Choose signals before you choose dashboards
If nobody can name the engineering decision a metric supports, it probably should not become a tile.
Notes
Field notes from telemetry reviews
Practical writing for platform and delivery leads — definitions first, dashboards second.
Recovery curves without vanity stages
Mean time to recover is useless if “recovered” means the job turned green for thirty seconds.
Schema rules that survive three tool migrations
Breaking property renames are cheap on day one and expensive on day ninety.
When quality gates lie about exposure
If a gate metric fires on schedule start instead of after the check runs, your pass rate is fiction.