Create a recurring audit.
An audit sweeps a window of your events on a schedule and files findings for
the failure patterns it sees. schedule_interval_secs must be between 3600
(1h) and 604800 (7d) and defaults to daily; lookback_window_secs must be
between 3600 (1h) and 7776000 (90d) and defaults to 7d. schedule_anchor
fixes the phase — runs land on anchor + N * interval — and defaults to the
next 09:00 UTC; an anchor more than 365 days out is rejected. The first run
is queued immediately on create regardless of the anchor.
Authorizations
Body
Reference context to store WITH the definition. Create only — an update sending it gets the same 422 as the two fields above.
It lives here for one reason, and the reason is a race, not convenience:
the insert trigger queues an eval job whose next_attempt_at defaults to
now(), so a new enabled audit is due the instant its row commits. A
client that created the audit and then wrote its context in a second
request could be overtaken by the dispatcher, and the first run — the one
an operator watches — would argue without the brief they just typed.
Writing both in one transaction removes the window rather than narrowing
it: the job row and the context rows become visible together.
Why an update must NOT accept it: _audit_to_body (CLI) and
auditToInput (dashboard) rebuild this whole body from an allowlist
before every save, so a context field on the update path would be wiped
by any unrelated edit — 200 OK, no error, brief gone. Create has no
read-merge step and therefore no such hazard. See the module comment in
routes/audit_context.rs.
Fixed phase for the schedule: runs land on anchor + N * interval.
Omitted on create → default_schedule_anchor() (next 09:00 UTC, matching
the 09:00 backfill in 20260721000200). Omitted on update → unchanged.
NULL in the DB is still tolerated by the dispatcher (legacy drift path)
but nothing this route writes can produce it any more.
Response
Audit created. Returns id and created_at.

