Systemic Tau as a label-free trigger for adaptation in an event-driven spiking controller: a closed-loop demonstration

An embodied neuromorphic system must decide when to adapt as well as how. This preprint tests whether Systemic Tau (τs), the mean pairwise Kendall rank correlation of module activity on a common event index, can make that decision without task labels inside a closed sensorimotor loop. A simulated differential-drive rover seeks a relocating scalar source; its sensor streams are encoded online as Bandt–Pompe ordinal patterns that emit events only when the active pattern changes, and two leaky integrate-and-fire motor neurons are updated only on those events. A fixed adaptation burst is released when the mean absolute change of Kendall τb on pre-specified edges leaves its calibration band. Across 50 seeds, after deletion of half of the sensorimotor synapses the τs-triggered controller reached a post-perturbation performance of 0.669, against 0.208 without adaptation and 0.586 with the same number of bursts on a fixed schedule, and was non-inferior to an alarm driven by the task error (difference +0.001, 95% CI −0.005 to +0.008). After actuator damage it exceeded no adaptation and the fixed schedule but fell short of the task-error alarm (−0.035, 95% CI −0.054 to −0.021), which used 3.5 times as many bursts. Under sensor gain drift the ordinal controller was unaffected and adaptation conferred no benefit, whereas a magnitude-coded controller lost performance and recovered under the same trigger. The analysis plan was committed before the reported seeds were run. The code, configuration, run-level records and figures are archived separately under the MIT license at doi:10.5281/zenodo.23254710. No funding is declared. The text is distributed under the Creative Commons Attribution 4.0 International license.