a numeric character reference normalizes to its named replacement spelling where the reference substitutes both alike

Ruled by Dan (2026-09-06, on #217): not a preserved choice. Normalize only where the reader has recorded a fact that the respelling is render-equal; keep the author's spelling where no fact licenses the change. Before any code: this row's proof source, the generated whitespace battery (#239) run over the pair's domain through both programs, with the negative shapes pinned; the fact joins the block-level record the whitespace-fact change (#240) establishes; the printer never decides safety itself (no fact, no rewrite).

The pair

The named replacement spelling over its numeric character reference where the reference substitutes both to the same output: (C) over ©, (R), (TM), -- and ... over their numeric forms, under the replacements table (asciidoctor.rb REPLACEMENTS). Domain: the lens decodes numeric references, so render-equality must be proved on the decoded output, and the em-dash row consumes flanking characters (see the whitespace rows A7).

Start with the row's proof source and the generated whitespace battery from #239, including its negative shapes, then inspect the replacements table in asciidoctor.rb. Trace how the decoded pair and whitespace facts are recorded in the block-level record from #240. Done means only fact-licensed, render-equal spellings are normalized and the printer does not decide safety itself.