# `PhoenixKitAI.GuardWatch`
[🔗](https://github.com/BeamLabEU/phoenix_kit_ai/blob/v0.26.1/lib/phoenix_kit_ai/guard_watch.ex#L1)

Notices when a task's model stops fitting it, and signals it
(`PhoenixKitAI.GuardEvents.model_unfit/4`: one event a day per task and
endpoint).

The process (started by `PhoenixKitAI.children/0`) keeps, for every
enabled task and every endpoint it uses now (`Tasks.endpoints/1`), the last
fitness that was known (`:fit` or `:unfit`; not knowing leaves the last
known one as it was). It looks again when the capability cache changes
(`{:capabilities_refreshed, _}`: a catalogue refresh, a model alias that
moved) and when an endpoint is created, saved or deleted. A pair that was
`:fit` and is unfit now is signalled. A pair first seen unfit — at start,
or an endpoint a task has just begun to use — is not: it is shown on the
Tasks page, and a consumer's form refuses to save an unfit choice.

Each node runs its own; the event's dedupe key keeps the signal to one.

It also drops this node's cached task limits when a task's limits change
anywhere (`{:task_limits_changed, task_key}` on `TaskLimits.topic/0`), its
copy of the active blocks when a block is set or lifted anywhere
(`{:guard_blocks_changed, group}`, same topic), and both whole when it starts.

After every look it emits `[:phoenix_kit_ai, :guard_watch, :look]`
(measurement `pairs`, metadata `trigger`: `:first_look`, `:capabilities`
or `:endpoint`), so whoever needs to know that a change has been looked at
can wait for it.

# `child_spec`

Returns a specification to start this module under a supervisor.

See `Supervisor`.

---

*Consult [api-reference.md](api-reference.md) for complete listing*