A git-history audit found 96.3% of AI-generated answer-key question blocks needed no post-release semantic repair¶

Claim · All claims
Evidence · 1 study · design 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? Of 273 distinct answer-key question blocks, 263 (96.3%) had no post-release repair across subsequent commits. → Jin-Guo Liu 2026

Evidence¶

Jin-Guo Liu 2026¶

Jin-Guo Liu, Shang-Qi Lu, Xin-Ran Shi, Long-Li Zheng, Wei Wang. (2026). Test-Driven, AI-Assisted Learning: Replacing Lectures with Weekly Closed-Book Tests. https://arxiv.org/abs/2606.23315 doi:10.48550/arxiv.2606.23315

q2 · i? · design · r2

Descriptive git-history audit of the repository's test and validation files, counting "273 distinct answer-key question blocks" and 10 semantic repairs. The authors call it operational evidence, not a controlled benchmark; no effect size printed.

"Subsequent commits contained 10 semantic repairs (proof logic, missing conditions, theorem references, or analysis), leaving 263 of 273 blocks (96.3%) with no post-release repair."

Discussion¶

Related Claims¶

• In a ten-request live textbook audit, citation integrity coexisted with substantive error: every citation resolved to its source, yet answers included retrieval misses and contradictions — related
• Evidence tracing links 96 of 114 generated claims to source chunks, but three claims with retrieval scores of 0.50 or below show retrieval similarity alone is insufficient — related
• Retrieval failure explained only 2 of vector RAG's 11 ungrounded answers; 9 answers retrieved the correct material and were still ungrounded, suggesting chunked context invites fabrication — related
• Model selection experiments show newer models are not strict improvements, with component-specific effects on quality metrics — related