A fixed ridge test did not confirm an EPSP width prediction gain from target time constants in 382 human layer 2–3 connections

We tested whether a target cell's membrane time constant improves prediction of excitatory response halfwidth in human layer 2–3 cortex. A fixed ridge model already used cell type, soma depth, and soma distance. Each test excluded one donor. Penalty selection used only the remaining donors. In the main sample of 382 connections from 18 donors, the added time constant reduced mean donor squared log error by 0.71%. A 95% donor bootstrap range that repeated all fits was −2.82% to +3.54%. The planned signal-to-noise ratio (SNR) check kept responses with SNR at least 3. Its 318 connections from 17 donors gave 7.27% lower error, with a complete-refit range of +0.82% to +13.59%. A larger penalty range and a common donor panel retained this sample contrast. Adding source time constants alone gave no clear gain. A simple passive membrane model also gave no clear direct gain over a training mean. Only 391 of 601 input-eligible connections had an available width. Slow responses and quality review both caused missing widths. These local slice recordings are not a complete anatomical connectome. The selected result does not establish prediction accuracy for unmeasured edges. The package preserves new code, source provenance, row selections, predictions, model fits, bootstrap results, figures, and repeat instructions.

human cortex · synaptic integration · membrane time constant · EPSP halfwidth · donor exclusion · connectome models · signal-to-noise ratio