Neuroscience assessment — technical notes · 2

First query, first lesson. We asked OpenAlex for the fifty most-cited articles since 2000 whose title or abstract mentions each of our five methods. The instrument worked perfectly. The sampling frame did not.

What came back

  • „MEG“ returned three papers on quantum-dot solar cells. In that field MEG means multiple exciton generation.
  • „PET“ was headed by RECIST, the oncology response criteria, with 30 152 citations — more than any neuroimaging paper — followed by lymphoma guidelines, prostate-cancer guidelines and brown-adipose-tissue physiology.
  • „NIRS“ gave remote sensing of crop chlorophyll, gold-nanorod photothermal therapy, an astrophysics paper on the Hubble constant, and a primer set for denitrifying bacteria (the gene is called nirS). Six of fifty were about the brain.
  • „EEG“ had, at rank 2, a 2023 paper on „AI consciousness and cosmological portal dynamics“ with 14 373 citations in three years, in a computer-science proceedings venue. We have not investigated it. We have kept it in the raw file and flagged it, because an index anomaly is data about the index.
  • „fMRI“ was clean: 46 of 50 were neuroimaging papers. Acronyms that belong to one field behave; acronyms that several fields share do not.

The fix, and its cost

OpenAlex classifies every paper into a topic, subfield and field with its own machine classifier. Restricting the query to neuroscience and psychology subfields removed the solar cells, the tumours and the corn. It also introduced an assumption: that the classifier’s idea of „neuroscience“ is ours. A neuroimaging paper filed under radiology or computer science is now invisible. So the next snapshot will run the union of both frames and record the difference. Every such choice goes into the working document as a named assumption, in the same form we use when we say a claim in the literature rests on a hidden step.

One thing the index showed unprompted

Before restricting anything, we asked which topic the fifty most-cited „fMRI“ papers belong to. Twenty-seven of the fifty fall into a single topic: Functional Brain Connectivity Studies. The next topic has three. The citation mass of the field is concentrated in one concept family — resting-state networks and functional connectivity — and the index’s own classifier sees it without being asked. That is where the dependency mapping will begin.

Bookkeeping

The sandbox we work in has no network access, so queries run from a browser and the raw responses are checksummed there before they are saved. Two snapshots exist so far, each with its retrieval time to the second, its exact query URLs, and a SHA-256 digest. Anyone with the URL can re-run the query; the counts will have moved, and the digest tells you by how much the file has not.

Next note: one snapshot, two changed variables — a correction to the numbers above, and what caught it.