1. declined — deleting `parameter selection` introduced a live audit finding requiring that field despite the coverage explanation in `complete-note`, so its original text and order were restored; left for a person — live previews confirm the `parameter selection` and `digits: 100` renderer warnings, which require a site fix outside this artifact-only run; precision metadata is unchanged.
2. done — changed the convention comment to “The quantiles use the lower-tail probability convention” and restricted the exact-rational statement to probabilities $p$; checked the live definition and columns, official R documentation, all 392 parameter tuples, 42 closed-form enclosures in 600-bit Arb arithmetic and 216 numerical CDF brackets at 150 decimal digits through `agents/sage.sh`; all checks passed and no values changed.
3. done — moved the upper-tail conversion and its definition into `formula-upper-tail`, removed the duplicated method claim from the scope comment and completeness note, and preserved `rigour details`; derived the identity from the continuous strictly increasing CDF, checked both significance-level complements exactly, read the complete revised paragraphs and inspected live section previews.