For equal activation and linear sums, a layer rule gave no clear prediction gain for 30 Hz input shares in 54 mouse V1 maps

We tested whether cortical layer labels predict changes in the relative share of excitatory input during repeated activity. We used open mouse V1 connection maps at 30 Hz. The test had 907 connections in 54 maps from 33 recording dates. All 69 eligible maps from 39 dates supplied training data outside each test date. We assumed equal input activation and linear voltage sums. Mean total variation between first and second layer shares was 8.48%, with a descriptive 95% date range of 6.82% to 10.15%. The largest source layer stayed the same in 51 of 54 maps. Mean variation across individual connection shares was 20.00%. A rule based on target type and source layer reduced excluded-date squared share error by 0.030%. A date bootstrap that repeated model fitting gave a range of -101.4% to +23.5%. A later nested shrinkage test gave 7.69% lower error. Its fixed-prediction range was -4.94% to +19.65%. Thus, neither rule showed a clear gain over static input shares in this sample. Ratio errors, response thresholds, partial input sampling, young slice data, and the linear sum assumption limit interpretation. The result does not establish in-vivo routing or a release mechanism. The package preserves source data, version and license records, code, all predictions, training ratios, numerical checks, figures, and offline repeat steps.

visual cortex · short-term plasticity · cortical layers · synaptic input maps · paired-pulse ratio · input shares · model validation · open data