Why LLM-Based Deep Research (Still) Fails to Meet Scientific Standards | _A Living Research Protocol v.1_

This paper reports on a small, internal exploratory study on the effectiveness of LLM‑supported “deep research” under real‑world conditions of academic due diligence in the EU. The central claim is that LLMs do not automate scholarly work but instead transform it into a mode of epistemic stewardship, where human researchers supervise and audit machine‑generated suggestions. While the production of initial candidate lists is significantly accelerated, a “verification tax” emerges from the audits required to detect hallucinations and relevance errors, which often neutralises the initial time gains. Our study DOES NOT CLAIM statistical representativeness or full compliance with formal scientific standards but documents failure patterns as observed in everyday research practice.