"Luo Si" and "Min Chi" lose their family name: Si and Chi read as credentials #791

si and chi are unambiguous entries in SUFFIX_ACRONYMS, so a trailing Si or Chi is read as a post-nominal and the person has no family name:

2.0.0 through 2.5.0 read it the same way as 2.6 dev does (measured 2.5.0 from the wheel, per AGENTS.md's comparison recipe). The bare form reports given-or-family, which is the wrong fork; the comma forms report nothing.

Both are common surnames. From tools/name_usage.py:

• Si (司, also Vietnamese): 1,166 US Census 2010 bearers, and 275+ Wikidata people with it as a family name.
• Chi (池/齊/遲): 7,273 Census bearers, and 219+ on Wikidata as a family name.

Both entries came in the 2019-12-11 bulk Wikipedia post-nominal import (af5bdab, #93) with no expansion recorded, and like rai, cha and mai they were never checked against surnames.

Found by running probablepeople's hand-labeled names (datamade/probablepeople person_labeled.xml) through the parser.

The two answers in decisions/suffix-acronym-collisions, measured with a private lexicon:

The open question is which outcome the relative-frequency criterion picks. Removal follows if the surname reading dominates; marking follows if SI and CHI are live credentials too. Marking reads every row of this grid correctly. Removal costs the bare and comma-form credential, as John Smith RAI already does.

What moves: no name in tools/differential/corpus*.jsonl (5,786 names) holds si or chi as a word, so nothing moves in the snapshot under either answer. The case rows above become the tests.