JSON is a simple de facto standard cross-language textual format used to represent, exchange and store data and documents in computer systems. JSON Schema is a description language, based on JSON, proposed to describe JSON types and validate JSON data. We investigate over 57,800 distinct public schemas for various defects through static analysis, and identify cases of mistyping, misplacing, misnaming, misspelling, misversioning and other miscellaneous issues. Over 60% of schemas are defective, allowing in the worst case unintended data to be validated. These findings suggest to make key changes to the current JSON Schema draft so as to limit potential issues. It also leads us to design JSON Model, an alternative compact and expressive JSON data structure description language.