Evidence journal (no paid deep research run; built from UniProt and cached abstracts).
Reviewed all 18 existing rows and the core functions with the review and annotation-reviewer workflows. Preserved every machine-sourced term, evidence code, qualifier, source identifier, WITH/FROM entry and row order. No new annotations were added in this audit.
PMID:2199330 supports acetylornithine aminotransferase chemistry and native yeast sequence. This enzyme catalyzes the fourth ornithine-producing reaction; the review no longer calls it the last step. Broad mitochondrial location is refined consistently to matrix.
Normal publication caching completed. Normal Falcon research with Perplexity-lite fallback was attempted for this gene; authentication/quota failures prevented a generated report. Manual review used the actual cached primary sources and curated UniProt, and no provider-named research file was fabricated. Per-gene attempts and validation logs are in .cache/module-audit-logs/arginine-yeast-ARG8-*. Abstract-only or bibliographic-only sources are distinguished from full text in reference reviews; unavailable source details are not treated as experimental refutations.
Actions: ACCEPT 13, MODIFY 4, REMOVE 1. Record sections were backfilled; focused schema/reference validation passed on the initial complete panel. Final targeted checks and HTML rendering are recorded in the audit logs.
Restored supported GO:0005739 mitochondrion rows to ACCEPT under the explicit root-organelle policy; existing matrix rows remain separate. Reasons distinguish organelle screens, direct targeting, publication-linked curated evidence and unavailable full text. No original annotation payload changed. Cytosol decisions use target-specific fractionation/processing evidence rather than relationship qualifiers or generic precursor transit. For Aga1, matrix inference does not imply universal post-import autoproteolysis.