Evidence behind the platform

A critical review of why oral assessment — long sidelined in mass education because it does not scale — deserves reconsideration as a response to generative AI. We synthesize the evidence on oral versus written assessment, the cognitive science of retrieval and dialogue, and early work on AI-mediated oral assessment, and argue that conversational AI may relax the cost constraint that once confined oral exams to elite settings. We are explicit that this central claim is a hypothesis to be tested — not an established result — and we set out the research agenda that would test it. Not yet peer-reviewed.

The empirical companion to the review above. With partner schools, we are studying how AI-mediated oral assessment compares to experienced teachers' judgment (validity), whether outcomes differ across student groups (equity), and how the format shapes the way students prepare and learn. Data collection is underway; we will report findings — and their limitations — only when the evidence supports them.

We document our research methodology — study design, measures, and analysis — and share it with partner schools on request.

AI scores are calibrated against a panel of experienced teachers, with ongoing review of agreement rates.

We actively test for score disparities across language background, disability status, and socioeconomic indicators. If we find them, we investigate them.

Viva scores are advisory signals. Final grade decisions remain with the teacher. We are explicit about this in every interface.