DORA Metrics
Five software-delivery performance measures used together to examine change throughput and instability for an application or service.
Download Markdown (.md)Soroush’s Product Wiki · https://wiki.soroush.uk/topic/dora-metrics/
01What it means
DORA metrics are five measures of software delivery performance. Throughput includes change lead time, deployment frequency, and failed deployment recovery time. Instability includes change fail rate and deployment rework rate. This current model evolved from the widely known original Four Keys and is intended for improvement in context.[1][2][3]
- 01Change lead timeThroughput · change to production
- 02Deployment frequencyThroughput · production cadence
- 03Failed deployment recovery timeThroughput · restore after failure
- 04Change fail rateInstability · deployments causing failure
- 05Deployment rework rateInstability · unplanned remediation
The current DORA model contains five contextual improvement measures, not universal targets or individual productivity scores.
02Why it matters
The measures give cross-functional teams a common view of how safely and efficiently changes reach production. Examined together and over time, they can reveal constraints, test whether delivery improvements work, and counter the assumption that speed must come at the expense of stability. They describe outcomes, not the practices that caused them.
03When to use it
Use DORA metrics for a specific application or service with a team that can influence its delivery system. Establish consistent definitions and a baseline before changing tooling or process. They are most useful in improvement conversations and retrospectives, not as individual performance targets or simplistic rankings across unlike systems.
04How it works
- Choose one application or service and agree on deployment, failure, recovery, and rework definitions for its context.
- Collect the five measures from delivery and incident data, beginning with an approximate baseline if precise integration is expensive.
- Map the delivery flow and discuss where delays, risky batches, failed changes, and unplanned remediation originate.
- Select one meaningful constraint, make a bounded improvement, and add local leading indicators where useful.
- Review all five trends with qualitative evidence, then repeat without turning benchmarks into quotas.
05A practical example
A payments service deploys weekly and often needs emergency follow-up releases. The team measures all five metrics, maps commit-to-production flow, and finds large review batches plus manual rollback steps. It trials smaller changes and automated rollback for one component, then checks whether lead time, recovery time, and both failure measures improve together.
06Where it breaks
DORA metrics do not measure customer value, developer experience, code quality, or individual contribution. Aggregating unrelated services hides context, while targets invite gaming and unsafe shortcuts. Data definitions can differ across tools. The metrics should support diagnosis and learning alongside product, reliability, security, and human-centered measures.
Common mistakes
- Still describing DORA as only four metrics without noting the current five-metric model.
- Comparing teams or services with materially different contexts as a league table.
- Making one metric a target and ignoring balancing signals.
- Investing heavily in dashboard precision before starting an improvement conversation.
07Sources
- DORA's Software Delivery Performance MetricsDORA · Official measurement guideDORA's Software Delivery Performance Metrics
- A History of DORA's Software Delivery MetricsDORA · Official research historyA History of DORA's Software Delivery Metrics
- DORA ResearchDORA · Research program archiveDORA Research