During active transfer, the application rejected learner coaching and export requests and redacted earlier project content in all 120 episodes¶

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? In all 120 episodes, active-transfer responses passed the redaction check, coaching requests were rejected without reaching the provider, and learner exports were rejected, while the facilitator retained inspection access. → Nizam Kadir 2026

Evidence¶

Nizam Kadir 2026¶

Nizam Kadir. (2026). LearnAdapt Praxis: Controlled AI Assistance and Evidence Traces for Adult Workplace Learning. Technical research report. https://arxiv.org/abs/2610.02699 doi:10.48550/arxiv.2610.02699

q2 · i? · design · r2

Results of the 120-episode synthetic rehearsal (Results section, workflow and authority group: 1,080 transfer-restriction checks with zero unexpected failures). The report states "All 120 active-transfer coaching requests were rejected without calling the injected provider"; no effect size applies.

"All 120 active-transfer responses passed the composite redaction check. All 120 active-transfer coaching requests were rejected without calling the injected provider; all 120 learner exports during that phase were rejected."

Discussion¶

Related Claims¶

• A frozen synthetic rehearsal of 120 serial project episodes passed all 3,990 recorded checks with zero unexpected failures — related
• Injected provider failures produced recorded unavailability and zero fabricated hints, while valid and reentrant conditions stored 30 synthetic hints each — related
• GPT-4 over-redacts famous names, locations, and mythological creatures that are not PII in educational forum discussions — related
• GPT-4 achieves high recall (average 0.958) but low precision (average 0.526) when de-identifying MOOC forum posts — related
• Supervised machine learning de-identification still outperforms GPT-4, though GPT-4 exceeds class-list/regular-expression and transformer approaches in recall — related
• Authentication regressions exposed a session-expiry defect in the preceding release, and the recorded release establishes a fresh anonymous session after timeout