Skip to content

Record whether each request went through headroom in its usage record #396

Description

@Mearman

A /v1/messages usage record carries the session id, model, token counts and both cache fields, but not whether the request crossed headroom. Without that, the cache-read ratio of headroom sessions cannot be compared with non-headroom ones, so the assumption that headroom's compressed prefix is byte-stable across turns cannot be checked, and the later cache-aware decisions have nothing to be measured against.

Add the flag to the record (and its schema and reader), set by the door where it decides to apply the hop. It also lets the per-session cache-write reporting in #157 separate headroom effects from identity switches.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions