Synthetic checkout tests run in real browsers on real networks. Occasional flakes happen — proxy blips, slow third-party scripts, a momentary CDN hiccup. If every blip pages Slack, merchants mute the channel. Then a real store-side fault goes unseen.

Confirmed repeat, same fingerprint

Friction Observatory only treats a critical merchant fault as pageable after the same issue appears again on the same shop, market, and device. The first signal is labeled for confirmation; the repeat is when we page.

• Same store — not a different client shop
• Same market — US flake ≠ GB outage
• Same device — mobile vs desktop can diverge

What we do not page for

Infrastructure and monitor-side problems (proxy, rate limits, our runner) stay out of merchant critical alerts. Those are our issue to fix — not a store accusation. Weekly digests summarize health without the noise of every run.