This submission presents an engineering architecture for exploratory AI systems operating under costly verification, delayed evaluation, and constrained review capacity. It integrates deferred evaluation, priority-guided reexamination, minimal-state resume, selective irreversible evaluation, and frontier-based restoration into one rule-compliant flow framework. The manuscript includes numbered flowcharts, block reference tables, baseline comparisons, and ablation design, with concrete illustrations for EDA exploration and distributed orchestration.
Koji Mochizuki (Thu,) studied this question.