DORA metrics are the best framework we have for measuring engineering effectiveness. They're research-backed, platform-agnostic, and they correlate with business outcomes. So of course, everyone is now optimising them, and of course, they're starting to lie.
The problem isn't DORA. The problem is that DORA is a measure of system capability, not team health. A team can hit all the DORA targets and still be burning out. A team can deploy once a week with 2% failure rate and still feel broken. DORA tells you the system works. It doesn't tell you if the people running the system are okay.
DORA measures capability, not sustainability
Deployment frequency — how often can you ship? Lead time — how fast from commit to production? Change failure rate — how often do those deployments break things? Recovery time — how fast can you fix it?
These are all system metrics. They measure the infrastructure. They measure the process. They measure what the system is capable of doing. And they're important. A system that can't deploy safely, quickly, and frequently is broken.
But here's what they don't measure: Can the team sustain this? Are they burned out? Are the best people leaving? Is the rework piling up? Is someone oncall at 3am every week? Is the cognitive load unsustainable?
A team can ship eight times a day with 1% failure rate and still be dysfunctional. They might be running on fumes. They might have just lost three senior engineers. They might be paying the piper for six months of technical debt. The DORA metrics look perfect. The team is about to collapse.
The human layer: what DORA doesn't measure
Nicole Forsgren, who literally invented DORA, has been clear on this point in recent research: teams need throughput AND stability AND satisfaction. Not just in principle, but in practice. If you have one without the others, you have a problem.
"Deployment frequency is important. But a team deploying fast while burning out is not winning. Satisfaction — whether people want to stay and believe in what they're doing — is as predictive of performance as the technical metrics. You need both."
Nicole Forsgren · 2025 State of DevOps Report
Satisfaction is the layer DORA doesn't measure. It includes things like: Do people feel like their work matters? Are they learning? Is the workload sustainable? Do they have autonomy? Are they respected? Do they believe in the direction?
These are softer signals. They're harder to measure than "deployments per day." But they're more predictive of whether your team will still exist in six months and whether they'll be coherent when they do.
How DORA gaming happens and why it's worse than before
The moment DORA became the goal, teams learned to optimise for it. They started deploying smaller changes more frequently to improve deployment frequency (while shipping less value). They automated everything to improve lead time (while creating technical debt in the automation itself). They made changes behind feature flags to reduce change failure rate (while increasing complexity).
None of this is dishonest. The teams aren't cheating. They're just rational. They're optimising for what the system measures. And they're right to do it — if DORA is what leadership cares about, then making DORA work is how they survive.
The cost is hidden. It shows up six months later when the best people have left, when the codebase is unmaintainable, when every "small change" takes three people and two weeks to implement. But by then, the focus has shifted to the next metric.
The three-layer framework that actually works
Layer 1: Capability (DORA) — Can the system deploy safely and frequently? This is your floor. If you can't do this, nothing else matters. But this isn't your ceiling. Having DORA doesn't mean you're winning.
Layer 2: Predictability — Can the team forecast accurately? Do they finish what they commit to? This is where teams start to feel in control. They're not just moving fast, they're moving predictably. Plans actually mean something.
Layer 3: Sustainability — Do people want to stay? Is the work sustainable? Are they learning? This is the layer that determines whether you can actually keep the best people and maintain momentum over years, not just quarters.
DORA gets you layer 1. But layer 2 and 3 are what make layer 1 sustainable. And they require different measurements. Predictability requires tracking sprint commitment completion, impediment resolution rate, and rework volume. Sustainability requires regular team sentiment checks, engagement scores, and willingness-to-recommend metrics.
What to measure if you want DORA plus the human layer
Keep DORA. It's valuable. But add:
Predictability signals: Sprint completion rate (did we finish what we said?), cycle time (how long is work actually taking?), impediment resolution rate (are blockers getting cleared quickly?), and rework percentage (how much of our effort is fixing previous work?).
Satisfaction signals: Engagement in retros (are people showing up and speaking up?), action completion (do we actually fix things we identified as problems?), attrition risk (are senior people quiet or looking around?), and learning investment (are people getting time to grow?).
These aren't soft metrics. They're hard signals of team health. And they're predictive. A team with high DORA but dropping predictability and declining satisfaction isn't winning — they're about to lose velocity and people.
The trade-off you're actually making
When you obsess over DORA without the other layers, you're making an implicit trade-off: short-term capability for long-term sustainability. You're saying "move as fast as possible right now," and the team responds by burning the furniture to heat the house. It works for a while. Then you're cold and out of furniture.
The high-performing teams measure all three layers. They understand that DORA matters for executive confidence, but predictability matters for team confidence, and sustainability matters for keeping the people who built the capability in the first place.
DORA won't save you. But DORA plus predictability plus sustainability — that combination actually works.