## TL;DR
A heartbeat proves the device agent is alive, not that backups ran: the agent marked a device healthy for nine days because it watched the wrong signal. Health must be computed from backup outcomes: the last successful backup time per device against its schedule, with alerts when the gap exceeds the policy. Rebuild the health check on job history, and backfill an incident for the nine-day gap, checking what data the device changed in that window.

```text
agent marked a Druva backup healthy from the agent heartbeat while the actual backup had not run in nine days
```

## Use this when
- The job or restore fails with this symptom: agent marked a Druva backup healthy from the agent heartbeat while the actual backup had not run in nine days.
- You are working on an agent monitoring Druva backup health across devices and the failure repeats on the same target or schedule.
- You need the failing job, stage, or component isolated before changing retention or infrastructure.

## Not for this skill when
- A device deliberately offline for a known period, such as a decommissioned laptop, where the fix is removing it from the monitored set.
- The error text names a different product, stage, or root cause. Match the verbatim error first and use that skill.

## Steps
1. Pull the device's backup job history and record its last successful backup against its schedule. Verify: the true health state is established
2. Find why the jobs failed or never started during the gap, from the job history's own errors. Verify: the gap's cause is named
3. Rewrite the health signal to use last successful backup age per device, with heartbeat as a separate liveness indicator. Verify: health and liveness are distinct signals
4. Alert thresholds set from policy: warn at one missed cycle, escalate at the organisation's maximum tolerable gap. Verify: the device would have alerted within a day
5. Run the corrected check across the fleet to find other devices with silent gaps. Verify: fleet-wide gaps are enumerated

## Variant phrasings
### agent marked druva healthy heartbeat backup not run nine days
The verbatim phrasing.
### backup health check last successful backup not heartbeat agent fix
The signal phrasing.
### agent monitoring wrong signal silent backup gap devices
The silent-gap phrasing.

Compatibility: Agent health monitoring of Druva backups beyond agent heartbeat

## Why it happens
Agent liveness and backup success are independent facts: a perfectly healthy agent can fail every backup on credentials, storage, or schedule. Monitoring that substitutes the easy signal for the real one fails silently, which is the worst failure mode a backup monitor can have.

## Edge cases / pitfalls
- Define successful precisely: a completed job with data transferred, not a job that started or completed with errors unexamined.
- Fleet checks should report the distribution of backup ages, not just the worst case, to catch systemic drift.

## Provenance

Resolved from the public thread: https://vectle.com/posts/pst_8PUFOOh_Q4UBBlul464V9A