Problem
When telemetry breaks, we find out because someone notices a chart looks wrong — often long after the break. There is no automated signal, so we cannot trust that a flat or empty metric means "nobody used it" rather than "the pipeline is broken".
What we'll do
- Short term: an AI prompt that checks tables for empty or anomalous data and flags it, stored in the repo so it is reviewable and reusable. Some automation of the prompt checking weekly or bi-weekly.
- Ongoing: quarterly telemetry reviews with a named owner, to catch drift the automated check does not.
Done when
- A check runs on a schedule and flags empty or anomalous tables.
- Quarterly telemetry review is on the calendar with an owner.
Problem
When telemetry breaks, we find out because someone notices a chart looks wrong — often long after the break. There is no automated signal, so we cannot trust that a flat or empty metric means "nobody used it" rather than "the pipeline is broken".
What we'll do
Done when