Last good value — fall back to the previous success

Pattern

A dashboard refreshes every 30 seconds. If a refresh fails — DON'T show an empty screen. Keep the last known good data, but honestly mark it "stale, last updated 2 minutes ago".

The problem it solves

Imperative code often resets data = null before a refresh, and after a failure the user sees emptiness instead of yesterday's data. Bad UX. On the other hand, hiding the fact that it's stale is also bad.

Operators and why they matter

• catchError(() => of({type: 'error'})) inside the refresh — a failed request becomes an event, not a stream failure.
• scan — holds the last successful value between attempts. On success it updates; on error it keeps the old one + marks status=stale.
• startWith('ok') — the initial refresh.

Gotchas

• catchError OUTSIDE the trigger stream — the first error stops all future refreshes. The stream dies.
• Don't hide the stale/error status. The user must know they're seeing old data.
• For critical data (account balance, medical data), a last good value may be business-unacceptable. Decide per case.