PN18: Assessment – What Is This Task Actually Evidence Of?

We do not mark capability. We mark something a student hands in, and then reason backwards from that submission to the capability we are prepared to certify. That gap between the claim and the evidence has always been there. Generative AI has widened it unevenly, without telling us which of our tasks it has touched. Nearly two-thirds of UK undergraduates now say assessment has changed significantly in response (Stephenson and Armstrong, 2026). Ours may not have.

This note is a method, not an argument. Work from the outcome you claim to the task you set to the inference the task asks you to make, then ask what a GenAI tool does and does not disturb along that path. Corbin, Dawson and Liu (2025) reduce the matter to a sentence: validity means an assessment reflects the student’s capacity to do or know something. Bearman, Nieminen and Ajjawi (2023) supply the organising logic for where the digital belongs in a task.

Authenticity will not rescue us. A task that resembles professional practice can still produce the wrong conclusion. Audit one task properly, name the inference, and add evidence rather than subtracting the assessment.