A structured mixture-of-experts candidate improved some corrections on fixed-evidence cases but introduced new factual and continuity regressions, so activation was held¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2–r3 · 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 structured Qwen mixture-of-experts response had a faster warm median (3.00 s versus 3.47 s for 8B) but review found three regressions across twelve fixed-evidence cases. → Sidney Shapiro and Joshua Lindemann 2026 q2 i? The candidate corrected earlier accounting and pricing mistakes but turned a qualified benefit of holding cash into a guarantee and lost continuity on a reference to a prior example. → Sidney Shapiro and Joshua Lindemann 2026 (2)

Evidence¶

Sidney Shapiro and Joshua Lindemann 2026¶

Sidney Shapiro and Joshua Lindemann. (2026). On-Premises Multi-Course RAG Tutoring for Business Education: Hardware–Software Trade-offs in a Campus AI Tutor. arXiv. https://arxiv.org/abs/2610.02510 doi:10.48550/arxiv.2610.02510

q2 · i? · design · r3

A separate fixed-evidence diagnostic on twelve frozen cases changed model, prompt, response contract, validation, and transport together; the decision summary reports "three regressions across twelve fixed-evidence cases" and activation was held.

"Warm median was 3.00 s versus 3.47 s for 8B, but review found three regressions across twelve fixed-evidence cases."

Sidney Shapiro and Joshua Lindemann 2026 (2)¶

Sidney Shapiro and Joshua Lindemann. (2026). On-Premises Multi-Course RAG Tutoring for Business Education: Hardware–Software Trade-offs in a Campus AI Tutor. arXiv. https://arxiv.org/abs/2610.02510

q2 · i? · design · r2

The fidelity comparison section reports the mixed pattern: "corrected earlier accounting and pricing mistakes" alongside a new overstated claim and a continuity loss, so the configuration was not activated.

"The larger candidate corrected earlier accounting and pricing mistakes, but also turned a qualified benefit of holding cash into a guarantee and lost continuity on a reference to a prior example."

Discussion¶

Related Claims¶

• Current best learner performance models are severely biased outside the interval containing most of the data, hindering downstream adaptive policies and open learner models — related
• Logistic regression with the best feature vector outperforms all other approaches on 4 of 9 datasets while DKT leads on the remaining 5, and Markov process methods lag behind — related
• Larger dense generators passed smoke probes but all exceeded the +30% classroom latency limit in bake-off round 1 — related
• LLM output can be misleading or erroneous because a stochastic element in text generation sometimes selects less likely continuations, so it requires continuous supervision and critical evaluation by domain experts. — a broader claim this one bears on