attributeByName lowercased the store attribute name but compared it to the
raw lookup string, so 'figure' !== 'Figure' and Edge/Figure/Grade/Thickness/
Moisture all matched nothing - the Figure chip group (and the other attribute
pickers) rendered empty, so 'Mark Ready' could never enable against the live
store. Lowercase both sides so the synced attributes resolve.
The axis-preview draw reactive only depended on the image being complete, so
it fired before approvedMask was populated (canvas stayed at the default
300x150 and nothing drew). Add cap.approvedMask to the reactive deps so the
sheet-removed photo + tape line render once the frozen mask is present.
The axis step rendered only text and buttons - no image, so the owner had
nothing to confirm the length axis against and 'handing to step 4' felt like a
dead end. Render the calibration photo with the sheet checkerboarded (using the
frozen approvedMask) and draw the length-axis tape line in red from
axis.tapeLine.start to .end (source-pixel coords scaled to the canvas), so the
BG-removed slab and its axis are visible before approving.
Small isolated artifacts (e.g. a room corner the photo caught past the black
sheet) survived the flood-fill as separate slab islands and carried into the
rest of the workflow. After the border flood-fill, label the 4-connected slab
components and keep only the largest, with a 0.5%-of-frame area floor that
also drops speckles. Adds keepLargestComponent + unit tests proving an
isolated blob and a corner artifact are dropped while the main slab is kept.
Rapid sensitivity drags triggered a flood of async mask runs; an in-flight
run could settle after the owner pressed Approve and clobber the state, so
the Approve click read a snapshot with overlay:null and early-returned.
Debounce 500ms after the user stops adjusting so the mask settles before any
interaction and the click reads a stable state.
Key the auto-run off a controlKey (sheet:sensitivity) so it fires on mask
entry and on each sheet/sensitivity edit, but never re-fires when approveMask
settles the overlay - preventing both the empty-preview bug and an infinite
settle loop.
The mask screen never populated its overlay: pickCalibration set overlay:null
and the only thing that computed it (approveMask) was gated behind the
Approve button, which is disabled until overlay.slabPx>0 - a chicken-and-egg,
so the preview stayed empty and Approve was permanently disabled. Auto-run
approveMask() whenever the mask screen is active (and on sheet/sensitivity
change) so the overlay and preview render immediately.
The border flood-fill on a full-res 2256x4000 (9M px) photo was too slow in
JS and the async mask pass timed out into the catch branch (overlay stayed
null, 'Mask settled' never rendered). Downscale to <=1200px for the fill,
then nearest-neighbor upsample the slab bitset back to full size so
minAreaRect/slabBounds stay in source-pixel coordinates.
The green/black-sheet luminance threshold could not separate the slab from
the 'black' sheet, which in real photos reads at lum ~50 and overlaps the
slab's bark/shadow tones (measured on the user's samples: 0% green, 60-80%
dark, no brightness gap) - it removed 100% of the image. Replace the mask
pass with connectivity-based background removal: sample the sheet tone from
the border, flood-fill from the border marking sheet-connected pixels as
background, keep the connected remainder as the slab. Robust to tonal
overlap. Adds bgRemoval.ts (removeBackground + sampleSheetTone) with unit
tests proving a sheet-toned slab pixel survives via enclosure.
The mask screen showed only controls and a pixel-count line - no photo, so
the owner had nothing to adjust the sheet/sensitivity against. Draw the
calibration photo to a canvas and paint removed (non-slab) pixels as a
light-gray checkerboard (CAPTURE-45/50), redrawing when the overlay settles.
Verified in Chromium: mask step now shows the photo with the checkerboard.
The LAN dev URL is a non-secure, sometimes restricted browser where
createImageBitmap may be undefined; the pick-calibration -> mask decode
would then reject. Fall back to a canvas + Image decode returning an
ImageBitmap. Contract unchanged (still returns ImageBitmap).
The LAN dev URL (http://192.168.1.202) is not a secure context, so
crypto.randomUUID is undefined and adding a photo threw
'TypeError: crypto.randomUUID is not a function'. Add newId() (randomUUID
when available, else crypto.getRandomValues UUIDv4) and use it in draft()
and PhotoImport. Verified in Chromium.
`session.state` is a Svelte store; `$: state = session.state` bound the store
object (not its value), so state.capture was undefined and cap.k threw on
mount. Bind a local const to the store and use the $ prefix ($stateStore),
and use get() in App.onPhotosReady. Verified in Chromium: hero -> batch ->
sample uploads show thumbs.
The hero button called flow.set('capture') which renders the path-chooser
branch, so it stayed on the chooser and 'Choose photos' never appeared.
Point it at 'captureSession' so the hero action lands directly in the batch
screen. Verified in Chromium: hero button -> batch screen -> sample uploads
show thumbs.
`createCaptureSession()` returns {state: Store, ...}; Svelte's $ prefix
calls .subscribe on the object, not the store, so $captureSession.state
threw 'subscribe is not a function' on App mount → black page. Reference the
store directly (auto-subscribes): captureSession.state / session.state /
session.lineup(). Verified: tsc clean, build green, 150/150 tests, and the
live app renders in Chromium.
Add global crop_fill percent (default 90, 1-100) as a settings key with
FastAPI 4xx validation. Wires the value through to the inventory PNG
encode (non-retro: stored PNGs untouched). Adds a throwaway
TapePrototype.svelte behind ?tape-prototype for the #99 design sandbox.
Gates: check_requirements PASS, check_coverage PASS, settings tests 8/8.
Morning merge kept Review call sites (nextPrimary, keptInOrder, roleLabel, …)
and dropped the imports, so hydrate threw ReferenceError and the UI showed
"Could not load the slab." Live UAT redeployed; sample data preserved.
The merge auto-kept the branch's canCreateTitle reactive (speciesId) which is
undeclared in the merged file (main's canonical var is speciesSel) → ReferenceError
on Review mount → black screen. Also dropped the branch's duplicate gradeId (main
uses gradeSel) and repointed hydrate/patch to gradeSel. Live UAT verified: Review
now mounts and renders 'Review to Ready' with photo strip + Create buttons.
Reconciles two independent lines of work that both touched Review.svelte,
MultiSelect.svelte, and the thickness band:
- main's PR #97 (client-side thicknessBand.ts + MultiSelect names/commit +
syncSelects DOM race fix + box-sizing CSS)
- work/merge-89-90 (server-side thickness band in thickness_band.py,
Create-string #89, metadata_locked dropped, photo strip, wood primary)
Canonical resolution: the band is SERVER-side (hidden from UI, sent to Woo
on publish), so the client-side thicknessBand.ts + read-only band display
are removed from Review.svelte; PR #97's syncSelects() and box-sizing CSS
are preserved (orthogonal fixes). MultiSelect takes selectedSummary from
woodPrimary.ts (names in closed state) + the ms-count testid + onDestroy
timer cleanup; duplicate close()/summary definitions dropped.
Gates: backend 117/117, unit 88/88, tsc clean, e2e 6/6 (m3-create +
photo-strip).
Drop the metadata_locked column + "Save and lock" UI end-to-end. The
review endpoint now builds the deterministic title/short_description on
every call; the owner's Create title / Create short desc buttons fire
the build and write the result (force-overwrite contract). commitReview
auto-fills only empty fields (no owner-clobber) and folds the
deterministic copy into its single PUT (kills the client_rev 409
cascade).
Taken wholesale from the branch: photo strip, wood primary/secondary,
server-side thickness band (hidden from UI, passed to Woo on publish),
and fixes #80/#82/#83/#84.
- backend: remove metadata_locked from schema.sql, db sentinels,
slabs.py client fields/exposure, review.py lock action; review
always computes title/short_desc
- frontend: slabs.ts + Review.svelte drop the lock model, add
Create-string workflow + wood primary + photo strip
- e2e: m3-create + review-photo-strip rewritten for the Create-string
contract; thickness-band test asserts absence (per #90)
- deploy: recreate LAN binds (override-m3.yml was absent); verify
metadata_locked gone, review builds title, PUT persists
Gates: backend 117/117, unit 88/88, tsc clean, e2e 6/6 (m3-create +
photo-strip). Live UAT verified at 192.168.1.202:10881.
#88 — thickness band is now display-only, derived deterministically from
thickness_in against the synced Woo thickness terms (TV-9 / QUALITY-95).
Replaces the <select> with a read-only div. New pure function
deriveThicknessTermId + 6 unit tests. Docs (UAT-PRACTICE, contracts/UX)
updated to reflect the band is not an owner control.
#90 — MultiSelect closed state now shows the selected option names
(truncated to 2 + +N) instead of just a count. onCommit prop wired so
multi-select changes persist + re-review immediately (250ms debounce,
flush on Done, cleanup on destroy). Single-select binds use string vars
(speciesSel/edgeSel/gradeSel) to fix the number/string mismatch with
<option value={id}>; a tick()-deferred syncSelects() keeps the DOM
option in sync after programmatic writes (AI prefill, reload hydrate)
that Svelte's own select_option would otherwise reset.
The mandatory clean-cal and dimension-overlay photos now show a padlock glyph
in their keep box and an amber REQUIRED tag at the top of the card, so the
owner can see at a glance which photos cannot be removed (vs. optional batch
photos, which keep a plain checkmark).
REVIEW-5 requires both the clean calibration photo and its dimension-overlay
copy to ship. Their keep checkbox is now checked AND disabled (aria-disabled,
a lock glyph, cursor:default, tap is a no-op) so the owner can't remove them.
Only batch photos are removable. The submission minimum can no longer be
broken by the owner, so publish is never lineup-blocked by a missing required
photo.
The Capture→Review upload filtered out the overlay (role !== 'overlay') and
sent no role field, so the dimension-overlay copy was never stored and every
photo defaulted to role=batch. Review then fell back to a stored /processed
URL that had no dimensions drawn — the owner saw the overlay thumbnail
without its measurements. Now upload the full lineup with the correct role,
keeping the overlay's dimension-drawn blob as both original and processed.
Locks in the two live-bug fixes: the Primary (leftmost kept) photo's <img src>
must be a real URL (blob: for an in-memory capture blob) and never empty/
"object Blob"; and each strip card must carry the gray checkerboard
background matching the Capture lineup.
Bug #1: the first/Primary (calibration) photo rendered broken because its
in-memory capture Blob was assigned to <img src> directly (String(Blob)=
"object Blob"). resolvePhotoUrl() now mints a blob: object URL (revoked on
unload) and falls back to the stored photo endpoint when there is no blob.
Bug #2: strip thumbnails used a flat dark bg; now carry the gray checkerboard
from the Capture lineup (MeasurePanel) so transparent inventory PNGs read the
same on both screens.
Also harden missingMinimums to filter kept internally (pre-existing test 7
failed on the clean base because it passed un-kept items).
REVIEW-5 / PUBLISH-45: owner edits the stored photo lineup in Review — tap a
card to keep/remove, drag to reorder, leftmost kept photo = primary (Woo
cover). The kept set must contain a clean calibration + its dimension-overlay
copy; publish is blocked until it does.
- backend: role column on slab_photos (schema + idempotent _migrate ALTER),
role in photo CRUD + new DELETE endpoint, Woo image order by role-rank
(clean→overlay→batch) per PUBLISH-45
- frontend: Review photo strip (render from listPhotos, keep/remove,
HTML5 drag-reorder, dynamic PRIMARY badge from .primary class),
persistLineup() re-seq kept + drop removed before entering Publish
- tests: lineup.ts unit, e2e review-photo-strip.spec, backend role + order
Refs: #90, REVIEW-5, PUBLISH-45
The Review screen now has an explicit 'Lock & Build Title' step. Until the
owner locks, title/short_description stay empty (no 'ready' copy from an
unlocked AI draft). On lock the server validates the taxonomy, sets
metadata_locked=1, builds + persists the deterministic title/short_description
(owner prose wins), and returns the ready state.
- schema.sql: add metadata_locked INTEGER DEFAULT 0 to slabs
- db/__init__.py: bump _SCHEMA_SENTINELS for metadata_locked
- slabs.py: expose metadata_locked in GET; reject it in PUT (server-managed)
- review.py: POST /{id}/review action=lock path; regular review reports
stored (empty) copy until locked
- test_review.py: 8 tests covering lock gate, owner-prose-wins, stale taxonomy
- frontend/api/slabs.ts: Slab.metadata_locked, ReviewResult.locked,
ReviewInput.action, lockSlab() helper
- Review.svelte: lockMetadata() with correct commit-token ordering (bump AFTER
inner commit so the inner PUT isn't superseded), Lock button, locked banner
- e2e/m3-create.spec.ts: pare to single golden path (capture→review→lock→
ready→publish→done); mock respects optimistic concurrency (client_rev) so
stale PUTs 409 like the real backend
Ship locked expert Call 1 seed: order cats before edge, species=best
confidence, mill-count edge rules, Wild for cookies/burl, multi figure,
feature_tag_ids only (no free-text bags, no grade). LISTS keys match the
seed; feat- prefix stripped for the model. Gate and review UI apply
feature_tag_ids. Seed auto-activates on deploy when still seed-authored.
- frontend: uuid() helper (crypto.randomUUID -> getRandomValues) for plain-HTTP LAN UAT
- slabs.py: empty SKU at create -> NULL (no-SKU-yet is transient, filled in Review)
- review.py: ready-gate enforces REVIEW-45 (SKU 3-24 alnum+hyphen) in the middle tier
- DB CHECK stays as backstop only (referential/structural), per layering principle
Owner-directed, LAN/HTTP dev-test only. When the env flag is set,
GET /api/v1/settings returns decrypted secret values (string) instead of
{configured: bool}; the UI prefills and shows them in clear text. Default
off -> SETTINGS-75 masking and the existing tests stay intact. Revert
before any non-UAT deployment.
Scope .settings styles to the panel: light #eee text on the dark
#121212 surface for labels, and white inputs/selects/options with dark
text. Both pairs exceed WCAG 4.5:1. Add settings-contrast e2e as the
regression guard for the Settings form plus the sibling dev panels.
- test_exit_inference_off.py (QUALITY-20/120): full create->photo->review->publish
happy path with inference OFF, asserts zero inference_log rows and no raw codes/
tracebacks reach the owner; not-ready publish 422 with plain field-level message.
- test_integration_woo_mock.py (QUALITY-35): Woo mock covering success, error,
duplicate SKU, stale field (no raw code), and taxonomy anchor bump on change.
- m3-create.spec.ts (QUALITY-50): Playwright E2E — all M3 dev screens render,
Review/Publish controls load, Generate control visible only when enabled+reachable.
- deploy/uat_walk.sh (QUALITY-110): ready-to-run human walk for the real SLAB-UAT-*
draft (inference off) against the live store — NOT a mock.
Verified: pytest 61 passed; frontend unit 88 passed; e2e 4 passed; gates exit 0.
Ty