At a 20 ms time constant floor, removal of same-type T4 excitation reduced gain but kept 4 Hz direction selectivity in 13 units of nine fly visual models

We tested the role of recurrent excitation in 50 released flyvis 1.2.0 models. The primary set used the ten models with the lowest source task loss. Thirteen central T4a/T4b units from nine models met fixed response and tuning limits for opposite 4 Hz gratings. We removed the four same-type T4 connection rules and kept all other parameters fixed. With a 20 ms time constant floor, preferred response gain fell in all 13 units. The median retained gain was 0.718. All 13 kept their preferred direction and a two-direction Fourier index of at least 0.10. The mean index changed from 0.417 to 0.385, but five individual indices increased. Removing Mi4 input increased median preferred gain to 2.916 and left only three units above the same direction index limit. A finer solver step and the original finite eye kept all 13 T4-removal tuning calls. Native 5 ms settings gave the same qualitative result. Across all 50 models, same-type T4 removal kept positive direction preference in all 54 qualifying units; 50 kept the tuning limit. Gain fell in 52. These are fixed model results, not measurements from flies. They separate a consistent recurrent gain effect from a variable effect on direction contrast. The package preserves source archives, data provenance, code, traces, figures and repeat instructions.

connectome · Drosophila · motion detection · T4 · Mi4 · recurrent excitation · causal intervention · flyvis · model validation · direction selectivity