Token Profile: Descriptor Common Header

The shared field definitions now live in Typed Payload Descriptor. This page stays token-specific: it shows how the common descriptor fields are combined when the payload belongs to a discrete-sequence stream.

Key field combinations in token

Field	What matters in token	Typical values
profile_id	Tells the receiver that the payload belongs to the token family	token
schema_id	Selects the concrete token-sequence template, such as chat delta or tool-result streaming	llm.chat.delta.v1
schema_version	Pins the exact schema interpretation version	1
stream_semantics	Decides how the receiver should append, merge, or otherwise consume the token stream	ordered_incremental
offset / length	Places the current chunk back into the logical sequence	128 / 36
descriptor_flags	Exposes token-side hints such as explicit stop-reason presence	stop_reason_present

What to focus on in token

1. schema_id + schema_version decide how to interpret token_unit / sequence_start / sequence_end / terminal / stop_reason.
2. stream_semantics + offset + length decide whether the chunk should be appended in order, reassembled, or handled under another sequence policy.
3. descriptor_flags is the right place for extra chunk-level hints, not for sampling internals or model-private runtime state.

This descriptor layer anchors token chunks to the right profile, schema, and logical sequence; the actual sequence interpretation still lives in Schema and Body and Payload Frame.