An interview question: how many metrics does DORA have? The answer “four” sounds confident and has been wrong since 2025. There are five, one of the originals has been renamed, the Elite–Low levels are gone, and DORA itself no longer stands for anything: the acronym was retired and the annual report renamed from Accelerate State of DevOps to State of AI-assisted Software Development.
None of this is cosmetic. Every change came from exactly where metrics break teams most often.
Five metrics instead of four
The fifth is deployment rework rate: the share of deployments nobody planned, shipped because of a production incident. It catches what the original four missed. A team with a decent release cadence and a low change fail rate can spend half its deployments cleaning up after the previous ones — and by the old four metrics that team looks healthy.
MTTR is gone as a name. In its place is failed deployment recovery time: recovery from a deployment that required immediate intervention. The wording is narrower than before — the metric speaks about a deployment, not about any incident.
Then comes the detail almost every retelling gets wrong. The five metrics split into two groups, and the split is counterintuitive:
| Group | Metrics |
|---|---|
| Throughput | change lead time, deployment frequency, failed deployment recovery time |
| Instability | change fail rate, deployment rework rate |
Recovery time sits in throughput, not in stability. The logic is simple: shipping a fix also means moving a change through the system, and its speed is set by the same pipeline. Instability keeps two metrics about how badly the deployments themselves go. Cutting recovery time is helped less by a separate “response process” than by whatever speeds up an ordinary release, plus the ability to find the cause fast — there is a separate write-up on the tooling for that search.
The levels are gone
Elite, High, Medium and Low are retired. The 2025 report — a cluster analysis of answers from nearly five thousand people — produced seven team profiles in which performance, stability and how people feel are interwoven. At one end sit the “harmonious high-achievers”, strong on delivery and on well-being alike. At the other are teams with “foundational challenges” and those locked in a “legacy bottleneck”, where unstable systems eat away at morale.
The relabelling is not the point. Level thresholds were recomputed from scratch every year: “Elite” in 2019 and “Elite” in 2023 are different numbers. A team reporting “we are High” year after year was comparing itself against a moving target. Now there is no target, and there is a profile that includes burnout.
Where metrics break a team
DORA names two ways to ruin a measurement outright, and both live in the typical platform dashboard.
The first is turning a metric into a goal. Goodhart’s law never misfires: the task “raise deployment frequency” is cured by splitting one release into five empty ones. The number grows, delivery stays where it was. The healthy move runs the other way — lower the cost of a release so the frequency rises on its own: canary rollouts and atomic deploys move both frequency and change fail rate, because they change the mechanics rather than the reporting.
The second is comparing what cannot be compared. DORA recommends applying the metrics to a single application or service and warns against comparisons between systems of a different nature. Folding several teams’ metrics into one league table is exactly what a dashboard usually gets built for.
Hence the practical rule: the five metrics measure the delivery system, not people. A single team’s trend over time is a valid use. A leaderboard across teams is not.
Where SPACE fits
The SPACE framework (Forsgren, Storey, Maddila, Zimmermann, Houck, Butler; ACM Queue, February 2021) answers a different question: not “how fast does the system move” but “how do people experience it”. Five dimensions — Satisfaction, Performance, Activity, Communication, Efficiency & flow. The authors’ central claim: productivity does not reduce to a single metric, several dimensions are needed at once.
The most dangerous dimension is Activity. Commits, pull requests and deployments are the easiest things to count, which is why they reach management reports first. Activity is context, not a goal: “many commits” does not equal “much value”, and nudging that number is trivial.
Until 2025 the two models divided the work: DORA measured delivery outcomes, SPACE measured how it felt. Team profiles erased that border. Well-being now sits inside the same model that measures delivery.
What to do with the numbers in your slides
“Elite teams deploy 208 times more often and have a 106 times shorter lead time” — the figures are real, from Accelerate State of DevOps 2019. They are seven years old, and the clusters they describe have been abolished. A number like that on a 2026 slide describes the year the slide was assembled, not the industry.
The check takes five minutes. If a metric in your dashboard is called MTTR, if there are four metrics, and if teams are laid out on an Elite–Low scale, you are measuring by an edition the authors have already replaced. Renaming things is cheap. Unlearning the habit of comparing teams against each other costs more — and that table is what bred it.