Add Canadian provinces and territories to the jurisdiction data

[JonathanNus]

Bones ships US-only jurisdictions, so a Canadian project either falls back to the INTL generics or, worse, gets pointed at a US state whose frost line is too shallow. The concrete failure: prairie frost penetration runs 1800–2400 mm (71–94 in), deeper than any US state in the data (max 60 in), so footings come out a foot or more short with nothing in the UI to say so.

This adds 16 entries — all 10 provinces, the 3 territories, and a generic NBC 2020 fallback.

Data only

No engine changes. Everything rides the existing path: two JSON rows merged by profileFor(), then the same five levers in applyJurisdiction().

Notes on the choices

Codes are ISO-3166-2 style (CA-ON-S, CA-BC, …) for two reasons. Bare CA is already California, so a "Canada" code there would be a genuine footgun. And jurisdictionOptions() sorts by code, so the CA- prefix groups the Canadian entries together in the dropdown without touching the sort.

Ontario is split three ways (South / East / North). Its frost line runs 1200 mm in the GTA, 1800 mm in the Ottawa Valley and 2100 mm at Thunder Bay — a wider spread than most US states carry in one row, and the Ottawa Valley is also Canada's second-highest seismic zone. One row could not represent it honestly. The other provinces are single rows.

Canada is not an IRC jurisdiction. Every row sets ircBase: null and cites NBC Division B Part 9 ("Housing and Small Buildings"), the analogue of the IRC — the same shape the Wisconsin UDC row already uses for its non-IRC state code. The note field warns inference engines not to assume IRC section numbers or IRC braced-wall-panel methods, since NBC 9.23.13 selects lateral bracing from Sa(0.2) and hourly wind pressure instead.

Unit conversions are documented per row. Canadian codes state ground snow as Ss in kPa and frost depth in mm; converted here to the psf/inch fields this schema uses, at 1 kPa = 20.885 psf.

Timezones: guess.ts gains the Canadian IANA zones, and an en-CA locale now resolves to CA-GEN rather than INTL. One caveat noted in the code: modern tzdata folds Montreal into America/Toronto, so Quebec guesses as southern Ontario. The dropdown still wins, as with every guess.

Territories carry an explicit permafrost warning in frostLineNote (so it surfaces in the panel's notes, not just the data file) — a spread-footing depth is not a meaningful output where pile or space-frame foundations are the norm.

Sources

• NBC 2020 (NRC) — Division B Part 9, Appendix C climatic data
• Ontario Building Code updates — 2024 OBC adopts NBC 2020 + Ontario Amendments, effective 2025-01-01; SB-1 supplies Ontario climatic/seismic data
• Earthquakes Canada NBC 2020 seismic hazard tool
• Saskatchewan adopts the 2020 codes and Manitoba 2020 adoption — both effective 2024-01-01, both skipped NBC 2015
• Provincial codes matrix

Values are typical provincial figures, not site-specific: each row carries a caveat saying so, matching the existing disclaimer style. Quebec is flagged as still on the NBC 2015 base.

Tests

src/jurisdiction/canada.test.ts adds 10 tests. They pin the things that would regress silently — a missing data row degrades to the INTL fallback rather than throwing, so "it still renders" is not evidence a province landed. Also asserts profileFor('CA').name === 'California' so the collision stays fixed.

Full suite: 444 pass, 0 fail. tsc --noEmit clean.

One thing I found but did not fix

insulationByClimateZone is keyed "1"–"8", but wall-layers.ts looks it up with the raw stateClimateZone value, which reads like "5A (4A NYC/LI, 6A Adk)". Zero of the 51 existing US entries resolve, so the R-value label never renders for anyone. I left it alone — it is orthogonal to this PR and fixing it would change output for every US state. Happy to send a separate PR if useful.