---
source_url: https://arxiv.org/abs/2605.04727
ingested: 2026-05-13
sha256: b67746b48114826e129557cf9221663b9da090ab83e65dfd9b66a95320ae7ece
---

# Ensuring Reliability in Programming Knowledge Tracing: A Re-evaluation of Attention-augmented Models and Experimental Protocols
arXiv: 2605.04727
Authors: Kim, J., Kim, H.
Published: 2026-05-06
Categories: cs.LG, cs.SE
Venue: ITS 2026 (Springer LNCS)

## Abstract
Programming Knowledge Tracing (PKT) has advanced through hybrid attention+RNN architectures. This paper identifies critical issues: attention dimension misconfiguration, temporal causality violations. Under controlled experiments, the advantage of attention-enhanced models over standard DKT is significantly reduced.