Four spike history features gave no prediction gain in a ridge test of ten fly pairs with 5 Hz input bursts

We tested whether exact input spike history helps predict voltage increments after recording, burst number and spike rank are known. We used ten paired DM1 to lateral horn recordings from the Kim et al. fly dataset, version 1. The selected sample had 3,369 clean events and 93 trials at a fixed 5 Hz burst rate. Each complete test trial was excluded from all fitting and penalty selection. Four history features gave a mean error decrease of -0.059%, with equal recording weights. The ridge rule removed the timing block in 92 of 93 test fits. Planned window, overlap and model checks gave small changes. A strong constructed timing signal gave a 70.43% decrease, so the code can recover a large effect. A later comparison found a 3.44% gain from voltage slope before the spike. Moving that slope to separate earlier samples gave 0.53%, with a descriptive panel range across zero. An exact calculation shows that a shared voltage sample can predict 21.694% of increment variance under independent noise with no neural signal. That value is not a noise estimate for the recordings. The main result limits one predictor family within already recorded pairs and one protocol. It does not establish absence of timing effects or identify a release mechanism. Animal IDs were not verified, and anatomy came from another fly. The package preserves selected source voltages, complete saved spike lists, code, all results, plans, figures, source hashes and offline repeat instructions.

Drosophila · synaptic dynamics · spike history · connectome · voltage measurement · prediction · short term plasticity