[ GLOSSARY ]
What is knowledge tracing?
Definition
EvidenceKnowledge tracing
Knowledge tracing is a family of models that estimate whether a student has mastered a skill from their answers over time.
Why it matters for schools
It works best on clean question data; tracing from chat is harder. With GPT-4o tagging tutoring turns, next-answer prediction reached an AUC of 0.66 on 153 real dialogues and 0.77 on a larger set with simulated students. Misconception detectors raised false alarms about 8 times as often as true ones at realistic rates. Treat a chat-based mastery estimate as a hypothesis, verified by a short unaided check.
In Nexus
Nexus does not score mastery from chats, and we do not claim to know what each student has learned.
[ FAQ ]
Questions and answers
Can AI tell what a student knows from a chat?
Only roughly. Chat-based tracing reached AUC 0.66-0.77 in research, and misconception detectors raised many false alarms.
What is a better signal of mastery?
Success on a new item without help, ideally after a delay.
Should mastery scores drive placement?
No. A decision with major impact on a student should not rest on a single score.
[ SOURCES ]
Sources
Facts checked Oct 18, 2026.
- 1Scarlatos, Baker and Lan, dialogue knowledge tracing, LAK 20252025-03 · learninganalytics.upenn.edu
- 2Imran and Bulathwela, misconception detection, arXivJun 22, 2026 · arxiv.org
- 3AERA, APA and NCME, Standards for Educational and Psychological Testing2014 · testingstandards.net
[ AI FOR SCHOOLS ]
The AI your school can approve.