Comment from Amadeus Brandes
AnonymousSupportIndividual
Summary: Amadeus Brandes, an independent researcher, supports the pilot program by recommending that the FDA require a lightweight, machine-readable "decision and amendment evidence record" for participating trials. The commenter argues that this record would provide a model-agnostic way to ensure transparency, auditability, and reconstructability of AI-informed decisions without requiring access to proprietary model internals.
This comment recommends that the pilot require a lightweight, machine-readable decision-and-amendment evidence record for participating trials — capturing source trial state, AI recommendation, human decision, resulting protocol amendment, and later outcome, each pinned by hash and bounded in what it claims. It is responsive primarily to Decision Quality (II.B.2.a–b), Trustworthiness/transparency for proprietary systems (II.B.5.c), Participant Safety and Data Integrity (II.B.3.c), AI System Performance/drift (II.B.4.b), Operational Structure (II.A.4.b), and Knowledge Sharing (II.A.6). A key point: a provenance record is model-agnostic, providing a transparency mechanism that works for proprietary or otherwise non-disclosed systems where internal explainability metrics may be unavailable or insufficiently reviewable. Drawing on openly archived, reproducible work over public ClinicalTrials.gov amendment data (Zenodo DOI: 10.5281/zenodo.20816639) — 100 deterministic, independently verifiable Evidence Records — the comment also distinguishes certifiable provenance facts from non-certifiable decision-quality judgments. The full comment is attached. Docket No. FDA-2026-N-4390.