When QA passes a piece of work with screenshots, Council asks a panel of models to score its interface. Each member gives a score out of 10 and names what it would fix. The lowest score decides. Below the target, the work goes back to the Developer with those findings as fixes it must make, QA checks it again, and the panel scores it again.
Council is off until you turn it on. It appears in Quality Control once it is available for your organization.
Turning it on
Open Quality Control, then Council in the QA group.
- On the organization default, it applies to every project that follows the default.
- A project can have its own setting. Use the default brings the organization's back.
Settings
| Setting | What it means | Default |
|---|---|---|
| Target | The lowest score any member may give for the work to finish without a repair: 6 to 9 of 10. | 8 of 10 |
| Rounds | How many times the panel scores the work, the first time included. | 3 |
CoRoid's panel uses models from different families, so that one model's blind spots do not decide alone.
A round
- What the members see. The screenshots QA published for the work, each with its page, state and screen size. They do not see the code.
- What each member gives. A score out of 10, and findings: what is wrong, where it is, and which screenshot shows it.
- The result. The target is met, or the work is below it. When a member could not score the work, for example because its model was unavailable, the round is recorded as not assessed and the work is not held back.
From the second round on, the members also see the screenshots of the round before, so they can tell whether a repair made things better.
When the rounds end
The rounds end when the first of these happens:
- the lowest score reaches the target;
- every round is used;
- no member names a finding to fix;
- a repair fixes none of the findings, when QA checks them again;
- there is no room left for another repair within the work's limits.
If a repair makes the score drop, CoRoid goes back to the version the members scored higher, once, and the rounds end there.
What you see
- The task's thread. Each round has a line. It opens the members' scores, their findings and the screenshots they saw, beside the earlier round's screenshots.
- Quality. The task's issues list Council's findings, open or fixed.
- The pull request. A Council section says the result, the score of each round, and what was fixed and what stayed open.
- The merge plan. A merge grant never merges work that ended below the target. A person decides.
Cost
Each round asks every member to score the work, with the screenshots, so Council adds model use to each piece of work it scores. More rounds mean more model use only when the work stays below the target.
Settings export and import
Council's settings travel with the Delivery policy when you export and import your organization's settings. They apply where the destination has Council.