Does PCIst survive being sampled at the rate a worm is imaged at?

Run: gate/tempres.py → gate/tempres.json Date: 2026-09-22 Standard: METHOD 0.4.8 → 0.4.9 (adds §3.6, condition 12)

---

What was done

The register's only two non-mammalian deposits clearing all four census gates are both C. elegans, and both are the same experiment: Randi et al., Nature 2023, single-neuron optogenetic stimulation with whole-brain calcium imaging. This run set out to issue the register's first non-mammalian reading from it.

The deposit was opened rather than assumed. DANDI:001075 holds 223 assets; 113 are raw imaging volumes of 8–60 GB, and 110 are companion segmentation files of 2–5 MB that carry the tracked traces. Those were streamed by byte range — remfile over the DANDI API, no 55 GB download — then pulled in full (1.6 GB for all 110) and surveyed.

The deposit is a genuine M1 perturbational recording. It has a PatternedOptogeneticStimulusTable with start and stop times, a targeted ROI per stimulus, a tracked population response and a simultaneous red reference channel. The median worm gives 3,800 frames, 125 neurons, 32 minutes, 54 stimulations of 500 ms at ~31 s spacing, across 19 distinct targets with up to 5 repeats.

It is imaged at 2.0 Hz. All 110 worms, without exception.

That is the whole question, and it is not a question about worms. PCIst is a principal-component decomposition of an evoked response followed by a state-transition count, and both steps are bounded by how many samples the response window contains. The register's issued readings carry 399 samples (mouse) and 207 (human). A worm's calcium response runs for tens of seconds, so the physiologically honest response window is 20–30 s — which at 2 Hz is 40 to 60 samples.

Rather than assume, this run did to time what gate/decimation.py did to space. The sixteen mouse state-cells behind NS-0006, NS-0011 and NS-0027 were re-read at their issued parameter set with the resample rate swept down and nothing else changed. Two arms, because the worm has two deficits and conflating them would let one be reported as the other:

rate     resample swept so the response window holds N samples; trials untouched
trials   trial count swept to the worm's 2-5; rate untouched

Four controls, run first, with the run refusing to report on any failure.

---

The value does not survive. The direction does.

Control 1 reproduced all sixteen register values with worst fractional disagreement 0.000 — this is the pipeline that produced the register's numbers, not a lookalike.

Monotone, and tight enough that it is the measure rather than noise. At the worm's operating point the same recording returns about one third of its value, and not one of the sixteen stays within 20%.

But the ordering is untouched. Awake over isoflurane held in 6 of 6 subjects at every sample count tested, down to 10 samples — median contrast 2.70 at full rate, 1.82 at 60 samples, still 1.41 at 10.

Trial count behaves oppositely. Swept alone, at the 2–5 trials a one-stimulus-per-target paradigm affords:

The two faults are different in kind. Sparse sampling makes the number smaller and keeps its direction. Few trials make the number unreliable and lose its direction — at 2 trials one subject reads anesthesia higher than wakefulness, a 15× interquartile spread. A substrate carrying both carries both, and they cannot be summed into one caveat.

The nematode reading is refused

Not because the worm is a worm. Because 2 Hz imaging with a median of 2 repeats per target cannot carry this quantity onto the register's scale. A nematode PCIst printed beside a mouse 61.35 would be the §3 scale error committed in a new axis — the same mistake, made by the run auditing against it, that gate/RESULT-WIDENERS.md records.

What would be issuable is a within-worm contrast, since direction survives grain. The deposit affords none: every worm is in one state. The wild-type against unc-31 comparison the atlas offers is a manipulation of the substrate, not a state of it.

The register's own readings already span this axis, undeclared

This is the finding that outlasts the refusal. Realized response-window samples across every computed reading:

A 1.93× spread, sitting in the steepest part of the curve above, never declared and never checked. This is gate/RESULT-PITCH.md again in the time axis: §3.5 exists because spatial grain moves the number, and nothing had asked whether temporal grain does. It does, by more.

No claim in the register rests on a direct mouse-to-human absolute comparison, and §3.4's modality tiers already forbid several that would be affected. The defect is that nothing prevented one.

---

What is recommended next

METHOD §3.6 and condition 12 are added in 0.4.9 — declare fs_hz, resample_hz, response_samples, and trials with the number of distinct sites they are drawn from. No tolerance is set. The curve rests on mouse surface EEG, one cohort, one substrate, and a conformance tolerance nobody has measured is the 0.3.9 pitch defect repeated knowingly. Declaration now; ratio when it has been measured twice.

Three things follow, in order:

1. Run this sweep on a human grid. Until the curve is measured on a second substrate, condition 12 can require declaration but not comparability, and the register's own 1.93× spread cannot be converted from a noted defect into a correction.

2. Stop pursuing the nematode on this deposit. The obstacle is the acquisition, not the animal, and no amount of access work moves it. A non-mammalian reading needs a perturbational recording at electrophysiological rates with repeated trials per site.

3. Read the Farnes ketamine deposit instead (gate/RESULT-ACCESS.md). It is human TMS-evoked EEG, 62 channels at 312 Hz, 10 subjects, wakefulness against sub-anesthetic ketamine, CC0 — a state where behavior and experience dissociate, which is the state most likely to break a measure that tracks behavior instead. It was unreachable yesterday and is reachable today, and it is worth more to this register than the worm was.