A MongoDB 1,815 record is a document: an ordered list of field-value pairs that looks like a JavaScript object and travels the wire as BSON, a binary encoding with real types. That one decision explains most of this chapter. Queries reach into nested fields with dot notation because documents nest; indexes span array elements because arrays are first-class; writes to one document are atomic because a document is the unit the storage engine reads and writes.
BSON adds 64-bit integers, exact decimals, dates, binary blobs and ObjectIds — none of which JSON can express — and length-prefixes every document so the server can skip fields without parsing them. Knowing what each type costs keeps you clear of the two classic MongoDB bugs: money in floating point, and documents that grow without a ceiling.