Ninety days after go-live, review the setup with the customer. Read the measures below, compare them with the goals of the discovery document, and agree on what to change.
The measures
CoRoid does not show these measures together on one page yet. Read each one where the table says, and write it down for the review.
| Measure | Where to read it |
|---|---|
| Days from kickoff to the first pull request merged by an agent | The go-live checklist, where you recorded it |
| Share of agent slots in use, and tasks delivered per week | Slots: Active now and Queued in Settings → Billing & usage → Capacity & credits. They show the moment only, so read them at set times each week. Tasks: the factory's Work finished figure, with 4 weeks chosen |
| Quality Control passes on the first try | Factory → Run log, with the type Quality check chosen. Count the changes whose first run passed |
| Workflow cases closed without a handover | Each workflow's Statistics, for 3 months: Closed by the workflow |
| Support cases caused by a configuration mistake | The customer's own support records. Count the cases whose fix was a change of a setting |
See The overview for the factory's figures.
Find the cause
A weak measure points to a setting more often than to the product. Start with these places.
| What you see | Where to look |
|---|---|
| Few pull requests by agents, or a slow first one | The requests: are they clear enough to build? See Specification-driven development. Also how long work waits in Needs you |
| Many slots stay empty | Too few requests reach CoRoid. Check the request sources, mandates and automations. Or the plan has more slots than the work needs |
| Slots are full and tasks wait long | The work needs more agent slots. See Agent slots and capacity |
| Few changes pass Quality Control on the first try | AGENTS.md, the project context, and the build and test commands. Also checks that are set wrong |
| Few workflow cases close without a person | Paths and rules table rows that are missing. In Statistics, a step whose yeses were Approved as written can need fewer yeses |
| Support cases come from settings | The setting, and the part of the handover note that explains it |
Agree on what to change
- Go through the measures with the customer, against the goals of the discovery document.
- For each weak result, name the cause and the change.
- Give each change an owner and a date.
- Update the discovery document and the handover note with each change.
A need that no setting covers is a request to CoRoid. See A need that nothing covers.
Done when
The customer agrees with the list of changes, and each change has an owner and a date.