At a quality cutoff of 0.3, depth reduced full-field waveform error by 5.4% in 30 mice; a cutoff of 0.5 gave no clear gain

We tested whether bipolar terminal depth helps predict full-field glutamate responses when local responses are known. We used the Franke et al. main processed file v1. The primary sample had 6,185 release regions from 145 scans and 30 date-and-mouse groups. Both response quality indices were at least 0.3. Each outer test excluded one complete mouse group. All model choices used other groups. Adding smooth depth terms and products to a quadratic local-waveform model reduced mean squared error by 5.42%. The descriptive 95% mouse bootstrap interval was 2.78% to 7.69%. Twenty groups had lower error. A broad depth split reduced error by 3.34%. Fine depth gave an uncertain 2.15% reduction beyond that split, with an interval from -0.19% to 4.16%. The planned quality-0.5 refit used 995 regions from 25 groups. Its depth model increased error by 5.36%; its reduction interval was -10.59% to 0.51%. A later check of fixed primary fits on those regions also gave no clear gain. The package preserves code, compact inputs, predictions, figures, source hashes, and repeat instructions. All 29 source rebuild and result comparisons matched exactly. These results show a small prediction gain under one quality rule. They do not establish a stable fine-depth benefit, a causal depth effect, or individual inhibitory connections.

retina · bipolar cells · structure and function · terminal depth · surround response · glutamate · validation across mice