Faculty guide

Create an assessment

Create a Provenance assessment from your course in the LMS. Attach the assignment prompt, the rubric or criteria you want the review to connect to, and any guidance you'd give a colleague running the review for you — emphases, topics to avoid, what a strong explanation sounds like.

Everything downstream is grounded in what you attach here: questions are drafted against your assignment, your rubric, and each student's actual submission — not from a generic question bank.

Review and approve the questions

Once students submit, Provenance drafts two kinds of questions for your review:

• Common questions asked of every student, drawn from the assignment and rubric.
• Artifact-specific questions grounded in passages of the individual student's submission — each one shows you the passage it came from and the rubric criterion it connects to.

You can approve, edit, replace, or discard any question. No question reaches a student until you have approved it. This is the single highest-leverage step in the process: a few minutes of question review is what makes the evidence worth having.

The review session

Students complete a guided oral review: a structured, recorded conversation in which they walk through their own artifact and answer the approved questions. The session captures audio, a transcript, timestamps, and references to the artifact passages under discussion. You set the window in which students can complete their review, and time budgets keep sessions consistent across the roster.

Evaluate and validate

After each review, you receive an evidence package:

• The full transcript and audio, linked to the artifact passages each exchange refers to.
• The questions asked, with their rubric mapping.
• AI-assisted evaluation outputs: multiple independent evaluators compare each answer to your criteria, and a synthesis highlights where they agree, where they disagree, and what evidence each cited.
• Delivery annotations — observational notes only, such as pauses or self-corrections. They never interpret intent or state of mind.

You review the evidence, adjust or override any evaluation output, and validate the final result. Validated results return to your LMS gradebook. The AI outputs are inputs to your judgment, not a verdict — Provenance does not make misconduct determinations, and a weak review is not itself a finding about why it was weak.

Good practice

• Write rubric criteria as things a student could demonstrate aloud — explanation, justification, connection — rather than surface features of the artifact.
• Use the guidance field. A sentence like "press on the design trade-offs in section 2" measurably improves question drafts.
• Tell students what to expect and point them at the student guide. Reviews are better evidence when students aren't surprised by the format.
• Where an answer is weak, treat the record as the start of a conversation with the student, not the end of one.