This arXiv preprint (not yet peer-reviewed) combines an evolutionary model with a preregistered lab experiment using five-person groups to show how democratic redistribution of public-good returns can stabilize cooperation, yet participation costs rapidly shrink the active electorate and reward abstention. Unlike equal-share controls, zero-cost voting produced sustained contribution gains over time, but even modest costs shifted players toward democratic free-riding rather than anti-social voting. The study thereby reveals a structural mechanism—cost-induced thinning of the electorate—that makes institutional decay less visible to participants themselves. Prior work on endogenous institutions, such as Dal Bó et al. (2021) in the American Economic Review on voting rules and cooperation, documented similar pro-social effects of democracy but rarely isolated participation costs; the current model adds the missing dynamic by showing costs suppress contributor-rewarding equilibria under strong selection while simultaneously widening perception-behavior gaps. Related patterns appear in field data on real-world turnout costs (e.g., Braconnier et al., 2017, American Political Science Review), where small frictions disproportionately remove marginal cooperators and leave policy skewed toward low-engagement equilibria. Limitations include the online, one-shot group format and modest group size, which may understate repeated-interaction effects or network spillovers present in actual civic settings. The core insight—that costs erode the visible base of democratic maintenance without triggering corrective awareness—directly explains persistent shortfalls in collective-action domains from climate agreements to local budgeting.