Habitat amount control is necessary but not sufficient to resolve the fragmentation debate

Opposing conclusions from the same global multi-taxa dataset have intensified debate over whether fragmentation effects can be inferred independently of habitat amount in observational landscape studies. Gonçalves-Souza et al. (2025) reported lower local- and landscape-scale diversity in fragmented landscapes, whereas Fahrig et al. (2026) reanalysed the same dataset with continuous, scale-matched predictors and concluded that no independent evidence for fragmentation remains once habitat amount is controlled. I argue that this stronger adjudicatory inference is not established by the reanalysis. Controlling for habitat amount is necessary, but it is not sufficient when the realised predictor structure may still fail to provide the contrast required for separately estimated coefficients to be interpreted as independent ecological effects. Supplementary design-based analyses show that the continuous predictor redesign remains embedded in the original directional landscape contrast, that same-amount contrasts are scarce, and that benign linear collinearity diagnostics do not exhaust dependence in the raw predictor pair. The point is not that the opposite ecological conclusion has been demonstrated, but that a near-zero additive coefficient under these conditions is not self-interpreting. More broadly, the comment argues that landscape-scale studies using additive habitat-amount control cannot treat coefficient separation as proof that independent ecological effects have been identified unless predictor separability is demonstrated empirically in the realised design.