Interpreting Delayed Mastery Contradictions Under Alternative Measurement References
A short run of correct responses may support a mastery decision without establishing durable proficiency. We examined how the stable-state reference used to interpret later disagreement changes that interpretation. Four references crossed practice-history length with item-difficulty representation for the same 9,755 EdNet mastery decisions. The observed contradiction rate was 19.15%. Final-three references predicted 15.12%–16.13%, whereas all-history references predicted 19.56%–19.67%; the contrast held when item calibration, learner sampling, and prior estimation were propagated jointly. Within delayed sets that mixed practiced and new items, residuals for new items were about 14 points lower than for practiced items answered on the same day, a gap that a change in proficiency cannot produce. Simulations showed that offsets estimated from adaptively routed responses leave a harder delayed item mix uncorrected, producing a positive residual without decline, and that learning during practice can conceal a decline under all-history references. Candidate-level contradiction exceeded the final-three reference unless non-returners' odds were about 35% lower than comparable returners'. The residual quantifies disagreement with a specified model; interpreting it as forgetting requires linked items, planned follow-up, and a dynamic state model.
Authors
- Kang Liu (ORCID: https://orcid.org/0009-0004-3214-505X)
Institutions
- Shandong Normal University (CN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24184650
- Primary Topic
- Intelligent Tutoring Systems and Adaptive Learning
- Type
- preprint