The pilot puts the setup to real use, on a small scale. Choose one project and one workflow, and run them for two weeks. Real requests show what the discovery document missed.
When Workflows is not on for the organization, pilot the project only.
Choose the pilot
- The project. Choose a repository with a test suite that the customer trusts. CoRoid checks its work with the customer's tests, so good tests give a clear result.
- The workflow. Choose a process with many similar cases, a clear owner and rules that the customer can write down.
- The people. Name who asks for work, who reviews and merges, and who answers what waits for a person.
Agree with the customer what a good result is after two weeks. Use the goals of the discovery document.
What to watch
Look at these every day in the first week, then twice a week.
In the factory. Open the factory overview.
- Work finished, and how much of it finished without help. Work that needs help often points to a missing setting or a vague request.
- Pull requests merged, and how many CoRoid's work opened.
- Time to deliver, and how much of it the work waited for a person.
- Needs you. Decisions that wait a long time slow everything else down.
- A task that is stuck or failed. See When a task gets stuck or fails.
In Quality Control. A check that fails on most changes is often set wrong. A check that cannot run in a project says so on its row.
In capacity and spend. Settings → Billing & usage → Capacity & credits shows how many agents are active and how much work is queued. Usage shows what the work costs. See Agent slots and capacity.
In the workflow. Open the workflow in Workflows.
- Cases. The Needs you filter shows the cases that wait for a person.
- Statistics. How cases end, what waited for a yes, how fast the first reply went out, and the agent time per case.
- New situation. This notice says that cases came to a person because no path covers them. Add a path for this adds the missing path.
Fix the settings, not the work
When something goes wrong, find the setting behind it. Change the setting, and write the change and its reason in your notes. Do not fix the same problem by hand each time.
- A task did the wrong thing: check the project context,
AGENTS.mdand the build and test commands. - A check failed for no good reason: check its setup in Quality Control.
- A case went to a person for no good reason: add a path or a rules table row, then publish a new revision.
Meet each week
Meet with the customer once a week during the pilot. Go through what you watched and what you changed. Write new decisions into the discovery document.
Done when
The two weeks end, and the customer decides one of these:
- Go live, with the settings as they are now.
- Run the pilot for longer, with a list of what to change first.