Homeostatic bioelectrochemical stacks: living layers can self-balance against voltage reversal

Bioelectrochemical stacks connected in series are limited by voltage reversal: layers with unequal current capacity cannot match the imposed current, reverse polarity, degrade further, and drive the stack into a positive-feedback collapse. Every reported mitigation is electronic (balancing resistors, per-cell converters) and is unavailable to a passive membrane stack such as a reverse-electrodialysis (RED) device. We propose and model an architecture in which the proton gradient of each layer is regenerated in situ by a living biofilm rather than supplied by an external salinity feed, and we ask whether biofilm growth provides negative feedback strong enough to overcome reversal. Using a coupled electrochemical–metabolic model (layer polarization, Monte Carlo over layer-to-layer variance, logistic biofilm dynamics, and flux-balance analysis of a curated core metabolic network) we find that a passive stack diverges, with the coefficient of variation across layers rising from 18.5% to 66% and power collapsing, whereas an otherwise identical living stack converges to 10.3% and retains power. Convergence holds whenever the reversal damage rate and the growth rate satisfy k_rev/μ_max ≲ 0.8, a ratio that is essentially invariant with μ_max. The residual variance is set by structural (fabrication) dispersion, not by inoculation dispersion, which growth erases entirely. After the deliberate killing of one layer the modelled stack recovers 95% of its power in 12.6 days while a passive control does not recover. Metabolic calibration reveals the governing trade-off: every proton diverted to the external circuit is a proton that does not pass ATP synthase, so power and growth compete for the same budget; and because activity falls with acidity, an optimal ΔpH exists near 1.5–1.6 rather than "as large as possible". We do not claim competitive power density — silicon photovoltaics outperform the photosynthetic front end by an order of magnitude in area — but rather a distinct operating regime characterised by self-levelling, self-repair and growth-based manufacture. We set out a falsifiable bench programme whose first experiment, the measurement of k_rev, can refute the central claim. This is a modelling study. No experimental data are reported. All code is included as supplementary material.