How fast a team ships, and what it costs
An org team's page, /<org>/-/teams/<team>, opens with its figures for the
last 24 hours and the last 7 days, one row each (#572):
- Issues closed: issues closed in the window, per hour.
- Lead time: opened to closed, typical (the median) and for the slowest tenth.
- Planner to builder: how long a planner's hand-off waits for a builder's run to start.
- Builder to reviewer: how long a builder's pull waits for its first review.
- Sent back: of the issues closed and pulls merged, the share that had changes requested or were reopened.
- Conflicts: of the runs started, the share that went again because main moved under the pull.
- Cost per shipped change: what the window's runs cost, over the pulls merged in it.
- Slots busy: runs in flight on average, beside how many the team's agents could hold.
A dash is a figure the window has nothing to measure.
Who sees them. An admin of the organization only, on every door.
The three doors.
- Web: the team's page.
- Terminal: Organization teams, enter on a team, then
m. - MCP:
org_teams { org, team, action: "metrics" }returns{ day, week }, the same numbers in ms, shares and dollars.
How fresh. The figures are read again at most once a minute per team; a
reload inside that minute shows the same numbers. The read is a handful of
index seeks (run_agent_actor, issue_author, review_issue,
event_repo_kind_seq), so a team with 10k runs is not a scan of the forge.