signals-scout-replay-vision

Compare each enabled scanner's observation count this week versus the prior week on `$recording_observed` events, grouped by stable scanner_id. Then use vision-scanners-observations-list with status filters to distinguish a stopped scanner from one whose observations all fail or land ineligible.

Query daily yes-rates for monitors, mean scores for scorers, or tag shares for classifiers over the past 28 days, comparing the latest week against the scanner's own prior weeks. Require roughly 30 sessions per week before treating a shift as meaningful.

scanner_output_tags is a JSON-encoded string, not a native array, so a bare arrayJoin errors or returns garbage. Wrap it in JSONExtract(..., 'Array(String)') first, and union scanner_output_tags_freeform to avoid missing freeform tags.

Only succeeded observations write events, so zero rows can mean no scanners exist or that every observation is failing, ineligible, or quota-skipped. Check vision-scanners-list for enabled scanners and vision-quota-retrieve for org-wide exhaustion before concluding non-adoption.

No, its scopes are read-only for scanner configuration and observations. When an aggregate finding warrants a sharper standing watch, it recommends a scanner change in the report and lets the team decide.

Skip disabled scanners, drops explained by a dated config edit, low-volume scanners under 30 sessions per week, single-session findings owned by the push path, and anything already covered by a dedupe, noise, or addressed scratchpad entry.