Severity: warning · Fires after: 6h · Rule: central/prometheus/rules/visibility.yml
An engine's trailing-28-day impressions for a site are more than 25% below where they were a week ago. Because the gauge is a 28-day sum it moves slowly — a total loss of impressions only crosses this line on about day 7 — so this is a large, sustained fall, not a bad afternoon.
webmaster_impressions
< 0.75 * avg_over_time(webmaster_impressions[1d] offset 7d)
and avg_over_time(webmaster_impressions[1d] offset 7d) >= 50
The >= 50 floor silences tiny sites. Blind spot: if Prometheus has no data
7 days back (stack outage), there is no baseline and the rule cannot fire until
one exists.
The fleet was offline in transit 2026-08-20 → 09-08. Google's 28d impressions for wiki.wikantik.com went 961 → 241 → 7 → 1 across the following month and nothing alerted; it was found by reading the dashboard on 2026-10-01.
search-visibility dashboard → Google index coverage by state: pages
moving into "Crawled – currently not indexed" or "Server error (5xx)" means
de-indexing, not a ranking change.wikantik-traffic → What crawlers were served: are crawlers getting 200s?reliability dashboard, and up{job="wikantik"}.bin/visibility report --site <site> --engine <engine> for the query-level view.It clears by itself once the 7-day-old baseline has also fallen — i.e. it
reports the fall, not the low level. Recovery work is in Search Console
(validate fixes, resubmit the sitemap) and bin/visibility submit --commit for
Bing/Yandex.
wikantik.com and wiki.wikantik.com are overlapping properties and usually
fire together. Related:
IncidentSeoIndexedPagesDropped.