A Brown University economics professor's decision to move his course's final exam from a take-home format to an in-person, paper-based format produced a genuinely striking data point worth examining closely: the class average dropped from 96 percent on the take-home midterm to 48.6 percent on the in-person final, a 47-point swing. Perhaps even more telling, 22 of the 27 students who didn't show up for the in-person exam had scored a perfect 100 on the earlier take-home midterm, a pattern that's genuinely hard to explain as anything other than a direct, measurable signal about how the assessment format itself shaped the results.

This kind of direct, before-and-after comparison within the same course and the same student population offers considerably more concrete, compelling evidence about the real scale of AI-assisted misuse on unmonitored take-home assessments than survey data or general concern alone can provide. It's worth taking seriously as a genuine, data-grounded signal, not just anecdotal worry, about how much unmonitored take-home format specifically may be masking actual student understanding in some courses.

At the same time, this finding is specific to exam-style assessment under timed, unsupervised conditions, and it's worth being careful about how directly it generalizes to other assessment formats, particularly extended writing assignments developed over multiple drafts and checkpoints, which carry a genuinely different risk profile than a single, unmonitored take-home exam.

Why this particular data point is so compelling

What makes this finding especially notable is the built-in comparison: the same students, the same course content, and the same underlying material, evaluated under two genuinely different conditions. This design controls for many of the confounding factors that make it hard to draw firm conclusions from broader, less controlled comparisons, giving this specific case real, concrete weight as evidence about the actual scale of the gap between take-home and in-person performance in at least this particular course.

• Consider building in-person, monitored assessment checkpoints alongside take-home work, especially for high-stakes exams, given this kind of concrete evidence
• Recognize that this finding is specific to exam-style, unmonitored take-home conditions, and be careful about over-generalizing it to genuinely different assessment formats like drafted writing assignments
• Use in-person or monitored formats specifically to verify genuine understanding when a take-home assessment's stakes are high
• Watch for more instructors sharing similar before-and-after comparisons as this kind of direct, concrete data becomes more available
• Treat this as one compelling data point within a specific course context, not a universal, generalized conclusion about every take-home assessment

What this means for essay and extended writing assessment specifically

Extended writing assignments, developed over multiple drafts, checkpoints, and often tied to specific class discussion or source material, carry a genuinely different risk profile than a single, unmonitored take-home exam, since the multi-stage, process-documented nature of strong writing assignment design already provides real, ongoing evidence of a student's genuine engagement that a single exam moment doesn't capture. This finding is a strong argument for in-person or monitored formats specifically for exam-style assessment, and a complementary argument for process-rich, checkpoint-based design for extended writing tasks, two genuinely different but equally important responses to the same underlying concern.

For writing-heavy courses specifically, this reinforces the value of the kind of multi-stage, rubric-based grading and feedback process that naturally builds in ongoing evidence of a student's genuine work, rather than concentrating all assessment weight in a single, unmonitored moment.

A concrete data point worth factoring into assessment planning

This specific, controlled comparison offers genuinely compelling evidence about the scale of what unmonitored take-home exam conditions can mask, worth factoring directly into assessment format decisions this year, while keeping in mind that extended, process-based writing assignments call for a genuinely different, complementary response than a single high-stakes exam does.