A fourth state, and a deposit that cannot carry a reading

Rendered from gate/RESULT-FARNES.md

Run: gate/farnes.py → gate/farnes.json Date: 2026-09-22 Standard: METHOD 0.4.9 → 0.4.10 (amends §2.3)

What was done

gate/RESULT-ACCESS.md recovered the Farnes deposit from a Zenodo mirror after Dryad's API refused it, and named it the most valuable thing the access work turned up. This run computed it.

Farnes N, Juel BE, Nilsen AS, Romundstad LG, Storm JF, PLOS ONE 2020;15(11):e0242056. Ten healthy volunteers, 62-channel EEG, TMS-evoked responses recorded before and during sub-anesthetic ketamine, CC0. The archive holds 20 files — ten subjects × two conditions — each a channels × times × trials array at 312.5 Hz spanning −400 to +400 ms around the pulse, with 270 to 298 trials per cell.

The parameter set is the TMS-EEG one: baseline (−400, −50) ms, response (0, 300) ms. The window starts at 0 rather than at a declared artifact offset because the artifact was checked for and is not there — the archived data is artifact-corrected, and the global mean evoked amplitude rises smoothly out of a baseline whose standard deviation is 0.053 µV. Trials were rejected on the register's own rule, 5 robust SDs of the baseline peak. No resampling was applied, because resampling 312.5 Hz data upward would let this reading declare a grain it does not have.

The register's second external check on an absolute value, and it holds

	Median	Range	n
Awake	53.58	33.09 – 64.17	10
Sub-anesthetic ketamine	50.10	32.75 – 70.91	10
Ketamine / awake	1.012	0.70 – 1.40	10 pairs, 5 above 1

Two things check out at once.

The absolute value lands on the published benchmark. NS-0023 records Comolatti et al.'s alert-wakefulness figure of 47.89 across 108 subjects. This cohort's awake median is 53.58 on an independent dataset, an independent cohort and an independent lab. That is the register's second external check on an absolute value after gate/RESULT-STREEF.md, and the first on scalp TMS-EEG.

The direction reproduces the source paper's own unusual claim. Farnes et al. report that signal diversity of spontaneous EEG rises under ketamine while the evoked response does not. Computed here, the evoked ratio is 1.012, above 1 in 5 of 10 subjects — no effect, which is what the paper says. A control was set before the run to fail if this came back far from 1, because a pipeline that disagrees with the paper that produced its data is a reason to stop rather than to publish.

This matters beyond bookkeeping. Ketamine at these doses is the state where behavior and experience come apart — subjects are behaviorally unresponsive while experience is preserved and richly reported. A measure that tracked behavior rather than experience should have fallen here. It did not move.

The reading cannot be issued, and the reason is not the science

The deposit cannot supply a matched null. §2.3 requires the floor to be measured on sham onsets drawn from the same recording, ≥ 1 s clear of any stimulation, holding the admitted channels, trial count, parameter set and averaging constant. One such onset needs 300 ms of response window, 350 ms of baseline and 1,000 ms of clearance: 1,650 ms of continuous record. The archive contains 800 ms epochs. There are no onsets to draw.

The spontaneous recordings in the same deposit are not a substitute: they are 62 channels at 250 Hz against the evoked 60 channels at 312.5 Hz, so they match neither the admitted channels nor the parameter set the section requires be held constant.

This was a control, set to fail loudly, and it failed:

FAIL  c4_matched_null_available
      evoked_epoch_ms: 800.0
      ms_needed_for_one_sham_onset: 1650.0

So the best perturbational human dataset the access work found cannot carry an issuable reading — not because the experiment was wrong, but because only the stimulus-locked epochs were archived. This is a defect of deposition. METHOD §2.3 now says so, names the arithmetic as the test, and makes the corollary a standing request: deposit the continuous record, not only the epochs. An epoched archive is enough to reproduce the authors' own analysis and not enough to measure the floor that analysis sits on.

The values above are published alongside the refusal so the work is not lost, and they are marked as refused rather than issued.

A second substrate for the temporal-grain curve, free

The same cells supplied what METHOD §3.6 said it lacked. Swept at 312.5 Hz native, 94 samples in the response window:

Samples	80	60	40	30	20	10
Human scalp TMS-EEG	0.984	0.960	0.896	0.819	0.680	0.383
Mouse surface EEG	0.454	0.386	0.330	0.265	0.203	0.151

At 60 samples the two differ by 2.49×. See gate/RESULT-GRAIN-MULTI.md for what that became once a third and fourth sweep separated window length from substrate.

What is recommended next

Write to the Oslo group for the continuous recordings. They archived what their own analysis needed. The register needs 1,650 ms per epoch rather than 800, and a deposit that already carries 20 files of clean, artifact-corrected TMS-EEG at roughly 290 trials per cell is one supplementary upload away from carrying issuable readings in both states, and the contrast between them, in a state the register has never measured.
Publish the deposition requirement as a document of its own. This is the first case where an experiment good enough to read could not be read because of what was archived. It will not be the last, and a one-page statement of what a perturbational deposit must contain is the cheapest standards contribution the Foundation can make.
Do not weaken §2.3 to admit this deposit. The floor is the reason the register can say a value is distinguishable from the instrument at all, and the six mouse recordings behind gate/RESULT-FLOOR.md spread 12.2× — exactly the range that cannot be guessed. A reading published without it would be worth less than no reading.