Follow up on the lowest-scored work
Each week Jaah's AI review reads a sample of the work your organization delivered and scores each task from 1 (poor) to 5 (excellent), with a one-line reason. Do this when you want to check what went wrong in the work it scored lowest, and decide whether to fix it, change how an agent runs, or let it go.
Who can do this: Requires the org admin role. Only Owners and Admins see the command center.
- In the sidebar, select Reports. The command center opens.
- In the bar, set the range to
7dor30d. The review runs once a week, so the default24hoften holds no reviews. To look at one project only, pick it in Project. - Select the Quality tab and scroll past the tiles. If it shows only No reviews yet, with the date of the first run, no review has run yet and there is nothing to follow up.
- Read Review score: the score trend, one line per agent. An agent whose line drops is the one to look at first.
- Read Score spread: how many reviews landed on each score, from 1 to 5. Hover the reviews/week figure beside it for the weekly sample size and the next run. If it is red, the last run failed, and the hover says why.
- Read Lowest scored: up to ten tasks, each with its score and the reviewer's reason under the title. A score of 1 or 2 is red, and a score of 3 is amber.
- Select a task's title to open its page, and check its result against the reviewer's reason.
- If the work needs fixing, start a new task that names what to fix. See Start a task. If the same agent keeps scoring low, compare it with other setups before you change anything: Compare agents and models by outcome.
The range and project stay in the page's address, so a copied link opens the same view for another Owner or Admin.
Related
- Read the command center: every tab, the Quality tab's tiles included.
- Move an agent to its best-performing setup: switch a project's default model when another setup does better.
- How the command center works: where its figures come from.