Decide How the #2100 Recorded-Name Charset Lane Finishes After Its Review Budget #2146

The unattended lane for #2100 (handoff #2145) stopped at the local review budget. Two rounds of edits answered two local strict review passes, and the third pass still raised three findings on text this change wrote, so per local-strict-review "Disposing of Findings" editing stopped and the rest comes here. No pull request exists yet. The work is on branch feature/auto-2100 (head 6d537301), pushed, with every local gate green.

What the branch does now. The charset rule states the ruling script-agnostically: a recorded name keeps its letters and marks, and anywhere else a letter in no tier takes its ASCII form. The prose gate implements only the Latin subset the issue named as the minimum: a Latin-named letter that is not a compatibility form, plus a U+0300-U+036F diacritic (the grapheme joiner excluded) carried by a letter. Other scripts' letters, ligatures, subscripts, fullwidth forms, variation selectors, and a lone mark stay charset-unknown.

The three open findings, all small:

1. prose_lint.py's docstring and scripts/README.md justify reporting other scripts as "most often a lookalike", but Latin-named homoglyphs of ASCII (U+0251 alpha, U+0261 script g, U+0131 dotless i, small capitals, and the punctuation-like U+01C3, U+01C0, U+0294, U+A78C) pass the gate.
2. No test pins the category.startswith("L") guard. U+271D LATIN CROSS is category So with a Latin name, and only that guard reports it.
3. AUDIT.md line 92 places "apart from the letters of a recorded name" before the parenthetical, so the parenthetical reads as describing the name.

Question: how should the lane finish?

1. Recommended: fix the three as scoped and drive the PR to develop. Drop the lookalike justification rather than defend it, so the docs say the gate covers the Latin part of the rule and nothing wider. Add a U+271D case and reorder the AUDIT sentence. Track widening (other scripts, Latin homoglyphs) as a separate issue. Reason: it closes Accept Letters in Recorded Names in the Charset Rule and Prose Gate #2100's ask (Latin at least, lone mark reported, tier 1 still reported) with no new enumeration, the thing that kept producing findings here.
2. Also reject the Latin-named ASCII lookalikes before the PR. An explicit deny set for those letters. Reason: keeps the gate's homoglyph coverage close to what it was. Cost: a new enumeration that reviewers will keep growing, and a few of these are real letters in some orthographies (U+0294, U+01C3 in Khoekhoe names) that would then be reported.
3. Accept every Unicode letter, all scripts, as Accept Letters in Recorded Names in the Charset Rule and Prose Gate #2100 first worded it. Reason: matches the script-agnostic rule exactly. Cost: a Cyrillic or Greek homoglyph inside an ASCII word in code or prose passes the gate silently, which the first review round showed.

Blocks: the feature -> develop pull request for feature/auto-2100, not yet opened.

🤖 Generated with Claude Code