Title: Labora. Roga. Proba.: Protocol for AI-Assisted Homework
DOI: 10.5281/zenodo.23112360
Tags: education, youth, cognitive-atrophy
---
A protocol paper for upper secondary schools (ISCED 3, roughly ages 14 to 18), proposed for piloting as an open tool under CC BY 4.0. Its premise is that students will use AI for homework anyway, so the question is no longer whether they use it but whether they understand what they hand in. Text detectors, bans and full oral exams are reviewed as the existing responses.

The three steps give the protocol its name. The teacher sets in advance a level of the **Mechanism of Cognitive Stimulation** (MCS), a ten-step slider that assigns each stage of the work (goal, method, execution, verification) to the student or to the AI. The student works with an assistant set to that level (*labora*). At the next class the teacher questions a few students chosen at random (*roga*), who show that they understand (*proba*).

The central idea is **level-conditioned verification**: the MCS matrix is read in reverse, so each level points to the stage the oral question should test, and the same homework yields different questions at different levels. The teacher's AI drafts the question from a short form; answers are judged on a four-step scale adapted from the SOLO taxonomy, and the AI never grades. Three output types range from the bare question to indicative content, response levels and a second question.

An incentive analysis argues that random selection makes cheating costlier than learning, and that three of the four strategies open to a student end in learning. The paper gives the full teacher prompt, a browser-based prompt generator that stores no student data, related work with its known criticisms, and the three measurements a pilot should make, starting with whether the questions really differ between adjacent levels.