Agents-Team-Assisted Individualized Education
真正的瓶頸不是教師投入不足，而是證據被切碎
提案修改：從 AI 功能集合，轉為可治理的教育工作系統
Agent team 的教育價值：降低證據取得摩擦
前景供應答案；背景保存事實；critic 決定能否發布
RLM 探索長期證據；兩個閉環分別改善系統與教育工作
三種責任分離，避免 agent 自己產生、核准並發布結果
Git 追蹤系統變更，但不把可識別學生資料放進 repository
月度會議把查詢、查核、決定與新證據寫回同一閉環
第一年依 governance-first 順序建立可部署、可復原的系統
評估不只問答得準不準，還要問能否回溯、更新與復原
AI 整理證據；教育者保留決定權
Agents-Team-Assisted Individualized Education
The bottleneck is fragmented evidence—not educator effort
Proposal revision: from AI features to a governed education system
Agent teams reduce evidence-access friction
Foreground answers. Background preserves. Critic releases.
RLM explores longitudinal evidence; two loops improve the system
Separate duties so agents cannot self-approve
Git tracks system change—not identifiable learner data
Monthly review closes the evidence loop
Year 1 follows a governance-first path to deployment and recovery
Evaluate traceability, updating, abstention, and recovery
AI organizes evidence; educators retain authority