With fixed fits in 73 mouse retinal scans, adding depth gave a larger error decrease when at least one response quality index was below 0.5

We tested whether scan coverage explains a quality limit on a retinal forecast from anatomy. We reused fixed predictions from SCI f1c62e61-5328-4fb9-bf19-528914bb12ff. Each test mouse record was excluded from model fitting. All regions had local and full-field response quality indices of at least 0.3. High quality required both indices to be at least 0.5. We compared high-quality and lower-quality regions within 73 common scans from 25 date-and-mouse records. The sample had 995 high-quality and 3,486 lower-quality regions. Depth was constant within a scan. We weighted scans equally within each record, then records equally. The mean error decrease from adding depth was 0.001011 squared normalized response units larger in the lower-quality group. A fixed-prediction 95% record bootstrap range was 0.000261 to 0.001718. Eighteen of 25 records had a positive contrast. Neither quality group alone had a clear error decrease. An exact arithmetic decomposition placed 56.3% of the earlier all-region versus high-quality gain shift within common scans. Thus, changes in scan and depth coverage do not fully describe the shift under these weights. Stricter scan counts and two simpler depth models kept the contrast direction. Quality uses the recorded responses, including the target trace. The study cannot separate noise, cell mixtures, and response selection. It concerns fixed fits and normalized shape, not absolute release strength or a causal depth effect. The package preserves code, a compact loss dataset, source hashes, all results, figures, and repeat instructions.

retina · bipolar cells · structure and function · response quality · matched scans · glutamate · forecast evaluation