Audit trail for: A measurable trustworthiness audit for AI-driven scientific discovery

Complete audit trail underlying the manuscript "A measurable trustworthiness audit for AI-driven scientific discovery" (Sheng et al., submitted to Patterns (Cell Press), 2026). Includes all sub-agent JSON outputs, Tournament I and II checkpoint chains, Bradley-Terry-Luce input matrices, blind-scan logs, as-discovered and grounded citation pools, the position-bias decision log, retrieval-vs-reasoning control-arm data, the dAMD cross-seed (Run 2) and cross-model (independent vendor family) replications, and the ALS Run 2 substrate-conditional grounding data. See manuscript Data availability and Supplementary Information Section 7 for a file-by-file index. Companion code: https://github.com/SJT503/SDL-dAMD.