Opacity is observed in rule interactions besides counterfeeding and counterbleeding, one of which being ‘self-destructive feeding’. In this paper, I revise the typology of self-destructive feeding, and highlight two
characteristics shared among all cases of self-destructive feeding neglected by previous literature: they all involve instances of non-derived environment blocking, and attempt to simplify consonant clusters with a less common method. This paper offers a new analysis involving underspecification and contextual faithfulness capturing both selfdestructive feeding and all relating characteristics. The new proposal shows not only that opacity can be captured using Parallel Optimality Theory once the UR is more properly understood, but also that selfdestructive feeding is better analysed as an epiphenomenon of phonologically-conditioned allomorph optimisation, rather than a separate type of interaction between two rules that non-restrictively apply in a language. Finally, if all relevant phenomena can be reanalysed in alternative ways, self-destructive feeding might be dispensed with.