docs(docs): V1-2.6 UX owner ratification recorded, independent checks

This commit is contained in:
bchanot committed 2026-11-05 19:06:48 +01:00
1 parent e10f167a6d
commit bd444d6906
6 files changed
+713 -273

No files matched your search

+262 -94
View File
@@ -1,156 +1,324 @@
# V1-2.6 UX redesign: technical decisions still needed
# V1-2.6 UX redesign: technical decisions
Status: PROPOSAL, 2026-11-05, revised after the independent review. Companion of `V1-2-6_UX_PLAN.md`; this is the
single table of what blocks what. The 11 UX choices and the direct SVG interactions validated by the owner on
2026-11-05 are not reopened. Each item is a technical question they leave open, with a recommendation.
Status: OWNER ANSWERS RECORDED, 2026-11-05 (brief « FAUNOSTEO — V1-2.6 UX — Ratification des décisions
techniques »). Companion of `V1-2-6_UX_PLAN.md`; this is the single table of what blocks what. The 11 UX choices and
the direct SVG interactions validated by the owner on 2026-11-05 are not reopened. Nothing here authorises code: U0
(docs) and every later sub-step start only on a new explicit authorisation.
## Owner decisions
| Id | Question | Blocks | Recommendation |
|---|---|---|---|
| T1 | Fallback if the wasm budget is exceeded | U5 and U6 only if the measure exceeds 600 000 B; U9 certainly | `wasm-opt -Oz` as a pinned build tool, then a small D-R13 raise; the PDF split stays a separate track |
| T2 | Doctrine for optional category keys, and the owner's agreement to widen the snapshot reader | U6 | route A, no version bump, `v1` convention kept, rules amended |
| T3 | Opacity semantics | U6 | integer percent 10-100, point twins and zone fills, constants when absent |
| T4 | Tier of the fixed-category rules; residual gap on a first version | U6 | `edit.rs` refusals + `sheet_report` issues (existing verdict code), round-withdrawal migration; residual gap accepted and covered by review |
| T5 | Fixed categories of a new sheet: ids, names, default colours | U6 | `cat-bone`, `cat-joint`, `cat-vertebra`; Os / Bones, Articulations / Joints, Vertèbres / Vertebrae; equine colours |
| T6 | Equine working copy conversion | V1-2.8 | proposal below, decided at V1-2.8 |
| T9 | Replacing a crop by its own drawing | U5 | explicit choice when the `viewBox` differs, positions kept when equal |
| T10 | Category of an automatic access; sheet without category | U5 | the Joints category after U6; before U6 the first category; none → box unchecked |
| T11 | Fate of the level-1 owner trial | U1 | skip it; trial the redesign at U2 and U8 |
| T12 | Common text: source, versions, wording, responsibility | U7 | versioned module; wording supplied and validated by the owner |
| T14 | Multi-tab notice | none (follow-up) | not in V1-2.6; track as a follow-up |
| T15 | Header « Annuler » next to form « Annuler » | none | keep the label; accessible name « Annuler la dernière modification » |
| T16 | Curve handle editing in V1-2.6 | U9 | defer to the lot that first lets a path enter a document |
| T17 | Vertebral regions entry when no category has the vertebra role | U6 | keep a « Régions vertébrales » entry in the Annotations panel in that case only |
| Id | Question | Owner answer (2026-11-05) | State | Blocks |
|---|---|---|---|---|
| T1 | Fallback if the wasm budget is exceeded | not asked this round; measuring rule recorded (LRN-118) | open, used only if a measure exceeds 600 000 B | U5, U6 if a measure exceeds |
| T2 | Optional category keys; widening of the snapshot reader | agreement in principle, conditional on the proofs | ratified with conditions (§ T2 and T3) | writing the keys (U6 write lock) |
| T3 | Opacity semantics | integer percent from 10 to 100 | ratified with T2's conditions | U6 |
| T4 | Fixed-category checks | no residual gap; checks in the core and at the server boundaries of save and publication; nothing written before they are live | ratified; design below; the two declared SQL checks (save non-regression, publication) to confirm | U6 |
| T5 | Equine history; categories of a new sheet | historical preservation mandatory; roles on a working copy only; conversion tested without mutating the originals | ratified for the history; new-sheet ids, EN names and colours still to ratify | U6 (defaults), V1-2.8 |
| T6 | Equine working copy conversion | not asked | open, decided at V1-2.8 under T4 and T5 | V1-2.8 |
| T9 | Replacing a crop view's drawing | explicit choice: keep positions, reposition, cancel; no approximate automatic transformation | ratified; design below | U5 |
| T10 | Automatic access to a detail view | system navigation type distinct from the visible taxonomy; schema checked first; never Os, Articulations or Vertèbres | principle ratified; internal category to confirm | automatic access (moved to U6) |
| T11 | Fate of the level-1 owner trial | approved: old-UX trials stopped; anomalies and untested items tracked; level 1 not declared validated | ratified | none |
| T12 | Common text | principle of one versioned copy kept; legal text, source and validation to be supplied | principle ratified; text missing | U7 |
| T14 | Multi-tab notice | not asked | open, follow-up | none |
| T15 | Header « Annuler » wording | not asked | open | U1 (accessible name and tooltip) |
| T16 | Curve handle editing | deferral only if the audit confirms it; exact geometry; no destructive conversion; identified future lot | ratified: deferred to the curves lot | none in V1-2.6; the database count before U6 or U8 reaches any environment beyond the throwaway sandbox |
| T17 | Vertebral regions entry without a vertebra role | not asked | open | U6 |
## Implementer decisions (recorded at U0, no owner question)
| Id | Decision | Reason |
|---|---|---|
| T7 | Strip `role` and `fill_opacity` from every envelope snapshot, at `lean_document` and at the export gate | the PDF needs neither; the only alternative breaks N-1 reading of new PDFs |
| T8 | Path ops (U9) compute on the written decimal digits; output precision = max(input precision, 2 decimals for a user move) | keeps « no other point moves » and exact round trips |
| T8 | Path ops (curves lot) compute on the written decimal digits; output precision = max(input precision, 2 decimals for a user move) | keeps « no other point moves » and exact round trips |
| T13 | No sub-branches: commit in place on `feature/v1-2-anatomical-sheets`, rollback by reverting a sub-step's commits | the gitflow lib branches only from develop and `finish` merges into develop |
| T18 | Handle cap per shape (constant fixed at U2 from the equine maxima with margin); above it, outline and a message | bounded DOM for oversized shapes saved directly |
| T19 | Sliders and arrow moves commit on release, Enter or blur | BDR-050; the 8 MiB history holds about 20 equine documents |
## T1 · Wasm budget fallback
## T1 · Wasm budget fallback (open)
- Facts: the gate measures Node `gzipSync` level 9 (`packages/core-wasm/tests/size.test.ts:12-25`): 592 143 B of
600 000 B, margin 7 857 B (measured 2026-11-05). Without curves the new core code is estimated at 4.5 to 13.5 KB.
The plan builds `annotations_in_rect` first and measures, then `remap_view`, then the U6 changes.
- Facts: the gate is `packages/core-wasm/tests/size.test.ts:12-25`: Node `gzipSync(bytes, { level: 9 })` on the
shipped build `packages/core-wasm/pkg/faunosteo_wasm_bg.wasm`, at most `COMPRESSED_BUDGET_BYTES = 600_000`.
Re-measured 2026-11-05 with that exact call: 1 569 864 B raw, 592 143 B, margin 7 857 B (build of 2026-11-04,
after the last core change `bbab0e6`). GNU `gzip -9` gives a different figure and is never used (LRN-118).
Without curves the new core code is estimated at 4.5 to 13.5 KB. The plan builds `annotations_in_rect` first and
measures, then `remap_view`, then the U6 changes.
- Options if a measure exceeds the budget:
- (a) `wasm-opt -Oz`: gain unknown; `wasm-opt` is not installed (`RUST.md:4054`), so it becomes a pinned tool in
`scripts/wasm-build.sh`, the builder image and the release workflow.
- (b) raise the budget (D-R13) by the measured excess plus a margin: no work; a slightly slower first load.
- (b) raise the budget (D-R13) by the measured excess plus a margin.
- (c) lighter URL check (about 23 KB): must keep `new URL()` equivalence.
- (d) PDF code in a lazily loaded second module (about 55-60 KB): touches R4, the import worker, the CSP proof,
trap recovery and the readers of user data; its own track with its own owner checkpoint, never a prerequisite.
- Rule: the budget is never raised silently. (b) is an owner decision taken on the measured figure; a sub-step whose
measure exceeds stops for T1.
- Recommendation: (a), then (b) if still short. (d) only as a separate decision.
## T2 · Doctrine for optional category keys
## T2 and T3 · Optional category keys and opacity (ratified with conditions)
- Facts: `.claude/rules/schemas.md:9-11` (« additive optional fields keep it ») against `.claude/rules/sheet-model.md`
(« Document schema change → bump … No bump during V1-2 (BLK-008) ») and BDR-068 (`v1` freezes construction and
validation). Route B (v3) is excluded by contract §12.3 (`V1-2_ANATOMICAL_SHEETS.md:1779-1781`) and needs a data
migration and a `save_sheet_draft` change. The snapshot reader (`assert_snapshot`, used by `archive/read.rs`) is a
reader of user data: CLAUDE.md requires the owner's agreement before widening it.
- Recommendation: route A as specified in `V1-2-6_UX_COMPATIBILITY.md` §1: an additive optional change of the v2
document, no version bump, the `sha256-rfc8785-validated-v1` construction unchanged (existing values proven equal
by tests); `sheet-model.md` amended to say that an optional key omitted when absent is not a version change, with
the proof list. Route C (reserved ids, no opacity) contradicts choice 5.
- Owner answer: agreement in principle, conditional on proofs. Two optional keys in the v2 schema, omitted when
absent, never serialised as `null` by default; fill opacity an integer percent from 10 to 100; a controlled
widening of the snapshot reader. The effects on fingerprints, Rust, SQL, PDF, worker, the N-1 reader, R3/R4 and
BDR-068 are demonstrated; the impossibility of returning to the old application after the keys are written is
documented; this is never called full backward compatibility.
- Conditions, all met before any document carrying a key is written outside a test (the U6 write lock, T4):
1. every proof of `V1-2-6_UX_COMPATIBILITY.md` §8 green and reviewed;
2. the one-way compatibility written down (below, and COMPATIBILITY §0 and §7);
3. the snapshot reader widened only as bounded below.
- Serialisation: `#[serde(default, deserialize_with = "js::present", skip_serializing_if = "Option::is_none")]` and
`ts(optional)`, the `LocalizedOptionalText` pattern (`schema.rs:162-180`). An absent key is omitted on output; an
explicit `null` is refused on input. The serde readers also accept a struct written as a positional array and an
enum value written as a one-key map (`RUST.md:3326-3346`): the U6 proofs cover these shapes for both keys. No `SHEET_DOCUMENT_SCHEMA_VERSION` bump.
- One-way compatibility, not full backward compatibility. Old to new holds: the new core reads, writes and
fingerprints every existing document as today. New to old does not: once a document carries a key, the
application, the worker and the core of any earlier commit refuse it (`invalid_document`,
`verification.integrity.document`). Returning to the old application after the first write is impossible:
published versions are immutable and migrations forward-only (COMPATIBILITY §7).
- Controlled widening of the snapshot reader (`assert_snapshot`, used by `archive/read.rs`). It follows from the
shared `AnnotationCategory` type and is bounded: it accepts exactly `role` and `fill_opacity` with the document
reader's types and range; any other unknown key, an explicit `null`, an unknown role or an out-of-range opacity
still gives `SnapshotRejection::Schema` with the report kept (the positional-array and one-key-map shapes above
included in the proofs); no version is dropped (envelope v1, Report v2 and v3,
sheet document v2); the frozen N-1 copies are never edited. Writers still strip both keys (T7), so no envelope the
product writes carries them.
- BDR-068. The `sha256-rfc8785-validated-v1` construction and the validation rules stay frozen: no new
`validate.rs` error (T4 rules live in `edit.rs` and `sheet_report`), and an absent key is never filled with a
default in the validated document the fingerprint hashes (the layer applies its constants at render time). Every
existing identity stays equal, pinned by tests; a document carrying a key gets a new fingerprint under the same
convention, like any content change. Aliases still come only from frozen historical documents.
- Opacity (T3): when present it drives the fill of point twins and of zones; when absent the current constants stay
(0.65 main and 0.5 detail for twins, 0.3 for zones), so existing renders are identical. Below 10 a marker becomes
invisible while the 4 px rules assume it visible. Detail triggers keep 0.06; the selected state keeps its accent
at 0.85 (`apps/app/src/styles/index.css:14`). Written first in `docs/ANNOTATION-RENDERING.md`. The layer writes it
as `fill-opacity`, which the PDF highlight overwrites: no PDF effect.
- Doctrine: `.claude/rules/sheet-model.md` amended at U0 (an optional key omitted when absent is not a version
change, with this proof list). Route B (schema v3, excluded by contract §12.3, `V1-2_ANATOMICAL_SHEETS.md:1779-1781`)
and route C (reserved ids without opacity, contradicts choice 5) stay excluded.
## T3 · Opacity semantics
## T4 · Fixed-category checks (ratified: no residual gap)
- Recommendation: integer percent from 10 to 100 (below 10 a marker becomes invisible on the drawing while the 4 px
rules assume it visible). When present it drives the fill of point twins and of zones alike; when absent the
current constants stay (0.65 main and 0.5 detail for twins, 0.3 for zones), so existing renders are identical.
Detail triggers keep 0.06. The selected state keeps its accent at 0.85 (`apps/app/src/styles/index.css:14`).
Written first in `docs/ANNOTATION-RENDERING.md`. The layer writes it as `fill-opacity`, which the PDF highlight
overwrites: no PDF effect.
- Owner answer: no residual gap allowed. The fixed-category checks are defined in the core and at the appropriate
server boundaries for save and publication; the new model is not written before the necessary checks are live;
a progressive rollout or a functional lock is used if needed. The earlier recommendation (issues relative to the
base, residual first-version gap accepted) is withdrawn.
- Property guaranteed: no review round opens and no version is published on a draft whose categories break the
fixed model, whoever wrote the draft (the editor, a direct `save_sheet_draft`, « Garder ma version », a tab
running an older bundle). Fixed model: each role `bone`, `joint`, `vertebra` carried by exactly one category,
and that category carries its fixed FR and EN names (T5).
## T4 · Tier of the fixed-category rules
| Tier | Check | Where | New code |
|---|---|---|---|
| Core, verdict | absolute rule on the draft, no base needed: missing role, duplicate role, wrong name on a role-bearing category; error issues of their own code family `category.fixed.*`, mapped to the existing `verification.validation.rules` | `sheet_report` on the draft (`verdict.rs:188`), mapping `verdict.rs:277-296` | Rust; no verdict code, no `validate.rs` error |
| Core, editor | giving a role writes the fixed FR and EN names in the same update; renaming a role-bearing category is allowed only to its fixed names; removing it, changing or clearing its role and a duplicate role are refused; `navigation` (T10 B) is never assignable by an edit; repair commands insert a missing fixed category, restore the fixed names and clear a duplicate role, so every state a direct save can write is repairable | `edit.rs` | Rust |
| Server, save | non-regression check (recommended): a save that removes or duplicates a role present in the stored draft, or renames its category away from the fixed names, is refused; drafts without roles (pre-U6, equine working copy) still save | U6 migration, `save_sheet_draft`, a second declared SQL check under O22 | SQL, pgTAP; owner confirmation |
| Server, verification | the worker computes the verdict with the core above; a round opens only on a passed verdict | `complete_sheet_verification` (`20261101100000_sheet_verification_jobs.sql:300-345`) | none |
| Server, core gate | a passed verdict computed by a core older than the U6 core is refused at completion, so an old worker (rollback, stale image, second instance) never opens a round | U6 migration, keyed on `verified_with` | SQL, pgTAP |
| Server, publication | a declared SQL check `private.sheet_fixed_categories_ok(content)` (each role once, fixed names) at approval and at publication, beside the existing status checks, so a worker defect cannot publish a draft outside the fixed model (the reason O22 keeps declared SQL checks, contract `V1-2_ANATOMICAL_SHEETS.md:1705-1711`, `20261028110000:146-149`); the existing preconditions stay (`verdict.missing`, `verdict.stale`, `:984-990`; approved round on the current revision, `:1028-1072`) | U6 migration, `decide_sheet_review`, `publish_sheet_version` | SQL, pgTAP; O22 amended at U0 |
| Server, rounds in flight | rounds in `verifying`, `open`, `approved` of draft versions withdrawn with the reason; the approved round linked to each published version stays approved | U6 migration (contract `V1-2_ANATOMICAL_SHEETS.md:945-949`) | SQL, pgTAP |
- Facts: edit refusals act only in the editor (`edit.rs:1-4`); a direct save is checked structurally only. A
`validate.rs` error would invalidate existing documents (fingerprint, snapshot gate). Issues added in
`sheet_report` (the `GEOMETRY_MISSING` pattern, `report.rs:17-19`) map to the existing
`verification.validation.rules` (`verdict.rs:277-296`), so no SQL code change. The contract requires a migration
withdrawing rounds in `verifying`, `open` or `approved` when the integrity rules change
(`V1-2_ANATOMICAL_SHEETS.md:945-949`).
- Options: (a) `edit.rs` refusals only; (b) (a) + report issues relative to the base (a role of the base must remain;
a role appears once) + the withdrawal migration.
- Residual in both: a first version has no base; a direct save, or « Garder ma version » from a tab running an older
bundle, can strip the roles before the first publication. It shows in the review of the `general` group.
- Recommendation: (b), consistent with O22, and the residual accepted.
- Save boundary: the earlier text left save fully permissive and cited BDR-043; the governing rule is O22 (contract
`V1-2_ANATOMICAL_SHEETS.md:1683-1711`), which allows a declared SQL check. An absolute check at save would lock the
equine working copy and every pre-U6 draft; the non-regression check above locks none of them and stops stripping
by a direct save, « Garder ma version » or an older tab at the source. Recommended; owner confirmation requested
for both declared SQL checks: the contract says today « No other rule is mirrored »
(`V1-2_ANATOMICAL_SHEETS.md:1711`), so O22 is amended at U0 only on that confirmation.
- Core gate detail: every core so far reports `faunosteo-core 0.1.0` (`verdict.rs:164-166`, workspace version
`Cargo.toml:10`), so the gate needs a marker that tells the U6 core apart: a core version bump (owner's call) or a
rules epoch added to `verified_with` (free text of at most 100 characters for SQL today). Chosen at U6, pinned by
pgTAP.
- Functional write lock: the U6 migration exposes the rules epoch (a read-only RPC); the application reads it and
writes no key and creates no sheet with the new model until the epoch is present: an application deployed too
early creates sheets through the plain v2 path (no categories, as today) and the repair adds the fixed ones later. Order SQL, worker, application (COMPATIBILITY §7), with SQL first
mandatory: only the orders that start with SQL fail closed. Worker before SQL leaves rounds already `approved` on
old verdicts publishable until the U6 migration lands; that is why SQL goes first. U6 stays
on the sandbox until the owner decides otherwise.
- Scope: the rule judges drafts submitted after U6, never a published version. The historical equine version has no
roles and stays as it is (T5); the release replay (contract §15) filters the `category.fixed.*` family, so it
checks published versions on reading, fingerprint, validation and drawings only. Consequences: the equine working
copy cannot pass verification until V1-2.8 converts it (T6), which is that lot's scope; the existing tests that
expect a passed verdict or an `open` round on a role-less document flip and move to converted in-memory copies
(`V1-2-6_UX_TESTS.md` U6), the seed itself unchanged.
## T5 · Fixed categories of a new sheet
## T5 · Equine history and the categories of a new sheet
- Recommendation: `sheet.empty` creates `cat-bone`, `cat-joint`, `cat-vertebra` (core-chosen, unlike the page's
`cat-<10 base32>` ids, `apps/app/src/editor/ids.ts:1-21`, so no collision); names FR « Os », « Articulations »,
« Vertèbres », EN « Bones », « Joints », « Vertebrae »; roles `bone`, `joint`, `vertebra`; default colours, shown
as « Couleur d'origine » (board 07 A), taken from the equine categories: `#a41e65`, `#ff9500`, `#e100ff`. Names stay
stored (old readers show them) and the edit rules refuse changing them.
- Owner answer: historical preservation mandatory. The historical equine versions keep their ids, names, colours,
particular annotations and fingerprints; any role adaptation applies to a working copy only, and the conversion is
tested without mutating the originals.
- Untouched: the five equine categories (`articulations`, `os`, `charnieres-vertebrales`, `zone-colonne`,
`cadrans-du-crane`) with their ids, names and colours; the three own colours (atm and ssb `#ff8800`, hyoid
`#d507c4`); every annotation; the fingerprint `12d78872…` and the alias `27c7383c…`; every file under
`supabase/seed/anatomical-sheets/equine/` (`pnpm r5:import-equine --check` clean); the published historical
version (immutable, `sheet.historical_read_only`).
- Test: a conversion of an in-memory copy of the seed draft document (roles on the T6 categories) asserts before and
after that the seed bytes, the fingerprint, the alias and every review digest are unchanged, and that the copy
differs only in the converted categories (`V1-2-6_UX_TESTS.md` U6).
- New-sheet defaults, still to ratify (blocks U6): ids `cat-bone`, `cat-joint`, `cat-vertebra`, chosen by the core
(suffixes of 4 to 8 characters, so never a page id `cat-<10 chars a-z2-7>`, `apps/app/src/editor/ids.ts:1-21`);
names FR « Os », « Articulations », « Vertèbres » (validated on board 07, `docs/design/v1-2-6/ARBITRAGES.md:60`),
EN « Bones », « Joints », « Vertebrae » (proposal); default colours `#a41e65`, `#ff9500`, `#e100ff`, copied from the
equine categories (copying mutates nothing) and shown as « Couleur d'origine » (board 07 A). Names stay stored, so
old readers show them.
## T6 · Equine working copy conversion (V1-2.8)
- Facts: five categories (`articulations` 12 annotations, `os` 2, `charnieres-vertebrales` 4, `zone-colonne` 5,
`cadrans-du-crane` 5). `os` and `articulations` already carry the fixed names; `zone-colonne` is « Segments
vertébraux / Vertebral segments ». Three own colours: atm and ssb `#ff8800`, hyoid `#d507c4`.
- Bound by T5 (working copy only, originals never mutated) and T4 (the copy passes verification only with the three
roles on categories carrying the fixed names).
- Proposal: roles on `os`, `articulations`, `zone-colonne`; `zone-colonne` renamed « Vertèbres / Vertebrae » in the
working copy only (id unchanged); the two other categories stay free; own colours kept (choice 11) unless the
business review decides otherwise. The historical sheet and the Git seed never change.
business review decides otherwise. Decided at V1-2.8.
## T9 · Replacing a crop by its own drawing
## T9 · Replacing a crop view's drawing (ratified)
- Facts: the seed head drawing has exactly the crop's `viewBox` and the full drawing's path data: keeping positions
is right for it. A remap is right only for a drawing re-based on another `viewBox`.
- Recommendation: V equal to R → positions kept, no question. V different → the dialog asks « Même repère : garder
les positions » or « Dessin recadré : transposer les annotations », no default selected; transposition is uniform
(`meet`), centred, annotations to confirm. Labels to confirm by the owner.
- Owner answer: an explicit choice when an SVG is replaced, with three actions: keep the positions, only if the
framing is compatible and the checks pass; reposition the annotations when the operation is safely supported;
cancel without any change. No approximate automatic transformation. The earlier recommendation (no question when
the `viewBox` equals the crop) is withdrawn.
- R (the view's crop rectangle) is written in the coordinates of the drawing the view shows today; V (the
candidate's `viewBox`) in the candidate's own. The core cannot check that the candidate was cut from the same
master or shows exactly region R: both actions rest on a premise the dialog states, and both fail closed.
- Two core calls (the cleaning output has no `viewBox`, `clean.rs:50-68`; the asset id exists only after
registration, `drawingUpload.ts:124-169`):
1. Offer, before any upload: input = the document, the view id and the cleaned candidate text; the core reads V
from the candidate root and runs the frame checks on it; output = the actions offered, each with its reason.
2. Batch, after the drawing is registered and held: input = the view id and the asset id; output = the
`sheet.edit` commands of the chosen action, committed as one undo entry under the geometry accept
(`issues.ts:123-175`). A refusal there leaves the document unchanged and a registered drawing that nothing
references, as today's refused attach (`drawingUpload.ts:195-196`).
- The dialog opens on every replacement of a crop view's drawing, before the upload, no action preselected:
- « Conserver les positions »: offered only when V equals R (the seed head drawing case); the dialog states the
premise « ce dessin utilise le repère de la vue parente ». Applied as today's attach (`attachToView`,
`drawingUpload.ts:184-192`); the view's annotations turn `to_confirm` (redrawn-view rule, `edit.rs:792-857`).
- « Repositionner les annotations »: offered only when V differs from R with the same aspect ratio (compared on
the values rounded to the editor's 0.01 grid; any other case refused, `remap.aspect`, never centred), no path in
the view (`remap.path`), no descendant crop (`remap.descendant_crop`), numbers in range. Premise stated:
« ce dessin montre exactement la zone de la vue ». One scale and one translation; numbers on the 0.01 grid; a
point pushed out of the frame by the rounding refuses the whole remap; every moved annotation turns `to_confirm`.
- « Annuler »: always offered; nothing is uploaded, the document and the history are untouched.
- An action that is not available stays visible, disabled, with its reason. A candidate without a `viewBox` gets
neither keep nor reposition, only « Annuler »: stricter than today's attach, which accepts it (stated in U5).
- Labels: the owner's French words above; EN « Keep positions », « Reposition annotations », « Cancel » proposed.
## T10 · Category of an automatic access
## T10 · Automatic access to a detail view (principle ratified)
- Recommendation: after U6, the Joints category (the equine access to « Tête » belongs to Articulations); before U6,
the first category of the sheet; a sheet without any category → the box is unchecked with the reason, the access
is placed later. Changeable in the properties.
- Owner answer: prefer a system navigation type distinct from the visible anatomical taxonomy; check the existing
schema constraints before choosing an internal id or category; never categorise this access arbitrarily as Os,
Articulations or Vertèbres. The earlier recommendation (the Joints category) is withdrawn.
- Constraints checked (2026-11-05):
- The navigation type exists: `detail_trigger` (`crates/core/src/domain/sheet/schema.rs:439-462`, « Navigation
only: a trigger opens a detail view and never carries findings »).
- It requires `category_id` (`schema.rs:449`), which must name an existing category:
`annotation.category.missing` is an error (`validate.rs:287-292`), and an error blocks the fingerprint, the
snapshot and publication.
- Snapshots copy annotations and categories whole (`lean_document`, `snapshot_rules.rs:141-165`): every trigger
reaches the PDF envelope and the frozen N-1 reader, whose schema requires `category_id`.
- Page ids are `<prefix>-<10 chars a-z2-7>` (`ids.ts:1-21`); a stable id has 1 to 64 characters (`js.rs:158`).
- The catalogue shows a selected annotation's category name (`ZonePanel.tsx:45-48`); the layer colours a trigger
from its own colour or its category (`layer.rs:142-150`).
- Options:
- (A) `category_id` optional on `detail_trigger`. A required key becomes optional: the ts-rs type is no longer
assignable to the N-1 type (`contracts.types.test.ts:53-61` fails by construction), and a trigger without a
category in a snapshot makes N-1 drop the snapshot unless triggers are stripped from snapshots, which touches
the readers of user data again.
- (B) one internal navigation category, created by the core, outside the visible taxonomy: id `cat-access` (a
6-character suffix, never a page id), role `navigation` (a fourth value of the T2 key), fixed names FR « Accès
aux vues détaillées », EN « Detail view access », colour `null`. Old readers and N-1 read it as an ordinary
category (the role is stripped from snapshots, T7). The editor lists it in no category panel and no picker;
only detail triggers use it (an edit refusal and a verdict issue for an observable annotation in it, and for a
second category with that role).
- Recommendation: (B). It needs the owner's confirmation: a fourth role value and an internal category go beyond
the T2 approval, and its names are visible to old readers.
- Effect on the plan: automatic access moves from U5 to U6 (it needs the role). U5 keeps the name proposal and the
overlap list. Until U6, « Ajouter un accès » (U4) asks the editor for the category as level 1 does: an explicit
choice by the editor, never an automatic one. Existing triggers keep their category (the equine « Tête » access
stays in `articulations`, T5); moving the working copy's access is a V1-2.8 question.
- Automatic creation (U6): only when the overlap list is empty; the view, its access and, if missing, the navigation
category inserted in one batch; otherwise the box is unchecked with the reason and the access is placed from the
view's menu.
## T11 · Fate of the level-1 owner trial
## T11 · Level-1 owner trial (ratified)
- Facts: level 1 is built and served for the trial (`V1-2-6_OWNER_TRIAL.md`); not tried yet. U1 and U2 replace its
tab structure and numeric journeys.
- Recommendation: skip the level-1 trial; keep its infrastructure (store, save, recovery, conflicts, upload, modes)
as the base; trial the redesign at U2 and U8. Level-1 criteria the redesign does not touch (C13 recovery, C14 undo
bound, C16 narrow read-only, C34, C41) stay in force and are re-tested.
- Owner answer: approved. Complementary trials of the old UX stop and the redesign preparation continues; the
anomalies and the untested items stay tracked; the old V1-2.6 is not declared fully validated.
- Level 1 (lots A to O, criteria C1 to C48): built and gated (`.claude/tasks/plans/2026-11-04-v1-2-6-checkpoint.md`),
never tried by the owner, NOT functionally validated. Its infrastructure (store, save, recovery, conflicts,
upload, modes) is the base of the redesign. The criteria the redesign does not touch (C13, C14, C16, C34, C41)
stay in force and are re-tested at U1 and U8.
- Tracked in TODO.md (level-1 line), from the checkpoint report, the execution ledger and the trial guide:
- owner manual trial: not run, stopped by this decision;
- a real second device on the LAN: not tested;
- editing specs at 375 px skipped (C16 waiver), and 8 admin journeys at 375 px skipped;
- worker-gated sandbox suite skipped (no worker by contract);
- the keyboard spec focuses programmatically (no Tab walk);
- free-path attach repair covered by automated tests only (no trial step);
- Safari iOS `beforeunload` unreliable (only « Enregistrer » protects work);
- LOW security items kept by the fresh auditor (checkpoint report, « Audits and reviews »: cleartext LAN trial
traffic including the TOTP enrolment key, allow list including tests and node_modules, first `hostname -I`
address as default trial host, template SQL in e2e helpers, trial gateway without a Host check, an unproven
sign-out copy-rewrite window);
- review minors m4, m6, m7, m8, m10 (execution ledger);
- path point correction (`.claude/tasks/plans/2026-11-04-v1-2-6-path-repair-core-proposal.md`, not built), now in
the curves lot (T16).
## T12 · Common text
## T12 · Common text (principle ratified, text missing)
- Recommendation: a versioned module in the application (id, version, FR, EN), changed only by a reviewed commit;
wording supplied and validated by the owner or a person the owner names; the editor shows « texte commun,
version n » and never presents it as veterinary or legal advice approved by the platform. U7 starts only when the
wording exists. Contract §7.4 amended at U0.
- Owner answer: the principle of one versioned copy is kept; the final legal text, its source and its validation are
still to be supplied; U7 delivers no invented or unapproved normative text.
- Design: a versioned module in the application (id, version, FR, EN), changed only by a reviewed commit; inserted
once into empty fields; the editor shows « texte commun, version n » and never presents it as veterinary or legal
advice approved by the platform.
- U7 is blocked until the owner supplies the FR and EN text, its source, and who validated it and when. The commit
that adds the text cites that record. Until then no such wording is written in the application, the dictionaries
or the docs; U7's tests use a fixture text marked as a test value, kept under `tests/`, never shipped.
- Contract §7.4 amended at U0 to state this, without any normative wording.
## T14 · Multi-tab notice
## T14 · Multi-tab notice (open)
- Facts: no multi-tab detection exists; the server revision check and the conflict dialog prevent silent loss; the
boards show no such notice.
- Recommendation: out of V1-2.6, tracked as a follow-up. If kept later: heartbeat-based (a crashed tab never says
goodbye), wording consistent with « Garder ma version ».
## T15 · « Annuler » wording
## T15 · « Annuler » wording (open, needed by U1)
- Recommendation: keep the validated label; give the header button the accessible name and tooltip « Annuler la
dernière modification » / « Undo the last change ».
## T16 · Curve handle editing
## T16 · Curve handle editing (ratified: deferred to the curves lot)
- Facts: no document of the project holds a path; no product path creates one; the validated guarantee « aucune
courbe convertie » is held by display-only paths and the C45 refusal; board 05 shows curves as « exemple d'une
autre fiche ». The three path ops are the plan's heaviest work (610 to 1 020 lines, 10 to 21 KB) and need a
certain size lever and a round-withdrawal migration.
- Recommendation: defer U9 to the lot that first lets a path into a document (package import, zones from drawing
shapes, or a curve drawing tool), keep its specification (`V1-2-6_UX_PLAN.md` §2 D) and keep C45 and the closed
`remap.path` refusal meanwhile.
- Owner answer: defer the curve handle editing only if the audit confirms that no existing curved annotation and no
journey available today needs it; keep geometries exactly; prevent any destructive conversion; record full curve
editing in an identified future lot.
- Audit (2026-11-05, at `e10f167`):
- Tracked documents: the seed draft holds 22 circles, 6 polygons, 0 paths; the legacy import 21 circles, 6
polygons, 0 paths (`supabase/seed/anatomical-sheets/equine/{draft,legacy-import}/document.json`). The historical
sheet and its working copy derive from them.
- Journeys that write geometry: creation makes points and polygons only (`apps/app/src/editor/annotations/creation.ts:90-97`);
the editable geometries are circle and polygon (`geometryEdit.ts:9`, `34`); package import, zones from the
drawing's shapes and curve drawing are level 2 (contract §8.3, `V1-2_ANATOMICAL_SHEETS.md:1328-1338`).
- Journeys that meet a path: display by the core layer; the attach repair returns a path unchanged and keeps the
attach refused, document unchanged byte for byte (`repair.ts:55-69`, blocked ids `95-100`, C45, `editor-geometry.spec.ts:317`);
« Remplacer par un polygone » (`PropertiesPanel.tsx:138`, contract §8.3).
- Not inspected: the database contents of DEV and of the sandbox (docs-only brief, no query run). A path there can
only come from a direct RPC save or a test fixture. Before U6 or U8 reaches any environment beyond the throwaway sandbox, a
read-only count of path geometries in its drafts and published versions is run, by the owner or under an explicit authorisation; a
non-zero count reopens T16.
- Conclusion: the condition holds for every tracked document and every product journey; the deferral is adopted.
- Guarantee until the curves lot:
- a stored path keeps its `d` text byte for byte; no TypeScript parses it (C45); no operation moves, scales,
remaps, simplifies, samples or converts it;
- every operation that would have to change a path is refused with the document unchanged: attach repair (C45),
remap (`remap.path`), the U2 handle layer and zone move (a path gets no handle and no move);
- « Remplacer par un polygone » is the one way a path leaves a document: the editor draws a new polygon by hand,
one undo entry, published versions untouched. It replaces the curve by the editor's choice; nothing converts
it. It stays as ratified in contract §8.3 unless the owner removes it.
- Future lot « Courbes » (ex-U9): specification in `V1-2-6_UX_PLAN.md` §2 D, tests in `V1-2-6_UX_TESTS.md`
(« Curves lot »), with the path point-correction proposal. Recorded in TODO.md with its dependency rule: no
feature that lets a path enter a document (package import, zones from the drawing's shapes, a curve drawing tool)
ships before or without it. It needs a size lever (T1) and a round-withdrawal migration.
## T17 · Vertebral regions entry without a vertebra role
## T17 · Vertebral regions entry without a vertebra role (open, needed by U6)
- Facts: board 09 opens the regions from the Vertèbres category; the equine working copy has no role before V1-2.8.
- Recommendation: when no category carries the vertebra role, the Annotations panel keeps a « Régions