Mean time to resolution (MTTR), also called mean time to restore or recovery, tracks how long service takes to recover after an incident. Start the clock when the incident is detected and stop it when service is restored. A high MTTR can point to poor observability, difficult debugging, or an architecture in which failures cascade unpredictably. section 9.2.4 section 12.1.1
Why the average needs context
A team with a four-hour MTTR faces a different deployment risk from one that typically restores service in 15 minutes. But one overall average can mislead: many quickly resolved minor incidents may conceal slow recovery from critical failures. Track critical incidents separately. section 9.2.4
Use MTTR as a recovery signal, not a target to improve at any cost:
- Plot a rolling 30-day trend. A rising trend suggests that failures are becoming harder to diagnose or fix. section 9.4.1
- Pair it with incident frequency. Closing incidents early can lower MTTR on a dashboard while the same root causes keep producing failures. section 9.3
- Read it alongside deployment and other stability signals. Shipping faster is not an improvement if the running system cannot remain reliable. section 12.1.1
Table 9.3 Paired metrics as guardrails against gaming
|
Speed metric
|
Quality guardrail
|
Gaming looks like….
|
|
Deployment frequency
|
Change failure rate
|
More deploys, more incidents
|
|
Lead time
|
Activity ratio
|
Faster queue clearing, some waiting
|
|
MTTR
|
Incident frequency
|
Faster patching, same root causes
|
|
Cycle time
|
Rework rate
|
Faster merges, more post-merge fixes
|
|
Build time
|
Test coverage
|
Faster builds, fewer tests
|
The paired-metrics table places incident frequency beside MTTR so faster recovery does not mask recurring failures. section 9.3





