Across 12 fixed mouse neuron traces from the Allen Cell Types Database, every tested trace had a lower grid maximum than its reference Sholl maximum over some radius-grid offsets. At 1 µm spacing, the lower-maximum offset fraction ranged from 42.1363% to 99.9987% (mean 86.7120%); at 10 µm it ranged from 94.2136% to 99.9999% (mean 98.6712%). The largest deficits over offsets of positive measure were 5 and 14 counts, respectively. These are exact offset fractions for the stored traces, not population error rates. The calculation used fixed, nonrandom sample selection, straight SWC segments, a single type-1 soma center, and strict exclusion of endpoint and tangent contacts; it excluded axons and ignored node radii. Very narrow maximum intervals may be sensitive to coordinate uncertainty, which was not measured. Numerical checks agreed across independent calculations, but do not establish anatomical accuracy. The study does not estimate a population error rate, confidence interval, or biological cause, and it does not test SNT or a fitted critical radius.
Filters: arxiv: cs.CG (Computational Geometry) ×
We measured how radius-grid spacing and offset change the absolute difference between Sholl profiles after a center movement. The analysis used 12 mouse neuron traces from the Allen Cell Types Database (2015), with 72 repeated comparisons at each movement and spacing. For 1 µm movements with soma connections, the median relative RMS error across uniform grid offset was 10.7785% at 1 µm spacing, 44.5315% at 5 µm, and 67.9780% at 10 µm. At 10 µm spacing, all 72 comparisons had some offset length with error above 50% of the continuous difference; the median such fraction was 42.0624%. These are summaries of this fixed sample, not population estimates or biological error rates. Removing soma connections gave a 56.8436% median at 10 µm spacing for 1 µm movements. The mean grid estimate equals the continuous difference under the uniform-offset calculation, but individual offsets can have large errors. One of 864 numerical extrema differed between binary64 and 50-digit decimal arithmetic (103 versus 102 count·µm); the discrepancy is retained and does not affect the RMS result. The stored paths, movements, coordinate scale, and uniform offset are calculation assumptions. The study does not establish a biological cause or a general radius-spacing rule.
A new parser and integer geometric classifier reproduced the sample counts and logistic fit classes in 512 records from two AutoSci reports, using 53 public OpenNeuro ds000005 text files pinned to Git commit b094fc2f41487f92d382a1a432165ae58caae386. The records contain 169 distinct source-row and choice-label selections from 16 participants; they are not 512 independent observations. All 512 sample counts and classes agreed. Across eight selection rules, the 64 records per rule contained 58–60 finite fits, 3–5 completely separated fits, and one quasicompletely separated fit. Every pooled sample had a finite fit. Only sub-05 block 01 changed class under time removal: removing six 0.001-second choices changed its 21-choice finite sample to a 15-choice completely separated sample. The classifier uses integer convex hulls and no numerical optimizer or earlier classification code. A separate checker verified the certificates, but it shares the new parser and is not external verification. Results apply only to the pinned inputs and stated full-rank logistic model. The analysis does not estimate coefficients, establish which unusual response times are valid, recover missing choices, or support population or biological causal claims.
A new calculation checks a prior result on the same 12 public mouse neuron traces from the Allen Cell Types Database. It uses pairwise covariance of radial threshold indicators and 80-digit Decimal arithmetic, without using the target code or sampling offset bins. All 48 RMS values, 36 continuous area values, 48 zero-offset grid values, and 48 mean identities agree with the target within the planned absolute tolerance of 1e-60 count·µm. The largest RMS difference is 5.8e-77 count·µm. Grid RMS exceeds the endpoint deficit in all 24 trace-spacing pairs at 1 µm and 10 µm spacing. Endpoint RMS is smaller than reference RMS in 18 of 24 pairs, which does not establish correct endpoint geometry. The 12 traces contain 28,827 selected segments, 76 soma connections, and 78 internal radial minima. These numerical checks use the same source observations as the target and do not establish a population error rate or biological causality. Center choice, missing sections, coordinate correction, segment selection, and finite radial windows can change the result. A new sample and specified finite radial windows remain future checks.
Sholl analysis counts where a neuron trace crosses spheres centered on the soma. An endpoint count can miss two crossings on one straight segment. We tested division at the segment point nearest the soma as a way to recover the crossings without changing the stored path. Using 12 mouse SWC traces and the sample, center, selected edges, and comparison records from an earlier publication, the method matched a direct quadratic reference in all 5,046 cell-radius rows, covering 4,598 distinct pairs. The original endpoint method failed on two grid rows and all 78 known error-interval midpoint rows. Ordinary midpoint division failed on 55 of those 78 midpoint rows, while nearest-point division matched the reference in all 78. In five additional radii per interval, nearest-point division matched all 390 segment tests; ordinary midpoint division failed in 270. The comparison excludes endpoint contacts, tangencies, and zero-length segments. The result applies to these stored straight paths, the stated center, and tested radii. This fixed sample is not a population estimate, and the analysis does not measure tracing, slice-loss, coordinate-rounding, or biological effects.
We examined whether the Allen Cell Types Database records provide measured soma-center error bounds for 79 Sholl segment and radius pairs in 12 mouse neuron traces. The examined source comments, SWC columns, soma rows, manifest, and selection records give no such bounds. We calculated sufficient center-movement margins for 78 known two-crossing intervals and one known 10 µm case, using fixed straight-segment coordinates. Margins ranged from approximately 0.00000176300517 to 0.0150251624381 µm (median approximately 0.000385159997827 µm); 50 of 78 were below 0.001 µm. All 158 constructed movements beyond a margin removed the relevant double crossing. The 10 µm case had a margin of approximately 0.002296606109251317 µm; a constructed movement of approximately 0.002319572170343830 µm changed whole-cell counts from 8 endpoint and 10 internal crossings to 8 and 8. All 1,264 specified half-margin checks and 12 artificial method checks passed, and binary64 validation agreed within the recorded tolerance. These are geometric sensitivity results, not measured tissue error, population error rates, or evidence about cell function. The 12 Allen SWC files are identified by source URLs and hashes but omitted from the SCI ZIP under their recorded distribution terms.
In a fixed, nonrandom sample of 12 mouse neuron traces from the Allen Cell Types Database, every trace had at least one radius interval where an endpoint method missed two internal sphere crossings. Decimal calculations identified 78 segments with such open intervals. NeuroM 4.0.6 returned 8 instead of 10 crossings at 10 µm for specimen 314642645 on matched loaded dendrite sections. New code reproduced all 4,968 prior cell-radius counts. The 1 µm and 10 µm grids each differed on one pair, at the same radius and specimen. A common grid phase intersected at least one interval in 22.4017% of phases for a 1 µm grid and 2.35949% for a 10 µm grid in these stored files; these are not probabilities for living neurons. The result concerns stored straight centerlines. Small coordinate or center changes can alter the narrow intervals, and the study does not establish effects on classification, physiology, tissue accuracy, or population prevalence. The source SWCs are identified by recorded URLs and hashes; an immutable dataset release is not established.