The old buildZoneLabels wrapped the main-SVG fetch and every detail-SVG
fetch in a single try/catch. As soon as one source threw — for example
a 404 on a detail asset in a misconfigured deployment — the loop bailed
out and the labels Map stayed populated only with whatever the loop had
already set. That left every observation zone whose boneId belonged to
a failed source falling back to the raw, opaque "point-mpbyo7zq" string
in the rendered PDF.
The refactor:
- Loads every source independently in parallel.
- Wraps each load+parse in its own try/catch.
- Logs a console.warn on failure so the issue is observable in dev.
- Skips empty bone.fr values so the lookup never returns "" (which the
`??` fallback would treat as a value).
Exports buildZoneLabels so it can be unit-tested directly. The new
suite (4 cases) loads the real Horse + Horse_head SVG fixtures with a
mocked fetch and asserts:
- detail-popup boneIds resolve to their FR labels,
- main-SVG boneIds resolve to their FR labels,
- main labels survive a 404 on the detail asset,
- unknown species returns an empty map.
Co-Authored-By: Claude <noreply@anthropic.com>
When drilling through the findings catalog, intermediate categories
and sub-categories now carry a rounded accent-coloured badge that
counts every selected leaf in their subtree. Sibling nodes with no
descendant selections stay badge-free, so the practitioner can scan
a level and see at a glance which branches already hold findings
without re-opening each one.
Implementation is a small `pathStartsWith` helper plus a `countSelectedUnder`
lookup against the current observation's findings — restricted to the
picker's `zoneId` so two parallel pickers on the same card never
cross-contaminate counts.
5 new tests cover: count of 2 visible at root, the same 2 visible one
level deeper on the right sub-branch, no badge on un-selected siblings,
singular aria label for 1 selection, and zone isolation.
Co-Authored-By: Claude <noreply@anthropic.com>
ObservationCard previously bridged SVG zones to the JSON findings
catalog with a strict FR-name equality check. That broke as soon as the
JSON used "(ATM)" while the SVG carried "Articulation Temporo-
Mandibulaire (ATM)" or whenever either side picked up a minor casse /
accent / typo drift.
The new matcher (src/lib/zoneMatch.ts) normalizes both sides
aggressively (lowercase + NFD-strip accents + drop non-alphanumerics)
and scores each pairing across all four (bone-lang, zone-lang)
combinations:
0 exact normalized match
1 one normalized string is contained in the other
2+d Levenshtein distance d, capped at max(2, ceil(maxLen * 0.25))
∞ beyond tolerance — discarded
Preference order: English exact/substring first (no accents → fewer
normalization edge cases), then French exact/substring, then fuzzy
across both languages.
Real-world payoff for the current equine catalog:
bone EN "Sphenoid" → sphenoid (score 0)
bone EN "Spheno-Basilar Synchondrosis (SBS)" → ssb (score 0)
bone EN "Temporomandibular Joint (TMJ)" → atm (score 0)
bone FR "(ATM)" alone → atm (score 1)
bone EN "sphenoi" / "shenoide" (typo) → sphenoid (fuzzy)
bone EN "Hyoid", "Hip joint", "Stifle", ... → undefined (correct)
23 new unit tests cover normalization, Levenshtein, scorePair branches,
the full matchZoneFindings strategy ladder, and the absence of false
positives (Sphenoid ↛ SSB, Hyoid/Hanche/etc ↛ anything).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
SVG data-anno-id values are opaque generator strings (point-mpbyk...,
zone-mpbx...) that never match the human keys used in the findings
catalog (sphenoid, ssb, atm, ...). The picker therefore stayed hidden
whenever a zone was clicked.
ObservationCard now bridges the two via the FR label parsed from
data-anno-name-fr: it builds a name_fr → ZoneFindings index from the
loaded findings, looks up each observation zone by its Bone.fr, and
renders one picker per matched JSON tree (deduped on tree.id). Emitted
findings carry zoneId = tree.id (the JSON key), keeping the snapshot
contract intact.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Mark B0/B1/B2/B5 done on TODO. Journal 2026-05-20 entry with the three
commit SHAs and a per-phase recap. Three new design decisions captured:
JSON snake_case kept in TS (BDR-009), snapshot labels at selection time
(BDR-010), hard-fail v1 PDF with FR-language message (BDR-011). Two new
learnings: Zod recursive schemas via z.lazy + explicit ZodType<T>
(LRN-011), once-per-session UX modal via sessionStorage flag (LRN-012).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replaces the flat PathologyPicker with FindingsTreePicker — breadcrumb
navigation, drill into intermediate categories, toggle leaves, optional
vet-referral badge, optional description rendering. One picker per zone
that has a catalog entry; zones without a tree fall back to the free-text
input (renamed FreeTextFindingInput to match the new vocabulary).
EditorRoute now loads SpeciesFindings instead of a Pathology[] and shows
the scope_disclaimer once per session via a sessionStorage flag so the
practitioner is reminded of the legal perimeter the first time.
PDF observationPage renders findings via the snapshotted FR label and
prefixes ⚠ on vet-referral items, keeping the existing layout pipeline
intact (LRN-007/008/009 still apply).
Also drops the unused Toast and ZoneTooltip stubs, refreshes the
CLAUDE.md folder map, and adds a FindingsTreePicker test covering
navigation, leaf toggle, vet-referral flag, and back-out.
Co-Authored-By: Claude Opus 4.7 (1M context) <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>
Replaces the flat Pathology[] model with a hierarchical SpeciesFindings
tree (zones → categories → nested children → leaves) authored as JSON and
validated at boot. Equine catalog JSON is now wired through
SpeciesFindingsSchema.parse() so malformed data fails loudly instead of
corrupting an observation later.
Species manifest swaps `pathologies` loader for `findings`. JSON species
id renamed equin → equine to match the registry. Old equine pathology
test deleted; new findings test covers parse + FR/EN label coverage +
zone/category/leaf structure.
Foundational step of the v2 domain pivot; callers in observation/store/
UI/PDF migrated in follow-up commits.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Records the schema-v1.1 findings pivot decision, the inert JSON
drop, and the species.id reconciliation ("equin" vs "equine") to
resolve before the next loader migration runs.
Co-Authored-By: Claude <noreply@anthropic.com>
Adds the schema-v1.1 hierarchical findings JSON authored by the
domain expert at src/domain/species/equine/pathologies.json
(3-zone sample: sphenoid/ssb/atm — placeholder before full-body
authoring).
JSON is inert: not yet imported by the loader (still exports the
empty Pathology[]), so build and tests stay green.
TODO rewritten to sequence the V1 refactor:
- B0 findings schema + JSON loader with Zod parse at boot
- B1 ReportSchema v2 (observation.findings = path + snapshot)
- B2 FindingsTreePicker replacing flat PathologyPicker
- B3 i18n FR/EN site + PDF (locale store, dict, switcher)
- B4 PDF observationPage rendering findings
- B5 stub cleanup
- C full horse.json authoring + QA + docs + deploy
Known fixup: JSON species.id is "horse" but registry uses
"equine"; align JSON to "equine" at integration time.
Co-Authored-By: Claude <noreply@anthropic.com>
Scaffold-era checklist was stale: most Phase 2-5 items already
shipped (full editor flow, PDF roundtrip, Docker, identity edit
dialog, etc.). Rewrite to capture what's actually done vs what
remains for V1/MVP ship.
Remaining for V1: equine pathology DB (40-60 source-cited entries,
currently empty array), stub cleanup (Toast/ZoneTooltip unused,
ObservationEditor never created), PDF QA pass, Docker deploy
validation, doc finalization, Scaleway production deploy.
Post-V1 ideas (Capacitor mobile, accounts/cloud, RDV) captured
in Phase C so they don't get pulled into MVP by mistake.
Co-Authored-By: Claude <noreply@anthropic.com>
Three related editor polish bugs from PDF-import dogfooding.
(1) boneStates previously walked every observation, marking zones
in collapsed observations as 'linked' (faint) or 'multi' (bold) —
so deselected zones kept lingering color states. Now we read only
the active observation; everything else renders neutral.
(2) AnatomyCanvas applied highlight attributes once on mount; when
the SVG (esp. inside DetailModal) loaded after the boneStates
effect had already run, the freshly-mounted nodes had no
data-state. Track loadedSrc and re-run the highlight effect after
each SVG swap so detail-popup zones get colored on open.
(3) NewReportRoute had "← Accueil" on the right while EditorRoute
keeps it on the left — disorienting. Both headers now share the
same left-aligned back-button + clickable Brand pattern.
Deviation: hotfix rule caps at 2 files, this touches 3. Each fix
is mechanical and obvious; bundling is cheaper than three separate
patches.
Co-Authored-By: Claude <noreply@anthropic.com>
Multi-SVG observations previously stacked vertically and pushed
the zones/pathologies/severity/comment sections off the page.
Now SVGs are laid out 2 per row with cell width = (page - 2*margin
- col-gap)/2. Each cell scales to its own aspect, capped by a
per-row height budget that subtracts the reserved bottom (text
sections + bottom margin) from the available image area. Single-
SVG observations keep a full-width row.
Safety: the text-area reservation guarantees the sections always
fit; the cell-height cap prevents runaway when many details are
selected. A floor (IMAGE_MIN_CELL_H) avoids zero-height cells in
pathological cases.
Co-Authored-By: Claude <noreply@anthropic.com>
Header now exposes a back button and clickable brand title both
linking to /. New "Modifier les infos" button opens a dialog that
edits practitioner, animal, and owner via the existing forms with
cancel/save and required-name validation. Save also refreshes the
practitioner localStorage cache.
Co-Authored-By: Claude <noreply@anthropic.com>
Enables in-place edit of report identity blocks from the editor.
Mirrors existing setConclusion/setRecommendations pattern; no-op
when no report is loaded.
Co-Authored-By: Claude <noreply@anthropic.com>
Eyebrow OBSERVATION / SYNTHÈSE now baseline-anchored at y=height-95;
title baseline at y=height-122. 27pt separation between baselines so
the Times-22 ascender no longer reaches into the eyebrow line.
Severity next to the observation number is rendered as a filled pill
with white uppercase label. Background color encodes severity at a
glance:
- légère → sage green
- modérée → amber
- sévère → red
Co-Authored-By: Claude <noreply@anthropic.com>
Layout:
- Title baseline dropped from height-90 to height-115 so the size-22
ascender clears the header rule at y=760 with a real cushion.
- Image area starts further down to match.
Visual style:
- Eyebrow label (small bold accent) above each title: OBSERVATION /
SYNTHÈSE. Title uses Times 22; observation page shows the index
as a zero-padded number (01, 02, …) with severity as a muted
inline note next to it.
- 2pt accent rule (48pt wide) under each title.
- Section labels (ZONES, PATHOLOGIES, SÉVÉRITÉ, COMMENTAIRE,
CONCLUSION, RECOMMANDATIONS, PROCHAIN RENDEZ-VOUS) rendered as
small bold uppercase accent labels with the value on the next
line in regular body.
- Image area gets a "VUE GÉNÉRALE" / "VUE DÉTAIL — <id>" eyebrow.
Colors are inlined as rgb tokens (ink, muted, accent terracotta).
No new assets, only the three Standard fonts already embedded.
Co-Authored-By: Claude <noreply@anthropic.com>
Four coupled PDF rendering bugs:
1. Title "Observation #N" and "Conclusion" overlapped the header rule
(y = height - 80; rule at y = 760; size-16 ascender ate the line).
Lowered both titles to y = height - 90.
2. Long zones lines visually wrapped via maxWidth but the cursor only
advanced one slot per array entry, so pathologies/sévérité/commentaire
collided with the wrapped tail. Added wrapLines + per-line advance.
3. The rasterized anatomy still showed unselected zones (purple spine
fills, orange head trigger) because colorizedSvgString only mutated
selected elements. Non-selected annotated elements are now hidden via
display="none" and every data-anno-* attribute is stripped from the
serialized output (data-anno are runtime-only hints).
4. Main SVG was drawn even when every zone lived in a detail popup.
Render it only when at least one zone has no detailSourceId.
Layout constants extracted at the top of observationPage.ts.
Co-Authored-By: Claude <noreply@anthropic.com>
Detail bones were cleared on popup close, so zone labels reverted to
raw ids (point-mpbynls5✕) once the head modal was dismissed. Now the
detailBones map accumulates across opens, and a merged allBones map
(main ∪ detail) is passed to ObservationList so describeZone resolves
the FR label every time.
Co-Authored-By: Claude <noreply@anthropic.com>
When opened from a spine zone, the picker now lands on the matching
region tab instead of defaulting to Cervicales every time. Inference is
substring-based on the FR label first, EN as fallback (cervic→C,
thorac/dors→T, lombair/lumbar→L, sacral/sacre→S, caudal/coccyg→Cd),
case- and accent-insensitive. An existing selection still wins over
the inferred region so re-opening lands where the user left off.
Co-Authored-By: Claude <noreply@anthropic.com>
Three coupled bugs in zone interaction:
1. Clicks inside the head detail popup were ignored — EditorRoute
looked up the clicked id in the main-SVG bones map only, so
detail-only ids (point-mpbykomv et al.) returned undefined.
2. Spine zones did not open VertebraPicker on click; the picker was
only reachable via the per-zone "Vertèbres…" button.
3. The new SVG hint data-anno-special="vertebral-precision" was
present in Horse.svg but unread by the parser/UI.
Parser now reads data-anno-special into Bone.special. EditorRoute
parses the detail SVG on open and falls back to detailBones for id
lookups. When bone.special === 'vertebral-precision', clicking adds
the zone (if absent) and opens VertebraPicker for that zone index;
toggle-off via main click is disabled for these zones (use ✕). The
"Vertèbres…" button in ObservationCard is now shown only on zones
that support it. Convention doc updated.
Co-Authored-By: Claude <noreply@anthropic.com>
User asked that the app interact with anything carrying data-anno-*
attributes without hardcoding any specific category, id, or name. Future
SVGs may introduce new categories, and the existing SVGs may be updated.
Changes:
- `BoneCategory` is now `type BoneCategory = string`. The closed union
is gone; the parser accepts any non-empty value. New categories don't
require code changes.
- `VALID_CATEGORIES` and `isValidCategory()` removed from
svgConvention.ts. `validateBoneElement()` now only enforces presence
of `data-anno-id` and `data-anno-category`.
- EditorRoute no longer auto-triggers the vertebra picker when the
clicked annotation is `zone-colonne`. Any zone is added the same way
to the active observation regardless of category.
- ObservationCard shows a "Vertèbres…" link on every zone badge (was
conditional on `category === 'zone-colonne'`). The praticien picks
vertebrae only when relevant — the UI no longer knows or cares which
zones are spinal.
- Tests rewritten: removed the `isValidCategory` enum assertions and
the "rejects unknown category" branch. New test demonstrates that
parseSvgBones accepts a completely novel category string.
90/90 tests pass, lint clean, build clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User dropped new Horse.svg / Horse_head.svg that follow a simpler
convention:
- Categories renamed to longer human strings:
articulation → articulations
charniere → charni-res-vert-brales
spine-zone → zone-colonne
cadran-crane → cadrans-du-cr-ne
head, os-overlay → dropped (head zone is now category="os" + data-anno-opens-detail)
os → unchanged
- No more hitbox / visible class split; multiple elements may still share
the same data-anno-id but they all carry the same role.
- Trigger-only zones (`data-anno-opens-detail="head"`) may omit
`data-anno-name-fr/en`.
Changes:
- `Bone` now exposes a single `els: Element[]` instead of `hitboxEls +
visibleEls`.
- `parseSvgBones` accepts annotations without fr/en labels (defaults to
empty string) but still rejects unknown or missing categories.
- The "interactive vs decorative" filter is gone — every parsed
annotation is interactive, since the new SVGs only annotate the
clickable zones.
- Vertebra-picker trigger: `bone.category === 'zone-colonne'`
(EditorRoute + ObservationCard).
- PDF colorisation in `observationPage.colorizedSvgString` writes the
accent fill to every `els[]` entry (no more hitbox dual-pass).
- Dead `src/components/observation/ObservationEditor.tsx` removed (its
body lives inside ObservationCard since the accordion refactor).
- Tests rewritten for the new vocabulary; synthetic fixture uses the new
categories, real-SVG test asserts 18 + 9 unique zones.
92/92 tests pass, lint clean, build clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three issues from manual QA:
1. UX restructure — observations are now an accordion list. Each card has
a header that's always visible (title + chips + chevron + action
buttons) and a body that expands inline showing zones / pathologies /
severity / comment. Click the header (or the chevron) to toggle.
ObservationEditor has been merged into ObservationCard so each card
owns its own details. Only one observation is expanded at a time; the
chevron on the active card collapses everything. EditorRoute uses
activeObservationIndex === -1 to represent "all collapsed".
2. App used to start with 2 observations instead of 1. Root cause:
EditorRoute had a "seed first observation" useEffect that fired twice
under React 19 StrictMode in dev, both invocations seeing length=0 and
queueing addObservation(). Fixed by moving the seed into the store's
initReport (synchronous, one-shot) and removing the useEffect entirely.
Store tests updated to assert the seed behaviour (90/90 pass).
3. PDF export still failed with "The input is not a PNG file!" — canvg
was returning bytes that pdf-lib rejected at embedPng(). Switched
rasterizeSvg to the browser-native pipeline: serialize the SVG into a
Blob URL, load it with an Image element, drawImage onto a 2D canvas,
then toBlob → PNG bytes (with a toDataURL fallback). Validates the
PNG signature before returning. Tree-shaking dropped canvg from the
bundle (979 kB → 874 kB JS).
Tests updated for the new initReport behaviour and the new svgToPng
pipeline (Image mock + dataURL fallback case). lint + typecheck + build
all clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Five fixes from manual QA:
1. ObservationCard — clicking anywhere on the card now selects it.
The wrapping button only covered the text rows; empty space was dead.
The whole card is now role="button" with keyboard (Enter/Space) support;
inner action buttons (↑ ↓ 🗑) stop propagation so they don't double-fire.
This was also the root cause of "details look stuck" — the active
observation simply wasn't being switched.
2. NextAppointmentPicker — when value === undefined the "Date" radio was
still showing as checked because `kind` defaulted to 'date'. The three
radios now read from the raw `value?.kind` (which can be undefined),
so "Aucun" is the only checked option when no next RDV is set.
3. svgToPng.rasterizeSvg — `canvas.toBlob` can return null in some
browsers when the canvas state is unusual; the call now falls back to
`canvas.toDataURL('image/png')` and decodes the base64 payload. Also
paints a white background before canvg renders (so the PDF embed is
opaque) and clamps width/height to >= 1px to avoid zero-area canvases.
canvg errors are caught + logged instead of propagating, so the export
still produces a (possibly blank) page rather than throwing.
4. svgConvention.parseSvgBones — clickable bones are now distinguished
from decorative ones. Interactive = has a hitbox element OR category
is spine-zone OR head. Decorative bones (teeth, malar, mandible,
nasal-bone, facial-crest, individual vertebrae paths…) get
`pointer-events: none` applied to their elements and are excluded from
the returned map. Result: in Horse_head.svg, frontal/temporal/parietal/
sphenoid/occipital/atm/ssb/c0-c1/hyoid become clickable as intended,
while teeth / malar / mandible no longer intercept the clicks.
5. (covered by #1) — the per-observation details panel always read the
active observation; the bug was the click handler scope. Confirmed by
switching observations: the editor section now updates immediately.
88 tests still pass, lint clean, build clean.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
The app is open to any praticien who wants to use it (not 'single-praticien'),
and the architecture is multi-species — equine ships first, dog/cat/goat/cow/
pig/rabbit/donkey/sheep come as pure data drops following the same convention.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Three impactful fixes from the post-impl code review:
1. EditorRoute.onExport now catches PDF generation failures, logs to
console, and alerts the praticien in French — previously errors were
swallowed by try/finally with only `setExporting(false)`.
2. VertebraPicker onConfirm reads the active observation from the store
at confirm time instead of closing over the render-scope `activeObs`.
Prevents mutating a stale observation if the praticien switches active
observation while the picker is open.
3. readPersistedDraft validates the parsed JSON with ReportSchema instead
of a raw `as Report` cast. Corrupted drafts now return null cleanly
rather than poisoning the store.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Replace the placeholder with a full 7-step walkthrough: SVG convention,
file placement under public/, species manifest at src/domain/species/<id>/,
registry wiring, vertebral formula, pathology seed (empty allowed in v1),
and the two mirror tests (registry + real-SVG parse + pathologies schema).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Capture the plan §23.2 walkthrough as an unticked checklist with
deliberate TODO markers. Automated browser QA was out of scope for the
/init-project session; this file is the handoff for a manual pnpm dev
verification before delivery. Includes the matching automated coverage
table so reviewers know what is already enforced by CI.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Escape French apostrophes/quotes in JSX text per react/no-unescaped-entities
and replace the report?.observations.length dependency with a derived
observationsCount value so react-hooks/exhaustive-deps is satisfied without
re-firing on every report change.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Wire index.html to actually load /env.js before the Vite bundle so the
runtime config injected by docker/entrypoint.sh takes effect. Add a
public/env.js stub so local dev and static builds don't 404 — the Docker
container overwrites it via the entrypoint.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>