Safe and effective conflict resolution is critical for survival and reproduction. Theoretical models describe how animals resolve conflict by assessing their own and/or their opponent's ability (resource holding potential, RHP), yet experimental tests of these models are often inconclusive. Recent reviews have suggested this uncertainty could be alleviated by using multiple approaches to test assessment models. The mantis shrimp Neogonodactylus bredini presents visual displays and ritualistically exchanges high-force strikes during territorial contests. We tested how N. bredini contest dynamics were explained by any of three assessment models-pure self-assessment, cumulative assessment and mutual assessment-using correlations and a novel, network analysis-based sequential behavioural analysis. We staged dyadic contests over burrow access between competitors matched either randomly or based on body size. In both randomly and size-matched contests, the best metric of RHP was body mass. Burrow residency interacted with mass to predict outcome. Correlations between contest costs and RHP rejected pure self-assessment, but could not fully differentiate between cumulative and mutual assessment. The sequential behavioural analysis ruled out cumulative assessment and supported mutual assessment. Our results demonstrate how multiple analyses provide strong inference to tests of assessment models and illuminate how individual behaviours constitute an assessment strategy.