Why your alerts should get quieter every month
JournalTorus
Noise is a bug, not a cost of doing business
Written by Elsa Thorvald, Head of Reliability
Monday, 3 November 2026
Most teams measure monitoring by how much it catches. We think the better number is how little it says. An alert that fires and needs no action trains people to ignore the next one, and the next one might be real.
What a noisy pager costs
Every page that turns out to be nothing costs about twenty minutes of focus, and far more at night. Multiply that by a rotation of six engineers and a week of false alarms, and you have lost a sprint without shipping a line.
Where the noise comes from
Static thresholds written years ago, checks copied from a template, alerts for symptoms nobody can fix. None of it is malicious. It simply piles up, because adding an alert feels safe and removing one feels risky.
How we cut it down
- Group related signals into one incident
- Learn normal ranges per service and per hour
- Close alerts that fix themselves within a minute
- Review every page that needed no action
- Delete one old check for every new one
Key takeaways
- Quiet is a feature
- Every alert needs an owner and an action
- Learned baselines beat fixed thresholds
- Measure pages, not checks
Noise is a bug, not a cost of doing business.
