docs(i18n): add Spanish, Italian, Dutch, Norwegian and Danish

[helmihyo]

Adds Spanish, Italian, Dutch, Norwegian Bokmål and Danish. The site now carries ten languages: 21 pages each, 210 translated pages in total.

Why five at once

The machinery from #26, #30 and #33 had been exercised one language at a time. Doing five in parallel tests whether it holds when nothing can be watched by hand — and it found things a single-language pass would not have. The rules that leak between languages only become visible when the languages are written side by side.

What is here

docs/es/, docs/it/, docs/nl/, docs/nb/, docs/da/, five glossaries in solutions/translation/, five locale blocks in mkdocs.yml, and two new scripts described below.

Norwegian is Bokmål (nb), the variety about 85–90% of Norwegian writing uses.

Each glossary opens with the rules where the sibling languages are actively wrong for it, because copying a pattern that worked next door is the recorded failure mode in this repository:

	quotations	address	proper-name compounds
Spanish	«…»	impersonal — no usted, no tú	no hyphen at all
Italian	“…”	impersonal, infinitive instructions	no hyphen at all
Dutch	“…”	u	junction hyphen
Norwegian	«…»	du	junction hyphen
Danish	»…« — outward, the opposite of Norwegian	du	junction hyphen

Norwegian and Danish were translated in parallel by agents that could not see each other. Their quotation marks came out opposite, as they must.

Verification

Structure vs. the English source	0 deviations across 105 pages
mkdocs build --strict	clean
check_anchors.py	9750 anchors across 212 pages, all resolve
check_typography.py	clean in all nine translated languages
check_glossary.py	clean in all five new languages
translation_status.py	current=21 in all nine
Safety-critical passages read against English	1031 compared, 0 defects

That last number is the one that matters. Every warning, danger, caution and note admonition, every numbered procedure step and every prohibition on the five pages carrying fuse ratings, voltages, polarity and jumper settings was compared sentence by sentence against its English source. A zero is only worth reporting with the count beside it.

What the automation could not see, and now can

check_glossary.py finds a prescribed term nobody used. It cannot find a term that acquired a rival — the same connector under two names on adjacent pages, which no reader can reconcile. That gap was found by hand in Swedish and is now covered by a per-language sweep: 21 such concepts, 88 occurrences resolved to whichever word the glossary prescribes, 22 new glossary rows recording the decisions the glossary had not yet made.

Blind replacement would have been wrong in several places:

• Spanish balanceo is wrong for super-capacitor balancing and right for rocking a stuck HAT loose.
• Italian microcontrollore renders microcontroller, a different English word from controller.
• Danish leder is the conductor inside the cable and ledning the cable — a split the English source makes too.
• Norwegian had left 11 image alt texts in English. A screen reader would have spoken them in English on a Norwegian page. Measured centrally: every other language has exactly one untranslated alt text — the bare product name ![HALPI2] — and Norwegian had twelve.

Two new scripts

scripts/map_anchors.py rewrites English anchor fragments to the translated slugs. Translators leave fragments alone; this maps all of them once the language is complete. The mapping is positional — the nth heading of the English page and the nth heading of the translation are the same heading — because matching on heading text cannot work once the text is in another language. It rewrote 88 fragments here; the same work was 18 hand-written sed expressions in the Swedish branch.

scripts/check_typography.py measures each glossary's own rules. Every rule is scoped to the languages whose glossary asks for it, because the same pattern is correct in one language and a defect in the next: German compounds a proper name with hyphens throughout (NMEA-2000-Netzwerk) and its glossary calls a missing one the clearest sign of a translator who does not write German, while every other language treats that chain as an error. French requires a space before ; : ! ? and requires it unbreakable; everywhere else any space there is wrong.

Four times while writing these checks a rule looked violated and was not: »…« matched the gap between two correct «…» pairs; a space before a colon came from English comments inside a code fence; HaLOS-avbilder is a correct junction hyphen; HALPI2-hardware is a repository name. Each was caught by running the checker against a language that had already been reviewed and fixed — if it reports a defect there, the defect is in the checker.

What this needs from a reader

Nobody who reads Spanish, Italian, Dutch, Norwegian or Danish has read any of this. That was decided explicitly, with the scale on the table: 105 pages at once, carrying fuse ratings, voltage ranges, polarity instructions and jumper settings. Every mechanical check above passes, and none of them can tell whether a fluent sentence says the right thing. The 1031-passage audit is the strongest evidence here and it is still an unverified reading, not a native one.

Corrections belong in solutions/translation/<language>-glossary.md as well as in the pages, so the next translator inherits them rather than repeating the choice.

🤖 Generated with Claude Code