# The 12 alerts I delete first on any team

Free lead magnet for **Pager Diet Kit** ($197).  
This list is the hook. The full kit expands each item with delete criteria, replacement signals, and a 14-day rollout plan.

---

## 1. Disk absolute thresholds with no trend
**Why it is noise:** Volumes can sit at 78% forever and still be fine.  
**Prefer:** fill-rate (time-to-full), inode pressure, and user-visible write failures.

## 2. CPU > 80% for a few minutes on autoscaled fleets
**Why it is noise:** Autoscaling is supposed to absorb this.  
**Prefer:** latency/error burn, saturation that blocks work, or sustained inability to scale.

## 3. Restart loops without crash context
**Why it is noise:** Restarts are symptoms. A clean rolling restart should not wake people.  
**Prefer:** crashloop with reason, OOM correlation, or deploy-linked regression.

## 4. Certificate expiry pages for non-user-facing intermediates
**Why it is noise:** Expiry is a calendar problem, not a 3am problem, if you track lead time.  
**Prefer:** tickets at 30/14/7 days; page only for short-runway user-facing or signing roots.

## 5. "Pod not ready" storms during deploys
**Why it is noise:** Every deploy creates readiness churn.  
**Prefer:** readiness debt after rollout settles; failed canaries; error budget burn.

## 6. Single-probe synthetic failures
**Why it is noise:** One region or one probe flakes.  
**Prefer:** multi-probe / multi-region quorum before paging.

## 7. Queue depth pages ignoring lag and SLA
**Why it is noise:** Depth alone does not say whether users are late.  
**Prefer:** consumer lag vs processing SLO, age of oldest message, drop rate.

## 8. Error-rate alerts without traffic baseline
**Why it is noise:** 2/2 errors is 100%. Low traffic makes rates meaningless.  
**Prefer:** absolute error floors + traffic-aware rates + burn rates.

## 9. Dependency 5xx that fire on expected partial loss
**Why it is noise:** Partial dependency loss is often designed for.  
**Prefer:** page when user-visible SLO burn accelerates beyond budget.

## 10. Log keyword pages ("ERROR")
**Why it is noise:** Logs are not SLIs. Keyword fires train people to ignore pages.  
**Prefer:** aggregate symptoms users feel; use logs for diagnosis after.

## 11. Heartbeat misses without confirmation
**Why it is noise:** Missed scrape ≠ service death.  
**Prefer:** multi-interval failure + secondary probe.

## 12. Capacity forecasts that page humans
**Why it is noise:** Forecasts belong in tickets and roadmaps.  
**Prefer:** page only on acute user impact or hard safety limits.

---

## Next step
Run this list against your current paging policies. Mark each alert: **delete / demote / keep**.

Full system (playbook, templates, 14-day plan): **Pager Diet Kit — $197**