fix(cv): capitalise every word of a heading derived from a record key

[eduardstan]

Intent

Unify the two heading-capitalisation rules in scripts/build-cv-data.mjs. The generator had two: tableHeader capitalised only the first character of a row key, while keyWords (used for macro names and section headings) capitalised every word. Neither was documented as deliberate. The first-character rule made a two-word column header - 'Programme / Level' - unreachable from any plain record key: the only way to get it was to store the capital in the record key itself, which the website then publishes verbatim on its provenance line ('course, programme / Level, key topics, hours'). That is presentation leaking into the record and onto a public page, against this repository's facts-versus-design boundary.

Decisions made: both call sites now share one 'capitalise' helper - deliberately one rule, no third invented. tableHeader applies it per word via k.replace(/[A-Za-z0-9]+/g, capitalise) rather than reusing keyWords, because keyWords splits on non-alphanumerics and its join would eat the separator ('programme / level' -> 'Programme Level'); the replace preserves slashes and spacing. Explicitly ruled out of scope by the task: acronym handling, lower-casing joining words ('of', 'and'), locale rules, and any configurable casing policy - a heading the simple per-word rule cannot express is a finding to report, not a reason to build a casing system.

Constraint honoured and proven: the bundled example's printed output must stay byte-identical. cv/generated/cv-data.tex regenerates unchanged (git clean), all three PDFs were rebuilt with latexmk -xelatex, and npm run check:maintainer passes all three recorded baselines untouched - no baseline was updated. npm run check, npm test and npm run check:format are green. One new test in scripts/build-cv-data.test.mjs covers the multi-word key and the preserved separator; a third casing site at line ~331 was checked and is an unrelated bibliography prefix letter, left alone. AGENTS.md was deliberately not edited: the rule is documented at its call site and AGENTS.md forbids restating values that live in another file.

What Changed

• scripts/build-cv-data.mjs now derives every heading from one headingCase rule that capitalises each word of a record key and leaves the separators alone, replacing tableHeader's first-character-only casing; keyWords keeps its ASCII split for macro names only, and sectionHeading reads separators as spaces (field_work prints "Field Work"). A word is any run of letters or digits in any script, so macronised and accented keys survive.
• The website applies the same rule to the same keys: headingCase is exported from web/src/lib/cv-schema.ts, web/src/pages/cv.astro uses it for the teaching table's <th> and data-label, and announcements.ts uses it for the /lately/ feed kind label — display casing only, since the slug still lowercases and strips to ASCII. Both copies document each other; no module crosses the plain-node/Vite build boundary.
• New assertions in scripts/build-cv-data.test.mjs and web/src/lib/cv.test.ts cover multi-word and non-ASCII keys and the preserved separator; content/README.md documents that column keys stay lower case in the record. The bundled example is byte-identical — cv/generated/cv-data.tex regenerates clean and all three recorded PDF baselines pass untouched.

Risk Assessment

✅ Low: The final round changes one display label expression to the rule already shared by both surfaces, and every identity form (slug, anchor, CSS rule, radio id) is provably casing-invariant because it passes through slug(), with no effect on the bundled example's site or printed output.

Testing

Ran the targeted generator, cv and announcements self-checks (all green), confirmed cv/generated/cv-data.tex regenerates byte-identical and that all three PDFs rebuild and pass their untouched baselines, then proved the user intent end-to-end on an adopter-style record copy: a plain lowercase programme / level key now prints "Programme / Level" in the built PDF and on the rendered /cv/ page while the provenance line still publishes the key verbatim in lowercase, with the base commit producing "Programme / level" for the identical record as the before-shot; /lately/ shows the "Field Work" label with its filter slug unchanged. Visual evidence captured as PDF page renders and browser screenshots; no failures or flakes.

• Evidence: Built PDF: teaching table header from lowercase record keys ("Programme / Level", "Key Topics", "Hours") (local file: /tmp/no-mistakes-evidence/01KYTA3WJYEVCJ2JP52G8K9XWF/pdf-teaching-table-1.png)
• Evidence: Built PDF page 1 (full page, demo record) (local file: /tmp/no-mistakes-evidence/01KYTA3WJYEVCJ2JP52G8K9XWF/demo-cv-page-1.png)
• Evidence: /cv/ with Inspect sources on: provenance line publishes the record keys lowercase while the column heading is title-cased (local file: /tmp/no-mistakes-evidence/01KYTA3WJYEVCJ2JP52G8K9XWF/web-cv-provenance-keys.png)
• Evidence: /cv/ teaching table rendered from the demo record (local file: /tmp/no-mistakes-evidence/01KYTA3WJYEVCJ2JP52G8K9XWF/web-cv-teaching-table.png)
• Evidence: /lately/ kind filter chips, showing the "Field Work" label (local file: /tmp/no-mistakes-evidence/01KYTA3WJYEVCJ2JP52G8K9XWF/web-lately-kind-chip.png)

Evidence: Before/after: same record generated by base vs target commit

base 1b9056d: \cvcourses{Y Y Y Y}{\textbf{Course} & \textbf{Programme / level} & \textbf{Key topics} & \textbf{Hours} \}{ target ce975ec: \cvcourses{Y Y Y Y}{\textbf{Course} & \textbf{Programme / Level} & \textbf{Key Topics} & \textbf{Hours} \}{ /cv/ provenance (target): ... with keys course, programme / level, key topics, hours — and those keys, in the order the file writes them, are also the columns of the same table in the printed CV.

Same record, two commits of scripts/build-cv-data.mjs.

Record (content/cv.yaml, teaching[].rows[] — plain lowercase keys, no capitals stored):
        course: GEOL 273
        programme / level: B.Sc. Geology
        key topics: Palaeoclimate archives; proxy calibration
        hours: "18"

cv/generated/cv-data.tex, base 1b9056d (first-character rule):
\cvcourses{Y Y Y Y}{\textbf{Course} & \textbf{Programme / level} & \textbf{Key topics} & \textbf{Hours} \\}{

cv/generated/cv-data.tex, target ce975ec (one shared per-word rule):
\cvcourses{Y Y Y Y}{\textbf{Course} & \textbf{Programme / Level} & \textbf{Key Topics} & \textbf{Hours} \\}{

/cv/ provenance line for the same table, target commit (the record key is still published verbatim):
content/cv.yaml — teaching[], 1 entry holding 3 courses in its rows[] table. ...
with keys course, programme / level, key topics, hours — and those keys, in the
order the file writes them, are also the columns of the same table in the printed CV.

Bundled example unchanged: cv/generated/cv-data.tex regenerates byte-identical to the
committed file, all three PDFs rebuilt with latexmk -xelatex, and all three
scripts/check-cv-baseline.sh comparisons pass silently (exit 0) against the untouched
data/cv-baseline/ recordings.

Pipeline

Updates from git push no-mistakes

✅ **intent** - passed

✅ No issues found.

✅ **Rebase** - passed

✅ No issues found.

🔧 **Review** - 3 issues found → auto-fixed (2) ✅

• ⚠️ scripts/build-cv-data.mjs:236 - tableHeader's new per-word rule uses the ASCII-only class /[A-Za-z0-9]+/g, so a non-ASCII letter ends a match and the remainder is capitalised as a new word: key 'français niveau' prints '\textbf{FrançAis Niveau}', 'kōwhai level' prints 'KōWhai Level'. The previous rule (k[0].toUpperCase() + k.slice(1)) left the tail untouched and rendered these correctly, so this is a regression for any adopter whose column key contains an accented or macronised letter — the bundled record already carries macrons in values (content/cv.yaml Kārearea, Ōtepoti). Fix: k.replace(/[\p{L}\p{N}]+/gu, capitalise); capitalise itself is already Unicode-safe.
• ⚠️ web/src/pages/cv.astro:75 - The motivating leak is only half closed. The same teaching rows: table is rendered on the website by label = (key) => key[0].toUpperCase() + key.slice(1) (used at cv.astro:202 for <th> and :210 for data-label) — the exact first-character rule this change removed from the generator. With the plain record key programme / level, the PDF now prints 'Programme / Level' but the site prints 'Programme / level', so an adopter who wants both to read correctly is still pushed to store the capital in the record key, which is the presentation-in-the-record failure the intent names. The intent scoped the change to scripts/build-cv-data.mjs, so this is a scope question rather than a defect in the code written: there is no module shared across scripts/ and web/ (AGENTS.md keeps the website reader boundary separate), so the earliest supported fix is applying the same per-word replace inside label, ideally with the Unicode class from the finding above.
• ℹ️ scripts/build-cv-data.mjs:224 - Acknowledged ceiling, reported because the intent asks for it: a key the single per-word rule cannot express is now level of study, which prints 'Level Of Study' where the old rule printed 'Level of study'. Joining-word lowercasing is explicitly out of scope; noting it so the tradeoff is on record rather than proposing a casing policy. Bundled example unaffected — every rows: key in content/cv.yaml is one lowercase ASCII word, so cv/generated/cv-data.tex is unchanged.

🔧 Fix: fix(cv): capitalise headings per word in any script, on both surfaces
1 warning still open:

• ⚠️ web/src/lib/announcements.ts:256 - The sweep updated this line's character class to /[^\p{L}\p{N}]+/gu but left the casing first-character-only: singular still returns stem.charAt(0).toUpperCase() + stem.slice(1). It is a record key becoming a label — the announcements kind, used at announcements.ts:360 and shown as the feed's filter vocabulary (lately/index.astro:28,155) — so a two-word section key field_work: now reads 'Field Work' as the printed section heading (sectionHeading, build-cv-data.mjs:287) and 'Field work' as the feed kind on the site: the same key, two casings, across the two surfaces. It also leaves the original leak reachable there — the only way to get 'Field Work' in the feed is to put the capital in the record key. Pre-existing rather than introduced (sectionHeading already capitalised every word via keyWords), and no bundled output changes because every section key in content/cv.yaml is one word, so this is a follow-up, not a blocker. Fix is return headingCase(stem) after the plural strip; flagged ask-user because it changes a user-visible label for multi-word keys and the author edited this exact line and stopped short, which may have been deliberate.

🔧 Fix: fix(web): capitalise every word of a feed kind label
✅ Re-checked - no issues remain.

✅ **Test** - passed

✅ No issues found.

• node --test scripts/build-cv-data.test.mjs — 46 pass, including the two new multi-word/non-ASCII header cases
• node --experimental-strip-types web/src/lib/cv.test.ts — includes the new headingCase assertions
• LEDGERPRESS_RECORD_ROOT=src/lib/fixtures/record node --experimental-strip-types web/src/lib/announcements.test.ts
• node scripts/build-cv-data.mjs then git status --porcelain — cv/generated/cv-data.tex regenerates byte-identical to the committed file
• latexmk -xelatex -cd cv/cv.tex (and short.tex, teaching.tex) on a pristine copy, then bash scripts/check-cv-baseline.sh for each of cv.pdf, short.pdf, teaching.pdf — all exit 0 with no baseline modified
• Adopter demo: copied the repo to /tmp evidence dir, renamed teaching row keys to programme / level, key topics, hours and section fieldwork to field_work, regenerated data and built cv.pdf — header renders "Programme / Level", "Key Topics", "Hours"
• Same demo record under astro dev: screenshotted /cv/ teaching table and its Inspect sources provenance line (with keys course, programme / level, key topics, hours), and /lately/ showing the "Field Work" kind label
• Clicked the /lately/ "Field Work" filter chip — filters to 2 items, slug still field-work, so display casing did not move any anchor or CSS rule
• Regenerated the same demo record with the base commit 1b9056d generator for a before/after: "Programme / level", "Key topics"

⚠️ **Document** - 1 info

• ℹ️ AGENTS.md - Judgment call, agreed with the author's stated decision: AGENTS.md was left unedited. The change introduces two copies of one casing rule across the build boundary (headingCase in scripts/build-cv-data.mjs and web/src/lib/cv-schema.ts), kept in agreement only by parallel assertions in scripts/build-cv-data.test.mjs and web/src/lib/cv.test.ts. Both call sites document the duplication and name each other, and AGENTS.md forbids restating a value that lives in another file, so no entry was added. If a third consumer of the rule ever appears, the invariant would earn a load-bearing-constraints line.

✅ **Lint** - passed

✅ No issues found.

✅ **Push** - passed

✅ No issues found.