ata-validator

draft-04 spelled the exclusive bounds as booleans modifying minimum and maximum, and schemas in that shape are still in circulation, OpenAPI 2.0 documents among them.

The engines read the numeric form every dialect since uses, and the compiled path compared the value against the boolean directly, so true became 1. { maximum: 5, exclusiveMaximum: true } rejected 1, which is a wrong rejection, and { minimum: 3, exclusiveMinimum: true } accepted 3, which is a wrong acceptance. The interpreted path ignored the keyword instead, so the two engines disagreed on the same document.

Normalization now rewrites the pair into the numeric keyword before either engine sees it, so both read one thing. false leaves the inclusive bound alone, and a boolean with no bound beside it is dropped.