Skip to main content
Legia

Apps & Software

Four DORA Metrics for Measuring Release Speed and Stability

DORA metrics turn the question of how well our team releases into numbers that can be compared month to month. This guide explains the four, how to calculate them, and how to read them without being misled.

SEO and Content Specialist

6 min read

PHP code on a monitor screen

Short answer: the four DORA metrics measure software delivery performance: deployment frequency (how often releases reach production), change lead time (how long a change takes from commit to running in production), failed deployment recovery time (how long it takes to recover from a failed release), and change fail rate (what share of releases need immediate intervention). The first three describe speed, the last describes stability.

Many companies feel their application releases are slow or risky, but have no numbers to prove it, let alone to show that an improvement worked. DORA metrics answer that need. DORA (DevOps Research and Assessment) is a research program that has studied software delivery practices for over a decade, and its four metrics have become a common reference for assessing a release flow. This guide covers what each metric means, how to calculate it from data you already have, and the traps that make the numbers misleading. A basic explanation of the automated flow that affects these numbers is in What Is CI/CD and When Do Development Teams Need It.

The four DORA metrics and what they mean

The definitions below follow DORA’s official guide.

MetricDefinitionGroup
Deployment frequencyThe number of deployments over a given period, or the time between deploymentsThroughput (speed)
Change lead timeThe time for a change to go from committed to version control to deployed in productionThroughput
Failed deployment recovery timeThe time to recover from a failed deployment that requires immediate interventionThroughput
Change fail rateThe ratio of deployments that require immediate intervention after deploymentInstability

Two things need clarifying. First, in earlier versions the third metric was called mean time to recover or time to restore service. DORA renamed it failed deployment recovery time in 2023 and narrowed its scope: only recovery from failures caused by a change in production, not outside disruptions such as a data center outage. Second, DORA now has a fifth metric, deployment rework rate, the ratio of unplanned deployments that happen as a result of an incident in production. It was added in 2024 to test whether change fail rate is only a proxy for rework. Together with change fail rate, the fifth metric belongs to the instability group. To start, the first four are enough.

Illustration: web application

Where the data comes from

You do not need special tools to get the first numbers. Three sources usually already exist:

  1. Version control (Git): commit times and merge times.
  2. Deployment records: release pipeline logs, or handwritten records if releases are still manual.
  3. Incident records: tickets, chat channels, or team notes that mark problematic releases and when they recovered.

Before calculating, agree on two definitions and write them down, because they determine every number:

  • What counts as a deployment. A change that reaches users in production. Releases to a test environment do not count.
  • What counts as failed. Following DORA: a deployment that requires immediate intervention, such as a rollback or an emergency fix.

How to calculate each metric

MetricHow to calculateExample report form
Deployment frequencyCount production deployments in a month, or the average gap between deploymentsReleases per week
Change lead timeFor each change, measure the gap from commit to running in production, then take the middle value (median)Days or hours
Failed deployment recovery timeFor each failed release, measure the gap from when the failure was detected until the service recoveredHours
Change fail rateFailed releases divided by all releases in the same periodPercent

The median is used for lead time and recovery time because one extreme case does not shift a whole month’s figure. That is a practical choice, not a DORA rule, so state the method in the report so figures are comparable between months.

How to read the four together

A single number tells you nothing. A team that releases daily but sees a third of its releases fail is no better than a team that releases weekly with nearly all releases smooth. DORA advises using several metrics that have a healthy tension between them, and states that top-performing teams do well across all metrics, rather than trading speed for stability.

Common patterns and their practical meaning:

PatternLikely cause
Low frequency, low change fail rateReleases are rare but done carefully; likely large releases with queues of approvals
High frequency, high change fail rateToo little automated testing; fast releases without a safety net
Long lead time despite high frequencyChanges wait in review or approval queues
Long recovery timeNo quick rollback, late monitoring, or unclear responsible person

Mistakes that make the metrics misleading

DORA itself notes several traps:

  • Used at the wrong level. The metrics are designed for a single application or service. Combining very different applications, or comparing teams, produces numbers that are hard to interpret.
  • Turned into rigid targets. Broad targets such as “release several times a day” raise the likelihood that teams game the numbers (Goodhart’s law).
  • Turned into a competition between teams. The goal is improvement over time, not rankings.
  • Used to claim that a particular industry is different. DORA rejects regulatory compliance as a reason not to improve the release flow.
  • Built too elaborately at the start. Elaborate automated integrations before any improvement may not be worth it. DORA suggests starting with team conversations, the DORA Quick Check, or tools you already have.

Industry benchmark figures are deliberately not quoted here. What is more useful for your company is its own baseline and the direction of movement.

How to start in four steps

  1. Pick one application whose releases most often cause complaints.
  2. Write down the definitions of a deployment and of a failed release.
  3. Collect the past three months from Git, release records, and incident records. If the data does not exist, start recording from now.
  4. Review the numbers monthly with the team, then pick one blocker to fix, not all at once.

Work done in stages with routine releases, as discussed in The Agile Method in an Order-App Project, usually collects data faster because the releases are frequent.

How Legia does it

In the DevOps & CI/CD service, Legia measures four numbers before and after the work: release frequency, time from finished code to running on the server, recovery time after a problematic release, and release failure rate. The first step is mapping the current flow for 1 week, which produces a baseline report with reference figures. After that the most disruptive blockers are handled first, the release flow is built alongside the old way until proven stable, and your team is trained to release independently. The whole process takes about 9 to 15 weeks. Other services in this pillar are on the Applications & Software page.

Frequently asked questions

Are DORA metrics only for technology companies?

No. The metrics apply to any team that releases software, including internal applications. DORA states that this measurement can be used regardless of the technology stack.

What numbers count as good?

DORA’s official guide sets no thresholds. It only states that top performers do well across all metrics. Use your own baseline, then improve the direction.

Are special tools needed to measure them?

Not to start. Data from Git, release records, and incident records is enough for a monthly calculation. Automation is needed only once manual recording starts taking up time.

What about the fifth metric, deployment rework rate?

Add it once the first four are recorded steadily. It measures the ratio of unplanned releases caused by incidents in production, and complements change fail rate in assessing stability.

Sources

Category Apps & Software

Keep reading

Related articles

All Articles