When uptime percentages lie about reliability
A 99.9% figure can hide long tail pain if your application analytics never see the journeys that actually stalled.
Uptime percentages feel definitive. They are also easy to game with the wrong aggregation. If your application analytics only count synthetic probes hitting a health endpoint, customers can still stall on a dependent queue that never appears in the uptime rollup.
Start by asking which user-visible steps must succeed for a period to count as “up.” Then check whether those steps emit events your reliability metrics actually consume. Gaps between marketing uptime and operational reality usually show up as support tickets without matching alerts.
A practical fix is to pair availability SLIs with completion SLIs — for example, successful sync finishes per attempt window — and review both in the same weekly reliability pack.