Reviewed the complete existing annotation inventory while retaining the imported fields and row counts. The focused changes distinguish AR evidence from GR/PR chaperone models; no new process annotation was manufactured from a necessity phenotype. Existing activities and client systems outside AR remain in scope. UniProt source-section metadata was derived with the normal backfill helper.
The 90 imported rows are preserved. Direct AR evidence supports an optional/contextual p23 branch: ligand-binding capacity and promoter occupancy, with partial HSP90 independence PMID:22899854; newer primary work directly reports AR binding and protein stabilization in advanced prostate cancer PMID:41193657. The latter was normally fetched as abstract-only after PMC rate limiting, so detailed uninspected mechanistic claims were not added. This does not establish universal p23-dependent AR import. Preserve prostaglandin-E synthase activity and other chaperone clients; COX1 preference is not absolute COX2 exclusion. Migrated all 12 generic protein-binding legacy over-annotation actions to policy-compliant REMOVE or evidence-specific MODIFY, without denying interactions. Replaced unsupported deep-research quotation strings with primary evidence. The independent anti-aggregation assays [PMID:10543959; PMID:10811660] support a proposed ATP-independent protein holdase activity; they do not by themselves prove productive refolding. Refined broad true terms rather than rejecting them as false.
Replaced templated MODIFY summaries with the exact supported replacement: Hsp90 binding, cytosol, prostaglandin biosynthesis, PGE synthase activity, folding/telomerase complexes or telomerase-mediated maintenance. Existing actions and payloads retained.
All original annotation payloads were compared before/after and preserved; only reviewer assessments, source evidence, and the specified core-function location were changed.