Why every reading is a mammal, measured rather than guessed

Run: gate/substrate_census.py → gate/substrate_census.json Date: 2026-09-22.

What I did

Three attempts to issue a non-mammalian reading produced three different blockers, each found only by going after that deposit specifically — too few channels on the organoids, no continuous voltage in the cultures, an unstated license on the reptile. Discovering them one deposit per session is the failure this run ends.

Every catalogued deposit is now screened against the same four gates, in the order the standard applies them: source admissibility, whether a perturbation was delivered, whether the signal is the kind the reference computation takes, and channel count. The run records verification level per deposit and never conflates the two — census means the deposit's own electrode and acquisition tables were opened and read, registry means the catalogued description was used and the files have not been touched.

Along the way I fetched and opened DANDI:001611, twelve dissociated rat cortical cultures on HD-MEAs, which was the run I recommended last turn.

The culture deposit has the finest grain in the register and cannot be read

982 channels, median nearest-neighbour spacing 35.0 µm, minimum 17.5 µm, extent 4.31 mm. Against a human subdural grid at 10 mm that is a grain span of roughly 286× — far and away the widest the register could reach, and exactly what §3.5's comparison rule was built to survive. Electrical stimulation is recorded, 256 trials.

acquisition/ is empty. The deposit ships 950 sorted units and their spike times. There is no continuous voltage anywhere in the file, and the reference computation takes an evoked voltage response — a channels × time matrix. Spike times are a different object, and converting them is not a parameter departure but a different measurement. Refused.

The census, and two mis-specifications caught inside it

32 deposits, 14 clades, 10 of them non-mammalian. No non-mammalian deposit is confirmed issuable.

Two defects in this run's own design were caught before it reported, and both would have produced a confident wrong headline.

The clade split was a substring search. The first version matched species keywords against a blob of each deposit's name, modality, subjects and notes. It classified the mouse deposit as non-mammalian, a crow as a mammal, and put five human EEG studies in the non-mammalian column. A classification that decides the run's headline cannot be a substring search; it is now an explicit per-deposit mapping with a control that refuses to report if any deposit is unclassified.

perturbation: other was passed through as a perturbation. It is a catch-all that includes sensory paradigms, and a spatial oddball delivers nothing to the substrate. On that error the turtle and crow deposits appeared to clear all four gates, and the run would have reported two non-mammalian substrates as ready on a guess. They are now marked unverified and require a census before any such claim.

The structural finding: it is the field's data, not the standard

The dominant blocker is gate 2. Fifteen of thirty-two catalogued deposits deliver no perturbation at all, which caps them at M2 under §3.4 — within-subject contrast only, never comparable to an M1 value. An M2 reading cannot join a cross-substrate scale.

Stack that against gate 3 and the shape is clear:

• The substrates with the finest grain and the most interesting phylogeny — nematode, octopus, dissociated cultures — ship spikes or calcium, not voltage.
• The substrates that do record voltage mostly have no perturbation delivered.
• The intersection, perturbational voltage, is overwhelmingly mammalian clinical and rodent work.

So the register is mammal-only because perturbational voltage recording is mammal-dominated. That is a property of what neuroscience has deposited, not a choice this standard made and not an oversight in how it was applied. Only one of the six blocking conditions — the channel minimum — belongs to the standard at all, and it is already marked provisional.

The strategic consequence is the uncomfortable part. If the M1 tier requires a voltage response, the cross-substrate scale is structurally limited to mammals, and the AI bridge inherits that limit directly — an artificial system has no voltage either. The question this raises is whether METHOD should admit a second declared signal type, with its own tier and its own validation, rather than treating spike and calcium responses as outside the instrument. That is a larger change than anything in this standard so far and it is not made here.

The mechanism, so this is not rediscovered again

The census now runs as part of the standing cycle and its totals are reported on the register's face alongside conformance, rather than living in a result document that has to be remembered. A deposit's verification level travels with it, so the difference between screened from a description and opened and read is visible in the report rather than in someone's memory.

What I recommend next

Census the four perturbation: unverified deposits — turtle, crow, octopus, nematode — because they are the only candidates left where the answer is not already known, and between them they carry three non-mammalian clades. The turtle deposit is 27 MB, so the first of them costs minutes.

Then take the signal-type question seriously rather than deferring it again. The decision of whether a spike-train response can carry a Nooscope reading determines whether this standard ever measures anything that is not a mammal, and it should be settled on its merits before more effort goes into deposits that cannot clear a gate the standard may be wrong to have.