Reproducibility materials for "Actual-use verification and lineage-resolved fidelity in scientific AI"

This record provides reproducibility materials supporting the manuscript “Actual-use verification and lineage-resolved fidelity in scientific AI”. The study evaluates scientific-AI fidelity by distinguishing assigned scientific objects, realized scientific objects, lineage or pathway fidelity, terminal scientific fidelity and claim boundaries. The archive contains detailed evidence registries, comparison specifications, provenance and audit records, structured source records, manuscript source materials, display assets, and code for assembling and rendering the supplied publication materials. It also preserves the available historical CoralTemp/Coral Reef Watch daily-classification source excerpt used in the documented evidence chain. The released materials support inspection of the reported evidence, scientific-object assignments, realized-use verification, lineage-resolved comparisons and presentation-level reproduction. They do not constitute a complete executable end-to-end reproduction of all scientific analyses. Complete raw-input and terminal-reference payloads, full benchmark execution and scoring pipelines, original scientific figure-generation programs and the original pinned scientific execution environments are not included. Public upstream scientific products and their provenance are documented in the manuscript and archive. The archive should therefore be interpreted as a provenance-rich reproducibility and audit package rather than as a general benchmark of scientific-AI accuracy or reliability.