A v1 Constants entry that is only a CJK full stop raises ValueError at the first parse: c.titles.add('。') then HumanName('john smith', c) #582

A v1 Constants set entry or capitalization_exceptions key that consists only of a CJK full stop (。, ． or ｡) raises at the first parse from 2.3.0 on. 1.4.0 through 2.2.0 accept it. This breaks the shim's contract that it never raises on a config v1 accepted (AGENTS.md, "The shim TRANSLATES").

>>> c = Constants()
>>> c.titles.add("。")
>>> HumanName("john smith", c)
ValueError: Lexicon.titles entry '。' normalizes to empty ...

Cause. #322/#323 (2.3.0) made Lexicon._normalize strip edge FULL_STOPS (.．。｡), where it had stripped only the ASCII period. v1's lc(), which SetManager stores, still strips only .. So the shim hands Lexicon an entry that folds to empty, and _normset rejects it. This is the same two-folds gap as #541, on a different character class.

Measured (adding the entry, then parsing a text it acts on):

． and ｡ behave the same.

Reach. Only a config that names a full stop by itself as vocabulary. 1.4.0 honored such an entry only on a name token that is nothing but that full stop.

Options

1. Drop with the A Constants set entry with edge whitespace raises ValueError at the first parse: c.suffix_not_acronyms.add('ma ') then HumanName('John Smith', c) #541 warning (recommended). Extend _v1_matchable with "and Lexicon does not fold it to empty", so these entries join the same drop and the same UserWarning. The only reading lost is a lone-。 token as a title, particle, suffix or connective, which no real name carries. This departs from 1.4.0 on that one token, so it is recorded in decisions.md under the A Constants set entry with edge whitespace raises ValueError at the first parse: c.suffix_not_acronyms.add('ma ') then HumanName('John Smith', c) #541 entry.
2. Keep the raise, but spell it for v1, like the Case repair: letter masks for the exceptions map (phd → PhD, bsc → BSc), md leaves the map, and roman numerals past iv (john smith xi → Xi) #459 mask exception: a message that names constants.titles.remove('。'). It is honest, but still a raise on a config v1 accepted, which the doctrine ranks below translating.

Note, related but not in scope: since 2.3.0, a titles entry 'dean。' also matches a bare dean, which 1.4.0 did not (1.4.0 through 2.2.0: dean john smith gives no title; 2.3.0: title dean). Same cause, the CJK full stops in _normalize's fold. The behavior widens and nothing raises. Decide when fixing this issue whether to accept it as a widening, as decisions.md#R4's #541 entry does for capitalized keys, or to drop the entry the same way.

Found in the #581 review (decisions.md#R4, 2026-10-02 #541 bullet).