DORA metrics: reading the four keys without fooling yourself

The DORA metrics dashboard grades your delivery on the four keys — lead time for changes, deploy frequency, change failure rate, and time to restore — against Elite/High/Medium/Low tiers. Here's how to read it, and what to do when a tile turns amber.

deckgauge · Board · Intelligence
Lead time (p50)
31h
High
Deploy freq
4.2/wk
Elite
Change-fail
9%
High
Restore time
3.5h
Medium

DORA tiles with tiers.

What it measures, and where the data comes from

Four tiles, each benchmarked to a tier. Three of them are measured from production deployment records when your repositories have them (GitHub deployments, Azure DevOps release pipelines) and fall back to a proxy when they don't. A proxied tile is marked (proxy) — hover it for the reason — so an estimate never passes for an observation.

  • Lead time for changes — median hours from a PR being opened to the production deploy that shipped it. Repositories with no production deploys are measured to merge instead, which leaves out any release queue; the tile then reads (proxy), or (partly proxied) when a board mixes both. Note the start point: DORA counts from the first commit, Deckgauge from PR open, because it does not link PRs to their earlier commits. Full definition.
  • Deploy frequency — successful production deploys per week, counting a multi-stage release once. Proxied from merged PRs per week when no production deploy can be identified.
  • Change failure rate — share of production releases that failed. Proxied from revert- and fix-style commit messages when no release carries a pass/fail outcome.
  • Time to restore — always a proxy: median hours from a bug being opened to closed, because no incident source is connected.
Read the proxies as trendsA tile marked (proxy) is inferred, not observed. Watch its direction over weeks — don't quote the absolute number in a board deck.

How to read it

Scan the tiers first, not the numbers. A wall of Elite/High is a healthy pipeline. The tile sitting a tier or two below the others is your constraint — and it's almost always a process gap, not an effort gap.

If lead time is worse than the rest, the bottleneck lives somewhere between PR open and production: slow reviews, oversized batches, a flaky pipeline, or a release queue. Fast coders with slow lead time is the most common pattern we see.

What it tells you over time

Give a tier at least a quarter before you trust it. A lead-time tile drifting from Elite to High to Medium over a quarter is a slow leak — usually reviews stretching out or PRs getting bigger. Catch it at High, not Low.

Example situations

1. Busy board, slow shipping

deckgauge · DORA
Lead time (p50)
6.1d
Medium
Deploy freq
5.0/wk
Elite
Change-fail
7%
High
Restore time
2.9h
High

Lead time lagging the other keys.

What you're seeing: three green tiles and one amber. Deploy frequency is Elite, so people are shipping — but lead time is Medium, so each change crawls from PR to production.

How to react: don't push people to "go faster." Pair this with the Review Pickup Time and PR Size widgets. If pickup is high, reviews are the jam; if PRs skew Large, batch size is.

Managerial playMake batch size the target, not review speed. Split the ticket into sub-400-line slices before anyone opens an editor — plan the work as a stack of mergeable PRs instead of carving up one giant diff under deadline pressure after the fact. Put a hard "one reviewer, one business day" expectation on the team: whoever's assigned either reviews it that day or hands it off, no PR just sits waiting. If the real jam turns out to be review latency rather than batch size, the Review Pickup Time guide has the SLA playbook for that — no need to run two versions of the same fix.

2. Change failure rate creeping up after a push

deckgauge · DORA
Lead time (p50)
22h
Elite
Deploy freq
6.4/wk
Elite
Change-fail
21%
Low
Restore time
7.2h
Medium

Change-fail proxy trending the wrong way.

What you're seeing: speed is Elite but the change-fail proxy jumped to Low right after a delivery crunch. Fast and fragile.

How to react: this is the classic "we hit the date by cutting corners" signature. Confirm it against Rework Rate and Bug Rate before acting — the proxy can spike on a burst of legitimate reverts.

Managerial playResist the urge to add a heavyweight process. Ask the team what they skipped under pressure — usually tests or a real review. Give them one slack sprint to pay it down, and stop rewarding date-hitting that ignores quality. Celebrate a boring, stable week out loud.

Frequently asked

What are the four DORA metrics?
Lead time for changes, deployment frequency, change failure rate, and time to restore service. The dashboard grades each against Elite, High, Medium and Low tiers.
Which DORA metrics does Deckgauge measure directly?
Lead time, deployment frequency and change failure rate are measured from real deployment records wherever your repositories have them; where they do not, each falls back to a proxy and its tile is labelled as one. Time to restore is always a proxy, because no incident source is connected. Lead time starts at PR open rather than first commit.
What should I do when a DORA tile turns amber?
Read the four together rather than acting on one. They are designed to be in tension: throughput improving while change failure rate climbs is a different problem from both moving the same way, and the combination tells you which.

Related widgets

Last updated