undocumented task in ds003848, found by enumerating tasks rather than checking the ones already believed to be there (gate/RESULT-SOURCES.md). It is a systematically different stimulation protocol:

Two subjects carry both. Matching on (subject, site, current) leaves 46 cells in which the patient, the implanted contacts, the stimulation current (2 mA throughout) and the behavioral state (awake) are all identical, and the protocol is the only thing that differs.

Geometry is depth electrodes (SEEG), not a subdural grid. Under METHOD §3.3 that bars these absolute values from being set beside the register's grid-derived ones, and leaves the paired contrast unaffected, because geometry is held constant within every pair.

The result

Paired on the matched variant, which equalizes trial count between the two arms:

Paired ratio CHOCS/SPES: median 2.069. CHOCS higher in 35 of 45 usable cells. Sign test p = 2.5 × 10⁻⁴.

Why the three variants matter here

The unmatched figure is inflated and should not be quoted. CHOCS1 contributes far more trials than SPESclin — 852 against 329 in RESP0749, 1,549 against 416 in RESP0800 — and more trials average to a cleaner evoked response, which raises PCIst independently of protocol.

That the matched and floor2 estimates agree at roughly 2.0, while the unmatched one inflates exactly as trial-count theory predicts, is the useful part. The effect survives at the lowest common trial count, so it is not a trial-count artifact; and the pipeline's three variants behaved as designed, which is a check on the pipeline as much as on the protocol.

The comparison that gives this its weight

Changing the stimulation waveform and rate moves PCIst further than losing consciousness does, on this deposit, in these units.

METHOD §2.1 fixes stimulation intensity per cell because intensity varies the quantity 5.4-fold. It says nothing about waveform or rate. On this evidence it has to: a standard whose protocol is unpinned admits a source of variation larger than the signal the instrument exists to detect.

The outlier, and why it does not carry the result

The ratio range is 0.248 to 3,588.7. The extreme is one cell — RESP0800 VL01-VL02, SPESclin PCIst 0.0166 against a 74.2 µV response, so PCIst collapsed despite a large evoked signal. It is the same failure mode as the three No components → PCIst=0 cells, all of which were also in the monophasic arm.

Trimming two cells from each end leaves the median at 2.069, unchanged. The sign test uses direction only and is unaffected by magnitude. Only one cell in 45 has a SPESclin PCIst below 1.0.

Limitations, stated

• Two subjects. The between-subject medians are 1.207 (RESP0749) and 2.585 (RESP0800) — a spread as large as the effect. Two patients cannot establish a population value, and this result should be read as establishing that the protocol matters, not how much.
• One current. Every matched cell sits at 2 mA, so nothing here says whether the protocol effect varies with intensity. Given §2.1, it very likely does.
• One geometry. SEEG only. Whether the same holds on a subdural grid is untested and, under §3.3, not assumable.
• Two factors confounded. Waveform and rate differ together. This result cannot say whether biphasic stimulation, 1 Hz repetition, or their interaction is responsible, and the deposit contains no cell that separates them.
• Pulsewidth also differs, by 25 µs (1.000 vs 1.025 ms). That is a 2.5% difference and is not a plausible source of a 2-fold effect, but it is a third uncontrolled factor and is recorded here rather than waved away.