Mean Time to Recovery (MTTR) measures the average time to restore normal operation after a production incident. It's the operational-resilience metric in the DORA set.
What it measures
Leanmote has no incident record to read, so it infers recovery from pull requests. A merged PR whose title contains hotfix, revert or HOT-FIX is treated as a recovery; the clock starts at the previous merged PR in that same repository that is not itself a recovery. MTTR is the average of those intervals.
How Leanmote calculates it
mttr = avg(recovery_pr.merged_at - previous_pr_in_repo.merged_at)
Recovery is recognised by title only, and the list is short:
hotfix,revert,HOT-FIX. A PR titled "fix: null pointer" is counted as a failure by Change Failure Rate but is not a recovery here — the two metrics use different word lists, so their counts don't line up.Matching is on any part of the title, so a PR about "reverting the copy change" also counts.
Reported as an average, in hours. There is no median, no percentile selector and no severity labels.
The interval is measured per repository, and consecutive recoveries are skipped over: three hotfixes in a row all measure back to the same original pull request, so the second and third read progressively longer.
A recovery with no earlier non-recovery merge in its repository is dropped entirely rather than counted as zero.
How to interpret it
Trending down means recovery is getting faster — operationally healthy.
Because it's an average with no outlier handling, one pathological interval moves the whole number. Open the drill-down and look at the longest rows before reacting to a jump.
The keyword list is the real limit. A recovery merged under a title that doesn't contain one of those three words is invisible here, and a routine pull request that happens to mention reverting is counted as an incident.
What to do about it
Improve observability — alerts that detect incidents earlier shorten the clock.
Clarify ownership and escalation paths so the right people are paged immediately.
Invest in runbooks for the most common failure modes.
Review post-incident: was the recovery time spent diagnosing or fixing? That reveals whether the bottleneck is observability or change-velocity.
Related metrics
Change Failure Rate
Deployment Frequency
DORA metrics overview