Check whether quality is improving
The tiles at the top of the command center's Quality tab answer one question: is the work getting better or worse? Each tile sets a figure for the range you pick against the same length of time just before it. Use them for a monthly review, or after you change how a project's agents run, to see whether the change helped.
Who can do this: requires the org owner or org admin role. Only Owners and Admins see the command center.
-
In the sidebar, select Reports. The command center opens.
-
In the bar, pick a range long enough to hold real work, such as 30d. To check one project, pick it in Project. Every tile follows both.
-
Select the Quality tab. The tiles sit at the top.

-
Read the first six tiles:
- Tasks done: tasks finished in the range.
- Success rate: tasks done out of tasks done or failed.
- Zero-touch: the share of tasks done with no question, no comment and no other step by a person.
- Cycle median: the middle time from a task's creation to its finish.
- Cost per PR: AI spend for each merged pull request.
- Review score: the weekly AI review's average, from 1 to 5.
-
Under a figure, read the line that ends vs before, such as −2.6 pts vs before. It is the change against the period before. A tile with no such line had nothing to compare with.
-
Hover a tile. The hover gives the exact figure, Window before with the earlier figure (or no data), and how the figure is worked out.
-
If your plan includes agent performance, read the four tiles after Review score. They judge each finished task by what happened to its change afterwards:
- Outcome success: the share of judged work whose change survived.
- First-pass merges: merges that went in on the first try, with no failing check and no commit by a person.
- Defects per 100: merges later linked to a defect, for every 100 merges.
- Cost per success: AI spend for each piece of work whose change survived.
-
Hover one of those four. After the formula, n is how many results it counts and the 95% interval is the range the true figure most likely falls in, such as n=19, 95% interval 69–97. A small n or a wide interval means a change against the period before may be chance.
A figure with no data shows — with the reason under it, never 0. A small amber dot after a figure means it is partial: some work behind it could not be counted. Hover the tile to read what was left out.
What to do next
- A falling Review score: see Follow up on the lowest-scored work.
- A falling Outcome success or rising Defects per 100: find which setup is behind it with Compare agent setups side by side.
- A falling Success rate: list the failed work with Find what needs attention.
Related
- Read the command center: every tab and tile.
- How the command center works: where its figures come from.