# Independent reproduction assessment

Verdict: reproduced for C1. The reference harness reran the complete declared analysis from an empty results directory in a network-disabled Docker container. All claimed result bindings agree within their declared tolerances; the run finished successfully in 66.353 seconds, within the declared ten minutes.

A separately written reconstruction, independent_primary.py, imports none of the author's code. It computes the energy score from the bundled expression and gene tables, reconstructs same-section nearest-endothelial distances from the primary-brain coordinate file, removes within-tile mean log distances using pandas group operations, and aggregates by subclass. It obtains rho -0.01896030378678668 across 239 subclasses and 2,066,795 eligible primary-subclass neurons. The maximum absolute subclass-excess difference from the reported table is 4.999e-10. The original harness rerun, rather than this narrower reconstruction, checks every secondary, bootstrap, and permutation result binding.

Scope: this reproduces the computation conditional on the supplied compact inputs. It does not independently regenerate those inputs from the multi-gigabyte atlas sources, validate the biological proxy for energy demand, or establish the population-level interpretation of the subclass bootstrap. Cell position, transcript-based demand, and endothelial-cell centroids are proxy measurements. A reproduced decision label is not itself a demonstration of equivalence across animals or of absence of a biological effect.

Integrity: the server reports no orphan numbers, missing sections/files, citation inconsistencies, or tabular anomalies. The harness reports no hidden text. The analysis code and corresponding plan code are byte-identical; the plan parameters reside at plan/parameters.json and match data/parameters.json. The supplied paper and plan disclose an author identity, so this reading was not fully blind; no author lookup was performed.

Hazard screen: none under the current rubric. Inspected files describe public mouse atlas data, aggregate gene-expression tables, array coordinates and annotations, analysis code, results, and citations. The five NPZ inventories contain numerical and Unicode arrays, are loaded without pickle, and have finite coordinates and the documented cell-kind codes. No human participant records, credentials, harmful capability, or identifiable unlicensed substantial copy was found. The source materials declare their public-data licenses. This assessment is not a general endorsement of every biological interpretation.