Canadian French (fr-CA) as its own locale, not a fallback

[distronode-com]

First of the features split out of #30, in the order of appetite you described on that thread.

fr-CA currently resolves to the France copy through the primary-subtag fallback. The differences are not stylistic:

	fr	fr-CA
email	e-mail	courriel
reschedule	reprogrammer / reprogrammation	reporter / report
personal data	données personnelles	renseignements personnels
before ! ? ;	narrow no-break space	no space
July, abbreviated	juil.	juill.

The third row is the one that matters most: renseignements personnels is the statutory term in Quebec, so the France word is wrong in a legal sense rather than a register one. The last row is CLDR's own disagreement, not a preference, which is why TestDateTablesMatchCLDR catches it.

The only structural point

This is the first regional locale. Resolve already prefers an exact tag before falling back to the primary subtag, so no matcher change was needed — but that means the risk is not in the new locale, it is in the eight that already work. A visitor sending fr-FR or plain fr must be untouched by this file existing.

TestResolve now pins both directions:

{"Canadian French resolves to its own locale", "fr-CA,fr;q=0.9", "", "fr-CA"},
{"European French is unaffected by fr-CA", "fr-FR,fr;q=0.9", "", "fr"},
{"plain fr is unaffected by fr-CA", "fr", "", "fr"},
{"fr-CA can be selected by the ?lang override", "en", "fr-CA", "fr-CA"},

Mutation-checked rather than assumed: with the locale file moved aside, the two positive cases fail (Resolve("fr-CA,fr;q=0.9", "") = "fr", want "fr-CA") while the two fr/fr-FR cases stay green, which is what they are for — they assert an absence of change and would be worthless if removing the feature moved them.

What is verified and what is not

The three guards the other eight locales pass apply unchanged: same key set as English (TestAllLocalesHaveTheSameKeys), matching printf verbs including the %[1]s/%[2]s positional pair no_available_times_host needs (TestAllLocalesHaveMatchingFormatVerbs), and date tables cross-checked against CLDR (TestDateTablesMatchCLDR). So the structure is verified.

⚠️ The wording is not. It is an unreviewed draft, like every non-English locale in this repository, and nobody who speaks Canadian French has read it. The CHANGELOG says so plainly, so a correction arrives as an ordinary pull request rather than as a bug report from a Quebec booker.

Currency and percent take a non-breaking space before $ and % in Canadian French. No key carries either today, so that rule is recorded in ARCHITECTURE §23 rather than applied — worth writing down before someone adds a priced-event string.

Scope

Nine files: the locale, four test cases, and the doc/CHANGELOG counts that say "8 locales". No handler, no matcher, no schema. internal/i18n passes; the full suite passes on this branch.

Adding a locale is still just adding internal/i18n/locales/<code>.json — init() globs the directory, and nothing else needed a list updating.

🤖 Generated with Claude Code

[pullfrog]

ℹ️ No critical issues — two stale locale counts outside the files this PR already updated.

Reviewed changes One commit adding Canadian French as a first-class regional locale (data + Resolve pins + doc counts). No matcher or handler changes.

• fr-CA.json locale table New 158-key file with the claimed fr vs fr-CA differences (courriel, reporter/report, renseignements personnels, punctuation spacing, CLDR juill.). Key set matches en; package tests pass.
• TestResolve pins both directions Exact fr-CA, European fr-FR/fr unaffected, and ?lang=fr-CA override — the half that would break silently if the matcher preferred the new regional tag too broadly.
• Doc/CHANGELOG counts CLAUDE.md, CONTRIBUTING.md, ARCHITECTURE §23, and the Unreleased CHANGELOG entry now say 9 locales / list fr-CA.

ℹ️ Locale counts still say 8 in two places this PR did not touch

README.md still markets "8 languages" and lists English through Swedish with no Canadian French — the most visible user-facing surface for the feature. docs/ARCHITECTURE.md (~L809, LiveKit "not translated" known issue) still says the booking surfaces "ship in 8 languages (§23)", which now contradicts §23's "9 locales". Historical "8 locales" changelog entries are fine to leave.

Technical details

# Stale locale counts outside the PR's doc set

## Affected sites
- `README.md` ~L193–195 — still "**8 languages**" and the bullet list without Canadian French
- `docs/ARCHITECTURE.md` ~L808–809 — LiveKit known-issue still "ship in 8 languages (§23)" while §23 now says 9

## Required outcome
- User-facing and cross-referenced live counts match the shipped locale set (`en es fr fr-CA de it pt nl sv` / 9)
- Do not rewrite historical CHANGELOG / ARCHITECTURE dated entries that correctly described the state at ship time

## Suggested approach (optional)
- Bump README to 9 languages and add Canadian French to the list (same register as the other names)
- Change ARCHITECTURE L809 "8 languages" → "9 languages" (or "§23's locales") so it stays consistent with §23

  ^{｜ Fix it ➔ ｜ View workflow run ｜ Using Grok ｜ 𝕏}

[shockalotti]

The locale itself follows the documented pattern (one JSON file, three guards green, honest unreviewed-draft label in CHANGELOG per §23's marketing rule) — no issue with the fr-CA copy as a draft. But this PR puts the docs in a self-contradictory state:

It updates the count to 9 in ARCHITECTURE §23, CONTRIBUTING, and CLAUDE.md, while README.md ("8 languages" bullet) and ARCHITECTURE §18 ("ship in 8 languages") still say 8. Merged as-is, the repo claims 8 and 9 locales at the same time. Please update those two spots to 9 in this PR.

Standing note (no action needed here): per §23's translation-quality rule, fr-CA stays unmarketed until a native speaker reads the booking page, manage page, and four emails end to end — the CHANGELOG already says this, good.

[pullfrog]

✅ No new issues found.

Reviewed changes One docs commit since the prior review, filling the two live locale-count gaps that review called out. No code or locale-table changes in this delta.

• README language bullet Count is now 9 and the list includes Canadian French alongside French.
• ARCHITECTURE LiveKit known-issue The §18 line now says the booking surfaces ship in 9 languages, matching §23.

  ^{｜ View workflow run ｜ Using Grok ｜ 𝕏}

[distronode-com]

Fixed in 6bd9e48: the README feature bullet now says 9 languages and lists Canadian French, and the LiveKit entry in §18 says 9. I left the two dated 2026-08-20 entries (the CHANGELOG and the ARCHITECTURE log) at "8 locales", since they describe what shipped that day.

Understood on the translation rule: fr-CA stays an unreviewed draft until a native speaker has read the booking page, the manage page and the four emails, and the CHANGELOG line keeps saying so.

[pullfrog]

✅ No new issues found.

Reviewed changes Changelog hygiene after merging main (which cut 0.9.0). No locale-table or matcher changes in this delta.

• fr-CA changelog placement The Canadian French entry now lives under [Unreleased], not [0.9.0].
• FRAME_ANCESTORS changelog placement Same correction for the admin frame-ancestors note, which landed on main after the 0.9.0 tag.

  ^{｜ View workflow run ｜ Using Grok ｜ 𝕏}

[distronode-com]

Merged main in (86a914f), and fixed something the merge did not flag. This branch was cut before 0.9.0, when its CHANGELOG entry sat under [Unreleased]. 0.9.0 renamed that section, and git carried the entry along with no conflict, filing Canadian French under 0.9.0. 60db2b3 moves it back under [Unreleased], text unchanged.

#40 hit the same thing when it was squash-merged: FRAME_ANCESTORS is listed under 0.9.0's Added on main, though 0.9.0 shipped a week before it. 707e4a4 moves that entry too, as its own commit, so it lands with this PR. If you'd rather fix main directly, say so and I'll drop that commit.