Before we automate WIL evaluation, we need to ask what we're actually measuring

Community-engaged Work-Integrated Learning (WIL) is increasingly positioned as a mechanism for advancing graduate employability and social responsibility, yet the frameworks used to evaluate it remain anchored to countable proxies: employment rates, satisfaction scores, and placement completions. This provocation argues that these metrics are structurally misaligned with what WIL practitioners actually know about placement quality: knowledge that is relational, contextually rich, and routinely invisible in institutional reporting. As Generative AI enters WIL evaluation, this misalignment becomes urgent. Automating existing frameworks will not resolve their limitations, and risks further marginalising the evidence that most reliably signals whether community-engaged WIL is working. Drawing on research into practitioner agency in learning analytics, this provocation calls for evaluation frameworks that legitimise practitioner judgement, centre community partner perspectives, and deploy GenAI in service of relational knowledge rather than as a substitute for it. Effective WIL evaluation is not primarily a measurement problem, it is a question of whose evidence counts.