Temporal Computation Audit for Spiking Neural Networks
Initial public release This is the first public software release accompanying the manuscript: A Theory-Grounded Diagnostic Framework for Temporal Computation in Spiking Neural Networks The framework separates four properties that are often conflated in the analysis of spiking neural networks: temporal capability; realized temporal utilization; representation-level temporal dependency; intervention-conditioned temporal provenance. Included in this release cleaned, import-safe Python source code; no notebook dependencies; no runtime package installation or cloud-storage mounting; English-only Python source files with module-level introductions; paired synthetic experiments for count, weighted-time, and GRU readouts; depth-2 and depth-4 confirmatory analyses; nested temporal-value estimation; nine exact per-neuron-count-preserving perturbations; activity-normalized downstream count-change analysis; two label-blind, count-exact restoration methods; SHD external validation; LIF and adaptive-LIF neuron-family controls; exact parameter-count analysis; raw-versus-normalized NCD comparisons; restoration-rule agreement analysis; decision-threshold sensitivity analysis; probe null and sample-size diagnostics; computational-cost measurements; SHD preprocessing and duration audit; frozen aggregate results, publication tables, figures, and provenance records; automated repository validation and unit tests. Frozen confirmatory provenance Final run ID: 20260720_114724 Frozen run key: b434371e5a55f5b8 Final code version: tds-final-2026-07-20-v1 TNNLS post-hoc code version: tds-tnnls-experimental-completion-v1 Primary hidden-path position: distance 1 from the readout Confirmatory seeds: 20 paired seeds Synthetic depths: 2 and 4 Validation The release package was checked using: python scripts/validate_repository.py python -m compileall -q src scripts tests pytest tds-final-pipeline --profile smoke --device cpu --research-root ./Research