What's fixed
collect_update() used a single 6-hour TTL for every result. That meant a release published shortly after a deploy stayed invisible for the full 6 hours, and a momentary GitHub blip stuck for the same window.
Split TTL by result:
| Result | TTL |
|---|---|
| Update available | 6h (unchanged) |
| No update / error | 30 min |
Worst-case API usage at 30m negative-cache: ~48 calls/day per host, comfortably below GitHub's 60 req/hr/IP unauthenticated limit.
If you're running v0.6.0 or v0.6.1 and want to verify the fix the lazy way, just leave it — your dashboard will surface this v0.6.2 update banner within 30 minutes. (Or restart the container to flush the cache and see it now.)
🤖 Generated with Claude Code