A fixed-threshold sequential risk policy concentrated persistent-failure alerts within restricted alert budgets¶

Claim · All claims
Evidence · 1 study (2 entries) · associational r2 · q2
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly

Subclaims¶

q2 i? The sequential student-history policy triggered on 17.8% of test states with precision 0.566 and captured 25.2% of observed persistent failures. → Shihao Wang 2026 q2 i? At the planned 20% budget, the consecutive-failures rule captured 75 persistent failures (precision 0.676) versus 62 for the student-history policy (precision 0.559) and 66 for logistic regression (precision 0.595). → Shihao Wang 2026 (2)

Evidence¶

Shihao Wang 2026¶

Shihao Wang. (2026). A Risk-Adaptive and Evidence-Constrained Framework for Generative AI Feedback in Programming Education. https://arxiv.org/abs/2609.29874 doi:10.48550/arxiv.2609.29874

q2 · i? · associational · r2

Sequential threshold replay over recorded test trajectories with a validation-fixed threshold of 0.500. "Applied to the test sequence, it triggered on 17.8% of states, with precision 0.566 and capture recall 0.252"; a more sensitive configuration reached 22.0% coverage and capture recall 0.333.

"Applied to the test sequence, it triggered on 17.8% of states, with precision 0.566 and capture recall 0.252."

Shihao Wang 2026 (2)¶

Shihao Wang. (2026). A Risk-Adaptive and Evidence-Constrained Framework for Generative AI Feedback in Programming Education. https://arxiv.org/abs/2609.29874

q2 · i? · design · r2

Fixed-capacity policy comparison at the planned 20% budget (111 alerts) on the labeled test pool. "Logistic regression captured 66 (precision = 0.595; recall = 0.297), while consecutive failures captured 75 (precision = 0.676; recall = 0.338)"; the student-history policy captured 62 with precision 0.559.

"Logistic regression captured 66 (precision = 0.595; recall = 0.297), while consecutive failures captured 75 (precision = 0.676; recall = 0.338)."

Discussion¶

Related Claims¶

• Recorded natural-language feedback was associated with lower odds of subsequent persistent failure than no feedback in the randomized course subset — related
• Threshold changes in early alert systems can drastically reshape flagged-student volume and advisor workload and should not occur informally — related
• Georgia State's GPS Advising paired prediction with advising capacity and retention grants, and its strongest reported outcome was narrowed equity gaps rather than prediction accuracy — related