Metrics glossary

What every metric shares

The window. The metrics pages share a window picked in the toolbar: a number of days, or this or last month, quarter or year. Changes are against the period before: as many days before, or the same stretch of the previous month, quarter or year (the whole one before, for a whole one). A change only shows once the history reaches back that far.
Medians and p75, not averages. Half of PRs were quicker than the median; three in four were quicker than the p75. One very slow PR doesn’t drag them about as it would an average.
Left out everywhere: repos and people unticked in Org settings › Repositories and People, PRs and issues you’ve hidden, and the repos outside the window’s project. Draft PRs are left out of the workload when Exclude draft PRs is on (Settings › General).
Bots: PRs by GitHub’s bot accounts are always left out of the PR metrics. PRs by ordinary accounts whose logins end -bot or [bot] are left out too, unless re-included in the org’s settings. Bot reviews never count as a first review.

Pull requests (Dashboard and PR flow)

Metric	How it’s worked out	Left out
Cycle time	First commit (or the PR’s creation, if earlier) to merge.	Unmerged PRs.
Time to first review (TTFR)	Ready for review (or creation) to the first review by someone who isn’t the author or a bot.	PRs merged with no review.
Coding	First commit to ready for review.
Waiting for review	Ready for review to the first review.
Rework	First review to the last approval before merge.	Stages most PRs skip show the share of PRs that had one and the median when it happened.
Merging	Last approval to merge.
PRs merged (throughput)	PRs merged in the window.
PRs opened	A count from GitHub’s search.	It ignores the org’s exclusions.
Merged without review	PRs merged with no review of any kind, as a share of merged PRs.	Repos marked No Review in Org settings › Repositories.
Size and Files	Each person’s median lines changed (added plus removed) and files changed per PR merged in the window. A large PR is over 400 lines.
Rushed large PRs	Large PRs approved within 15 minutes with nothing asked, or merged unreviewed.

Reviewers

Requests answered and time to review come from review requests: each is matched to that reviewer’s first review between the request and merge. The clock starts at the request, or at ready for review if that’s later. Requests withdrawn before an answer are dropped.
Waiting now is review requests still open on open PRs.
The team filter applies to the reviewer, not the PR’s author.

The workload (Dashboard’s Team, Inbox, People)

In flight for a person is their open PRs, the reviews requested of them, and assigned issues with an open PR. Assigned issues nobody has started are backlog and not counted. An issue closed by one of their own PRs counts once.
Stale PRs are open ones not updated for 7 days or more.

Needs attention (Dashboard)

Items ranked into Act now, Today and Keep an eye on: CI red on a default branch, scorecard goals off target, review requests waiting over a day or on someone off today, approved PRs not merged, changes asked for and not answered, stale PRs, committed issues overdue, and people carrying far more than usual.

Issues (Issue flow)

The issue workflow in Org settings › Issues picks the board and which of its statuses count as in progress.

Metric	How it’s worked out
Cycle time	Time spent in the in-progress statuses. Moving to any other status (Done, Backlog) or closing stops the clock; coming back starts it again. With the fallback on, an issue never moved on the board uses its first linked PR.
Lead time	Created to closed as completed.
Flow efficiency	The share of in-progress days with activity on a linked PR.
Scope creep	The share of sub-issues added after work started.

Issues closed as not planned are left out of cycle and lead time.

Scorecards

Each goal has a cadence (weekly, monthly, quarterly, annual), a source and a target (at least or at most).

A goal is judged on its last whole period. The period under way is shaded and judged on its share so far.
A count’s target is held to the column’s length, rounded down: 20 a week is 2 a day when viewed by day.
Hit is how many whole periods were on target.
Hand-entered numbers are judged only in their own periods, and added up (unjudged) in longer ones, or averaged for percentages.
From other sources: reviews waiting over a day (requested in the period and answered after a day, never, or still pending on an open PR), open PRs at the period’s end, issue cycle time (as above), and flaky CI runs (below).

CI (GitHub Actions)

Metric	How it’s worked out	Left out
Success and failure rates	Failure includes timed out and startup failure.	Cancelled and skipped runs. Only a run’s latest attempt is listed.
Duration	The latest attempt’s start to its last update; p50, p75 and p90.
Run time	Wall-clock time of runs, by repo, workflow or job.
Re-runs	Runs with more than one attempt.
Flaky	A run that passed only on a re-run, or a commit with both a failed and a passing run.
Default branch health	Runs on the default branch outside PRs: red now and since when, and time back to green.
Needs attention	Red default branches, often-failing ones, flaky workflows, and workflows with a median of a minute or more that ran 25% slower in the window’s second half.	Failures on PRs alone: that’s CI doing its job.

Runs older than 190 days are dropped, and excluded repos aren’t fetched.