The glossary: one table, two jobs
One glossary entry fixes how a term reads in your copy and expands what a resident's question can match in search — confusing the two is the mistake to avoid.
One glossary entry does two genuinely different jobs, and the screen exists specifically to keep you from conflating them.
The rendering is what your copy uses
Some words have no faithful translation. "Arnona" rendered automatically into English is wrong in a way a resident will act on — so the glossary's rendering for a term is the exact text your interface copy is allowed to use for it, in each language you publish.
The synonyms are what residents actually type
Separately, the same entry lists synonyms — the words a resident might type that mean the same municipal concept. Someone searching "city tax" and your corpus only saying "arnona" is a vocabulary gap no spell-checker or stemmer closes, because it isn't a spelling problem. A synonym here expands their question before search runs, so it can still find the right article.
Using a synonym inside your copy is exactly the mistake the glossary is built to catch — a synonym is what a resident says, not what your city says back. The two columns look similar in the editor and mean opposite things.
Do-not-translate defaults to on, and that's deliberate
A department name your city chose has no meaningful machine translation — an automatic rendering of it names a department that doesn't exist, and a resident sent looking for it finds nothing. New terms start with translation switched off; you turn it on only when you've decided a term genuinely should be rendered automatically elsewhere.
There's no delete button, on purpose
Retiring a term is the only removal action available, and it isn't cosmetic: the platform's own seeder re-inserts a genuinely deleted term on the next update, in its original wording, because seed data is insert-if-missing. A hard delete would silently undo itself days later with no warning. Retiring is real, visible, and reversible by restoring the same row.
Two checks run as you type
The editor warns you if a phrasing you're adding already fires for a different term — a collision that would make two glossary entries compete for the same resident question. It also checks for a missing "definite article" form of a Hebrew word (a bare "welfare" not matching how residents actually write "the welfare"), because the platform's own measured history shows that exact gap silently dropping a real answer's ranking in production.
The try-it box at the top of the editor runs the platform's real matching rule, not an approximation — if it says a phrasing won't fire, that's the same "no" a resident's real question would get.