# Adversarial review

Verdict for C1: major_issues. Significance: moderate.

## Scope and blindness

This is a reading and static-design review, not a computational reproduction. No assigned program was executed or imported. I examined the paper, claim and evidence mappings, frozen plan, parameters, preparation and analysis source, tests, materials, source manifest, environment, pilot, result tables and downloaded array schemas. Four externally delivered data files were downloaded and checked against their assigned byte counts and SHA-256 digests. All corresponding plan/source/input pairs inspected were byte-identical. The paper's Provenance and plan contain an author byline; knew_author is therefore true. I did not seek the author's identity or other reviewers' judgments.

## Main objection: the negative result's biological scope

The reported primary correlation and threshold decision are internally consistent: the supplied lower bound is greater than the preregistered -0.20 threshold. This supports a restricted statement about the chosen expression score and two-dimensional endothelial-centroid residual in this primary dataset, conditional on the inference assumptions. It does not establish that energy-demanding neuron types lack closer access to capillaries, nor that vascular supply matching operates instead through regional density or blood-flow regulation. Those mechanisms are not compared by this analysis.

The score averages standardized transcripts from a separate 10x population; it is not measured ATP expenditure, oxygen consumption, or firing-related demand in the spatial brains. The target is an endothelial-cell centroid rather than the vessel lumen, vessel surface or reconstructed capillary network, and the selected endothelial subclass is not restricted here by vessel caliber. Pericytes provide an additional cell-centroid proxy with similar sampling limitations, rather than independent measurement of the vascular structure. Shared proxy error can attenuate the association in all brains and sensitivities. A predeclared correlation cutoff does not validate either proxy.

Required fix: limit C1 and the title/discussion to the measured transcript-score versus local centroid-distance relationship, and make the biological interpretation explicitly conditional. To sustain the broader null, quantify sensitivity to realistic proxy error or validate against vessel geometry and an independent energy-demand measure. A positive-control or injected-effect calibration should establish that the measurement and local normalization can recover a biologically relevant proximity effect, not merely a shuffled null.

## Inference and the exclusion threshold

The primary bootstrap samples subclasses independently. The paper itself acknowledges lineage dependence, and the supplied output contains 239 subclasses in 27 classes. The class-centered secondary statistic changes the estimand; its bootstrap still samples individual subclass rows and does not model class-level dependence or uncertainty in the spatial brains. The division analysis repeats subclasses across divisions while resampling rows. Consequently these analyses do not demonstrate that the primary confidence interval maintains coverage under shared lineage/spatial structure. The claim that a moderate biological association is excluded needs this uncertainty addressed.

Required fix: provide dependence-aware uncertainty, for example a justified hierarchical or class-block sensitivity with animal-level replication estimates and their uncertainty. State what population the interval refers to. Restrict conclusions if the exclusion of -0.20 does not survive the appropriate uncertainty analysis. The large cell count is not a count of independent animals or independent energy-score observations.

## Local normalization and replication

Subtracting each tile's mean controls the stated local baseline by construction, but also conditions out part of any effect mediated by spatial placement of neuron types across regions. Tiles with little subclass mixing can provide limited within-tile information; no mixing/positivity diagnostic is supplied. The inference therefore concerns within-neighborhood placement, not all architectural matching of supply and demand. Report subclass mixing and effective spatial support, and explain the conditional estimand.

The four replication correlations have heterogeneous coverage (224, 169, 198 and 38 subclasses) and include a positive estimate of about 0.344 in the smallest brain. Counting negative signs and pooling by neuron number are useful descriptive checks, but do not constitute an animal-level equivalence test. Provide replication uncertainty rather than characterizing all five brains as a definitive absence of a relationship.

## Percent-distance language

The supplied subclass mean residual range is approximately -0.0395 to 0.0383. Exponentiating these mean log residuals describes a relative geometric-distance quantity under the local baseline, not a bound applying to each neuron's distance or to each pair of types. Specify this interpretation and do not infer uniform capillary access from small averages.

## Primary-source context actually examined

Tsai et al. (2009), DOI 10.1523/JNEUROSCI.3287-09.2009: the accessible primary-paper search text and PubMed abstract describe three-dimensional soma-to-microvessel geometry, illustrating the distinction from endothelial centroids in thin sections. https://pubmed.ncbi.nlm.nih.gov/19923289/ and https://pmc.ncbi.nlm.nih.gov/articles/PMC4972024/ . The full-text open was blocked by a browser check; no full-text-access claim is made.

Yao et al. (2023), DOI 10.1038/s41586-023-06812-z, and Zhang et al. (2023), DOI 10.1038/s41586-023-06808-9: primary abstracts and retrieved article excerpts support the atlas/taxonomy context, not validation of this new energy-demand proxy. https://pubmed.ncbi.nlm.nih.gov/38092916/ and https://pubmed.ncbi.nlm.nih.gov/38092912/ . Publisher full-text opens failed; this is not a comprehensive literature/novelty review.

## Integrity and overall assessment

The node reports no orphan numbers, missing required sections/files, citation-list mismatches or data warnings. My static inspection found clear parameterization, documented pilot and null tests, unchanged plan/code pairs and broad sensitivity reporting. These strengthen the descriptive analysis but do not resolve the identified measurement and inference gaps. I found no basis to assert a numerical mismatch, and performed no reproduction. The contribution has moderate significance as a focused new atlas analysis if its claims are restricted to what its measures and uncertainty establish. The broad biological refutation currently requires major changes.