With 5 to 10 calibration traces, separate factors did not clearly reduce second-response error in 66 mouse V1 inhibitory connections at 30 Hz

We tested whether each recorded output of a parvalbumin-positive (PV) cell needs a separate response factor when calibration data are limited. We used 66 PV-to-pyramidal connections from 25 sources and 17 mice in Znamenskiy et al., Figshare version 1. Odd and even accepted trials formed separate calibration and test groups, with five to ten calibration traces per connection. Both rules received the measured first test response. One rule fitted a shared factor across the source outputs. The other fitted a separate second-to-first ratio per connection. Mean mouse squared voltage error was 0.017160 for the shared rule and 0.019309 for separate factors. This was a 12.5% error increase. The central 95% whole-mouse resample range allowed a 15.9% decrease to a 55.7% increase. A partly pooled rule selected outside each test mouse gave a 5.9% decrease, with a range from a 20.2% decrease to an 8.0% increase. Fixed voltage windows gave no clear separate-factor gain. First-versus-last trial groups gave a 46.4% separate-factor error increase. All 66 source connections passed an independent raw-trace check. A full input rebuild and a separate numerical repeat passed. The result supplies a compact benchmark for dynamic rules on measured connection graphs. It does not establish uniform biological plasticity or predict intact-circuit activity. Selection, source annotations, short trace groups, and one study limit inference. The package preserves code, data provenance, inputs, selection, results, figures, resamples, and repeat instructions.

short-term plasticity · inhibitory connections · parvalbumin · source neuron · paired-pulse response · trial separation · mouse visual cortex · model calibration