The Innovation Generator subproject of AutoResearch connects source evidence, candidate actions and observed feedback. This update records literature exploration, a controlled computational diagnostic, and improvements to source acquisition and stop closure.
The project’s 2026-10-01 v7.3 research record covers authorized literature development and an eight-state computational diagnostic. Literature development used six live-model calls and three local acquisitions, ending with a PROPOSED_UNTESTED proposal on exploration stagnation.
The comparison includes a conditional ledger, fixed actions, flat presentation of the same information, disconnected feedback and ordinary active design. A/C/E passed 8/8 paired states and B/D passed 0/8. Each world evaluates one follow-up decision, supporting feedback control while leaving the independent value of full long-horizon frontier scheduling untested.
Engineering updates improve source retrieval, action fields, bounded context, acquisition receipts and stop closure. The next design concentrates model computation on explanation or action gaps; direct compilation and stable question-state separation remain research plans.