What changed: the status map is written after the commit, not when the probe returns. The API and the dashboard answer from that committed map, so an in-flight result can never appear as healthy — and a dead writer shows up as staleness, not a frozen green board. This is the property everything else builds on: what you see is what was durably observed.
Proof — the M1 exit drill, live: registered a monitor against the fixture;
up after the first probe; killed the target → three transport checks →
incident open + incident_opened outbox event in one commit; killed the
server mid-incident → still down after restart; target back → two fresh
passes → resolved + incident_recovered. Thirty-one tests, gate green.
First useful release.