A Dissociation Test for Causal Attribution Under Partial Observability: Separate Memory Channels for Rate Pressure and Capacity Damage in Li-Ion Batteries

This preprint presents a dissociation-based attribution framework for separating two latent degradation channels in lithium-ion battery capacity trajectories under partial observability. We define two lightweight memory proxies representing rate-type stress (charge-rate history) and baseline-type stress (manufacturer batch offset), and evaluate their selective identifiability using ablation tests, permutation controls, bootstrap stability analysis, and a negative identifiability case study. Results demonstrate cross-dataset dissociation under controlled conditions and principled non-significance under structural confounding. The work is positioned as a falsifiable test of a lifecycle attribution hypothesis rather than a claim of universal validation.