docs: localisation slice 2 release B (controller v0.253.0) — R-557 progress, R-575, R-576

All 179 Hungarian error messages carry a key; zero remain. 10-localisation.md gains §10.2: the
four properties util.MsgError had to have at once and the failure each one prevents, and the
plural rule as a BUNDLE rule rather than a per-call-site flag, with the answerable sentence and
what the other option would have cost.

Two instrument defects recorded rather than tidied away, because both shapes recur:

R-576 — the parity gate has a measured blind spot. The bulk converter dropped the continuation
of multi-line concatenations, damaging 7 producers, and the gate stayed GREEN: every surviving
fragment WAS a byte-equal base-commit literal, so its question ("is this text real?") was
answered yes while the CALL had lost half its sentence. Two behaviour tests caught it. The
general form: a structural gate over the TEXT cannot see a defect in the CALL.

And the script counting what was left was case-sensitive, so it said "0 remain" while five did —
R-565's shape inside the measurement. Every "no Hungarian left" claim in this slice is now made
case-insensitively and with both controls.

R-575 — the soft memory-overcommit warning has no error to carry a key and no language where it
is built, so it renders Hungarian on an English page. Named in the code, not hidden.

Live evidence includes a mistake I made and corrected: a probe of the deploy refusal INSTALLED
vaultwarden on demo-hp (the endpoint accepts before it validates), the same mistake the previous
session recorded. Removed through the product's own path with its data; verified gone.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_0159rPz1ZhFKsS53msqPYxtS

STATUS.md

@@ -1,46 +1,49 @@
 # STATUS — what works, what's broken, what's next
 
-**Updated 2026-09-18 — the sentences the program writes now follow the language.**
+**Updated 2026-09-18 (evening) — the sentences the program writes now follow the language, error messages included.**
 
 > **Ready for a volunteer: yes, unchanged.** A Hungarian household sees exactly what it saw
 > yesterday. I did not read the pages to decide that — I compared them, letter by letter.
 
-**Decisions I took.** None. One choice you were asked for stayed on its own stated default: notes
-saved to disk at night (last error, last warning) will be written in the box's language at the time
-they are written. A household that switches language sees last night's note in the old language until
-the next run. That is option 1 in the plan, and the plan says it is what happens if nobody decides.
+**Decisions I took.** None you have to reverse. One choice you were asked for stayed on its own
+stated default: notes saved to disk overnight will be written in the box's language at the time they
+are written, so a household that switches sees last night's note in the old language until the next
+run. That is what the plan says happens if nobody decides.
 
-**What I exercised, on the demo HP box.** The pages were already English. The SENTENCES the program
-writes into them were not — "Saved.", "The remote target is not set.", the country names, the yellow
-banners at the top. I moved 226 of them, and they now follow the language. Two I want to name:
+**What I did.** Two releases. The first moved the sentences the program writes into the pages — the
+little green line after you press a button, the yellow banners, the country names, the answers the
+page fetches in the background. The second did the **error messages**: about 180 sentences that are
+written deep inside the program and printed by whatever catches them. All of them now travel with a
+short code, so the page that shows them writes them in your language. **None are left in Hungarian
+only.**
 
-- **The little green "done" line after you press a button.** It travels inside the web address, so
-  it is written by one page and read by the next. It used to be written in whoever's language came
-  first. It now travels as a short code and the page you land on writes the sentence. A link you
-  already have open from before still shows its old sentence, word for word — I checked that live.
-- **The country list.** All 237 names, and the list re-sorts itself, so the English list really is in
-  English alphabetical order and not Hungarian order with English words in it.
+**My mistake, and what I did about it.** My first test of an error message asked the box to install an
+app with nothing filled in, expecting a refusal — and it installed the app instead, because that
+particular one needs nothing filled in. **This is the same mistake the session before mine made and
+wrote down, and I repeated it.** I stopped the app and removed it the way a customer would, with its
+data, and checked: no container, no folder, no leftover storage. The 23 apps that should be running
+are all running. The real lesson is sharper than "pick a different app": that button accepts first and
+checks afterwards, so it cannot be used to test a refusal at all. The tests were rewritten to use
+requests that are refused before anything is created.
 
-**How I know Hungarian did not change.** A new check freezes every Hungarian sentence in the program
-as it stood before I started — 7 467 of them — and refuses any sentence that is not exactly that,
-down to a comma. It caught a real mistake in itself on the first run and I fixed the check. Then on
-the box: ten Hungarian pages before and after. Six identical. The other four differ only in things
-that move on their own — a clock ticking over, an app that was unhealthy for a minute, and the
-"update available" line, which is true on the old version and not on the new one.
+**How I know Hungarian did not change.** A check freezes every Hungarian sentence as it stood before I
+started — 7 467 of them — and refuses any that is not exactly that, down to a comma. Then on the box:
+ten Hungarian pages before and after, six identical, the other four differing only in a clock ticking
+over, an app unhealthy for a minute, and the "update available" line.
 
-**What I deliberately did NOT translate.** Anything the central service reads and turns into an
-e-mail to the household. I checked, and found the plan was wrong about this: the service does not
-always write its own e-mail — for several kinds of message it sends the box's own sentence straight
-on. Translating those here would have changed an e-mail nobody asked to change. They are now frozen
-by a test, and they get their turn in the next step.
+**Two things went wrong in my own tools, and I am naming them rather than tidying them away.** My
+bulk converter quietly dropped the second half of sentences that were written across two lines — seven
+of them — and **the freeze-check did not notice**, because each surviving half really was a real old
+sentence. What caught it was two tests that read the sentence a customer would see. And my counting
+script ignored capital letters, so it told me "none left" while five were. Both are written down with
+a number.
 
-**What is not done yet.** About 900 sentences remain. 176 of them are error messages, which is the
-next release; the rest are notes saved to disk, which is the one after. Four smaller gaps I found
-while working are written down with a number each.
+**What is left.** The notes saved to disk overnight — one more release. Plus a handful of small gaps,
+each with a number.
 
-**Rows.** One closed, three opened. The register went from **264** to **266** open rows.
+**Rows.** One closed, five opened across the day. The register went from **264** to **269** open rows.
 
-**Needs you.** Nothing. The new version runs on the demo HP box only; the fleet floor is unchanged at
+**Needs you.** Nothing. Both new versions run on the demo HP box only; the fleet floor is unchanged at
 0.250.0. If you do nothing: the other boxes keep the version they have, and nothing breaks.
 
 ---