ClassLattice guide

Teacher-reviewed AI workflows: an evaluation checklist

“Human in the loop” is meaningful only when the educator can see the source context, inspect the proposed result, change it, decline it, and understand what becomes part of the instructional record.

01

Check the input boundary

Confirm which course, standard, source, artifact revision, and learner records the workflow can access. Look for clear limits on retrieved context, files, and external connections.

02

Check the review boundary

The proposed output should be labeled as a draft, preserve citations or source references where relevant, and remain editable. High-impact decisions should never be silently automated.

03

Check the operational boundary

Ask what happens during an outage, timeout, duplicate submission, or concurrent edit. The product should fail clearly and preserve the educator’s canonical work so the task can be resumed.