When Online Growth Adaptation Helps a Synthetic Continuous Fermentation

Abstract

Online parameter adaptation and biomass protection can alter a continuous fermentation differently during steady operation, activity loss and startup. We execute a two-state Monod chemostat simulation with 24 independently fitted kinetic models, five conditions and four dilution policies, giving 480 eighty-hour episodes. The policies are a frozen steady optimum, an online adaptive steady optimum, a protected adaptive rule and a truth-informed protected reference. No predictive controller, physical fermentation or organism-specific dataset is claimed. After a maximum-growth drop, mean cumulative biomass harvest is 25.54 for the frozen rule, 34.48 for adaptation and 37.78 for protected adaptation in normalized units; only the frozen rule crosses the biomass floor. Under substrate-poor startup, protected adaptation yields 86.07 versus frozen 77.13 and truth-informed reference 78.58, demonstrating that the reference is not a finite-horizon optimum. Adaptation instead reduces harvest under optical drift. The lower growth value used for protection is a heuristic, not a validated confidence bound or safety certificate. All figures and performance quantities are reproducible synthetic outputs, with a generated laboratory scene excluded from quantitative evidence.