What is change failure rate?

Plain DORA definition, what counts as a failure, and where the term differs from defect-escape rate and incident rate.

Oliver Wakefield-Smith

Founder, Digital Signet. Built siemcost / monitoringcost / itbudgetcalculator.

Updated 21 Jun 2026

Direct answer

Change failure rate (CFR) is the percentage of deployments to production that cause a failure, where a failure is defined as service degradation, rollback, hotfix, or unplanned incident attributable to the change DORA Four Keys2024. It is one of the four DORA keys and is paired with mean time to restore as the stability half of the framework.

The DORA definition, in full

DORA frames CFR as the share of changes that result in degraded service and require remediation (rollback, fix-forward, patch, or any user-visible recovery work) DORA Four Keys2024. The denominator is total deployments to production; the numerator is deployments that triggered a failure during or shortly after rollout. DORA explicitly excludes failures with no end-user impact when teams have the instrumentation to detect that distinction.

What counts as a failure

The most common inclusions: user-visible degradation, page or endpoint downtime, an SLO breach (latency, error rate, availability) traceable to a deploy, a rollback, a forward-fix that exists only to patch the previous deploy, an incident with severity above an internal threshold, or a customer-reported regression confirmed against a release.

The most common exclusions: feature-flag changes with no code roll, dark launches, internal-only tools deploys, externally caused incidents (third-party outage, upstream API breakage), and any rollback executed within an automated canary window where no traffic was served. DORA leaves these calls to the team but documents the standard convention Google Cloud CFR Guide2024.

CFR vs defect-escape rate vs incident rate

Defect-escape rate is the share of bugs that reach production from a release branch, regardless of whether they caused a customer-facing failure. Incident rate is the count of incidents over time, decoupled from deploy count. CFR is the joining variable: failures per deploy, with deploy frequency held constant. Two teams with identical incident counts can have very different CFR if their deploy cadences differ by an order of magnitude.

Why DORA shifted the framing in 2024

The 2024 DORA report sharpened the user-impact framing and stopped publishing the Elite / High / Medium / Low band labels in favour of percentile distributions DORA 20242024. The shift was driven by survey-data clustering that no longer matched the original tier boundaries; high-frequency deploy teams were consistently posting CFR in the 5-15% range that the old "Medium" tier would have penalised despite throughput-and-stability dominance.