Glossary
Format Adherence
A model's ability to consistently follow specified output formats, such as JSON, markdown, bullet lists, or length constraints.
Format adherence measures whether a model follows the structural requirements of a task — producing valid JSON when asked, using markdown headers in the right places, staying within a word count, or following a specific template. In many product contexts, format matters as much as content: a model that produces great summaries but occasionally forgets to use the specified JSON schema can break downstream systems. Format adherence failures are often invisible in manual testing (because individual responses look fine) but show up clearly in production where automated systems consume the outputs. For behavior architects, specifying format requirements precisely — and testing for them explicitly — is a basic but critical part of building reliable AI products.