A burrito is a collection of content plus metadata. That collection may be made available in various formats, such as a zip file, an Amazon S3 bucket or a series of API calls. The term “burrito” describes the content, not the distribution mechanism.
Burritos contain ingredients. An ingredient is a file-like resource with a mimetype and, optionally, a canonical scope. Burritos typically contain several types of ingredient.
Burritos exist in a number of flavors. A flavor describes a way to represent one class of entity as a burrito. Each flavor corresponds to a high-level entity, such as a Scripture text project or a sign language dictionary. A flavor specification typically includes ingredients with multiple mimetypes.
Flavors are typically quite broadly defined. Additional constraints may be added using conventions. For example, an audio convention may specify that audio files represent whole chapters of Scripture, or that they are arranged according to a specific hierarchy. Burrito creators should respect any convention they include in the metadata. Burrito consumers may use conventions to decide how or whether to process a burrito. The semantics of no specified conventions is caveat emptor, ie nothing should be assumed about the content of the burrito beyond what is specified for the burrito flavor.