signals-scout-observability-gaps

Use read-data-schema to list event volumes, then run execute-sql over system.insights checking whether each event name appears in any insight query. Events firing above roughly 100 per day with zero references are coverage-gap candidates.

Extract the events each saved insight filters on via execute-sql over system.insights, then measure recent volume with query-trends. An insight whose primary event has zero firings in 7 days while a similarly named event fires actively indicates drift from a rename.

The skill requires at least 100 events per day sustained across 7 complete days in the project timezone, plus meaningful distinct-user reach. Below that bar, candidates are parked in a watch scratchpad entry rather than reported.

No. Builtin events like $pageview, $autocapture, and $identify are already surfaced through PostHog product views such as Web Analytics and Feature Flags, so recommending custom insights for them is explicitly disqualified.

An empty run is a valid outcome. The scout closes out when the project has too few active events, when coverage is already saturated, or when every candidate matches a prior report, an addressed entry, or a disqualifier such as backfill spikes or internal test traffic.

No. It operates with read-only analytics access and only writes to its scratchpad and the report channel. It files recommendations with suggested actions and reviewer routing; a human decides whether to create the insight, dashboard, or alert.