Chain alert waits for three failed minutes; Composition app back up

A fix to the alerts of Satsuma's chain checker was deployed, in a session run with Claude in Claude Code. The checker is a job that asks the chain's RPC endpoint whether it is answering and whether new blocks keep coming. It watches the chain and raises an alert when either check fails.

The alert for an unreachable RPC endpoint now fires only after 3 failed minutes in a row, so a single slow request no longer raises it. The alert for a stalled chain is unchanged. The checker's first run after the deploy, at 16:40 UTC, reported the RPC healthy.

The same session brought back Composition, an app on the machine that writes a heartbeat every minute, a regular signal that it is still running. It had stopped: it quit cleanly at 23:47 with exit code 0, and the only crash reports on the machine are from October 24. It was reopened and has written its heartbeat every minute since 00:10. Nothing was lost while it was down, because the automated publishing jobs it normally feeds kept running without it.

Two things stay open. What closed Composition is unknown; the system log does not record it, and it may have been closed by hand. And "The keepalive job restarts it only after a crash, not after a normal quit, which is why it stayed down until the alert fired." A change so the app also returns after a normal quit was offered, with no decision recorded.