Across several languages that encode grammatical gender, an intriguing pattern emerges under ellipsis. Whereas certain noun pairs disallow gender mismatches altogether between the antecedent and the ellipsis site, a second set of noun pairs allows them freely, and a third set allows them only when the grammatically masculine noun is in the antecedent but not vice versa (Bobaljik & Zocca 2011). We illustrate this pattern through the lens of Spanish and argue that the empirical generalizations can be captured via a universal identity condition regulating ellipsis that is split into two statements, where each statement refers to different syntactic primitives. On the one hand, the identity condition requires featural non-distinctness, which is a weaker requirement than strict featural identity. On the other hand, the identity condition requires that ROOTs, unlike features, be strictly identical. Coupled with the independently needed mechanism of repair-by-ellipsis, we argue that the proposed identity condition can provide insight into capturing the microvariation that is attested across languages, within languages, and between individual grammars: whereas the identity condition does not vary, the featural representation of nominals varies idiosyncratically. We elaborate on the relevance of repair-by-ellipsis for this empirical domain and the identity condition, arguing that certain lexical gaps cannot be repaired (Mendes & Nevins 2022).