Object Structure

The logical structure of a serialized object is as follows:

An object is an unordered set of key/value pairs enclosed in a start-object and end-object delimiter.

object
  start-object *(key value) end-object

A key is an integer value.

key
  integer

A value can be a nested object, an array, or a primitive.

value
  object
  array
  primitive

An array is an ordered collection of values enclosed in a start-array and an end-array delimiter.

array
  start-array *value end-array

A primitive is an octets value (byte array), a Unicode string value, an integer, a float, a Boolean true or false, or a null value.

primitive
  octets
  string
  integer
  float
  true
  false
  null

That's it! The serialized representations of the undefined elements is determined by the specific encoding chosen at runtime.

Compared to JSON

The logical structure of a serialized object is almost identical to JSON, with the following exceptions:

The top-level element MUST be object. JSON allows the top-level element to be a bare value (which is never a good idea, so I'm just taking that gun away from you).
Member key types are integers as opposed to strings in JSON. This is a trade-off between human readability and protocol efficiency, and the winner here is efficiency.
JSON does not have an octets primitive type, so byte arrays must be converted to strings.

All of the differences from JSON apply, and additionally: