Root cause: src/pdf/fonts.ts embeds StandardFonts.Helvetica /
HelveticaBold / TimesRoman, which use WinAnsi encoding (Latin-1
only). PDF export threw `WinAnsi cannot encode "⚠" (0x26a0)` the
moment the text-height estimator measured a vet-flagged leaf
label (countSegmentedLines → widthOfTextAtSize).
Fix: replace the PDF prefix "⚠ " with WinAnsi-safe "(!) ". The
picker DOM keeps the original "⚠" since browsers render U+26A0
natively — the limitation is purely on the PDF side.
Regression test exercises embedFonts + widthOfTextAtSize on a
realistic prefixed label; would catch any future non-WinAnsi
marker re-introduction.
Follow-up (not in scope): switch the PDF body to embedded TTF
(Inter / Fraunces per CLAUDE.md spec) so the catalog can carry
arbitrary glyphs. CLAUDE.md states TTFs should be used; current
StandardFonts is a drift from intent.
155/155 tests green, typecheck clean.
Review surfaced 3 critical paths where hardcoded French leaked into
the EN UI:
- src/pdf/import.ts ImportError default message
- src/pdf/attachment.ts IncompatibleReportVersionError message
- src/app/routes/HomeRoute.tsx propagated err.message verbatim
Refactor:
- ImportError carries a typed `kind` discriminant
('invalid_pdf' | 'incompatible_version' | 'no_attachment') plus
optional `foundVersion`. Removes hardcoded localized strings.
- IncompatibleReportVersionError keeps `foundVersion` only; message
collapses to an internal sentinel.
- HomeRoute resolves the user-facing message via the active i18n
dict using `err.kind`, with `{version}` template substitution for
the incompatible-version case.
- 3 new dict keys (fr+en): home.import.error_invalid_pdf,
home.import.error_no_attachment, home.import.error_incompatible_version
- pdf.roundtrip v1 case asserts ImportError instance + kind +
foundVersion instead of grepping the message text.
154/154 tests pass, lint clean, typecheck clean, build OK.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
3 new cases:
- exports with locale=fr roundtrips
- exports with locale=en roundtrips
- both produce non-empty byte streams and the imported JSON payloads
match (locale lives outside the Report, so finding labelFr/EnSnapshot
are byte-identical between exports — only the static dict labels
rendered into the PDF differ, verified manually since pdf-lib font
subsets make byte-grep unreliable per LRN-016)
154/154 tests pass.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
ObservationSchema now carries a hierarchical Finding (zoneId + path +
labelFrSnapshot + labelEnSnapshot + requiresVetReferral?) plus a renamed
freeTextFindings list, replacing the flat pathologyIds /
freeTextPathologies pair. Snapshot labels keep already-exported reports
readable if the catalog is later edited.
ReportSchema.version bumped from 1 → 2. PDF import surfaces a distinct,
French-language error message on legacy v1 attachments instead of the
generic "not an osteo PDF" line, via a new IncompatibleReportVersionError
detected before the strict Zod parse runs.
Store seed and tests updated for the new shape. The standalone Pathology
Zod schema and its test are removed — the new findings tree is the single
source of truth for what an observation can carry.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>