# Would you sign this AI-written memo?

A six-minute teaching case by Frank Riemer · FrankX

You may copy and adapt this case for your class. All records are fictional. No accounts or software required.

## Give the class this memo

“75% of attendees want us to run the lab again.”

## Source packet

- REG · registrations.csv · 24 registered. Distinct names on the registration list.

- ATT · check-in.csv · 18 attended. Registered participants who checked in.

- EXIT · exit-slips.csv · 12 returned a slip. 9 answered yes and 3 answered no to “Run it again?” One slip per participant.

## Run the room

0–2 min · Show the memo and all three records. Ask each person to publish, revise or request the source. Ask for the numerator and denominator.

2–4 min · Remove EXIT. Read the identical memo aloud. Ask what can still be signed from the two remaining records.

4–6 min · In pairs, write one defensible sentence for each run. Exchange answers and ask a partner to reproduce both judgments.

## Facilitator key

Run 1 · REVISE. 9 of 12 respondents asked for another lab. Six attendees left no slip. Their preferences are unknown. A second 75% appears in the register: 18 of 24 registrants attended. That answers a different question.

Run 2 · REQUEST. 18 of 24 registrants attended. Demand for another lab is unknown. Registration and check-in can establish attendance. To publish a claim about demand, recover the exit slips or collect new responses and identify them as new evidence.

## Mark the reasoning (one point each)

- Names the population and denominator for 9/12.

- Distinguishes six missing responses from six negative responses.

- Withholds the demand claim when EXIT is removed.

- Preserves the independently supported attendance finding using REG and ATT.

## Transfer to an AI system

Keep the prompt and memo fixed. Change only the available source IDs. Log the claim, source IDs, reviewer action and accepted wording. A model that repeats its first answer after EXIT disappears fails the second case.

Classroom link: https://www.frankx.ai/partnerships/universities/source-check

University programmes: https://www.frankx.ai/partnerships/universities
