The user-facing goal status is maintained in STATUS.md. Update it when implementation, validation, blockers or next priorities change; keep this file as the detailed backlog and evidence history.
Branch fix/handoff-zero-length-walls (PR #120). A production check of link
sharing from the Storage cutover (October 16, 16:30 UTC) to October 20, 08:00
UTC counted 785 accepted shares and 23 refusals (422), with no 429 or 5xx on
/api/handoffs. Every refusal logged walls[N].dimensions must contain valid
positive width/height and nonnegative depth. They came from 9 iPhone clients
(IP and user agent); 8 sent no later successful share, and two tried 6 times
each. 422s at 5–25 a day were already logged October 9–15, before the server
recorded why it refused a capture (October 16).
Cause, reproduced off-device by compiling the iPhone app’s own model, drag and
export code (openplan3d-ios 09eaa13): the editor can leave a wall of length 0,
which PlanDocument+Export.swift writes as [0, height, thickness]. Ways found:
setLength);Nothing removes such a wall on save, load or export. The phone shows it as a
dot and draws none of its openings (segment(of:) returns nil for length 0),
and the Length field cannot fix it. Raw scans are checked on the phone and
cannot fail this way.
must have a positive height), so later
refusals show which one failed.handoff_tolerated (collapsedWalls, walls, prepared)
for a capture with such walls that passes validation, before admission.
handoff_invalid now also logs prepared. It is true for any prepared plan
(an edited iPhone plan or a sibling planner’s export) and false for a raw scan.tests/fixtures/handoff-collapsed-wall.json and handoff-collapsed-jog.json).
Both are accepted and import without the collapsed wall. Each still encloses
one room, and passes the project check, packaging, re-export and Add to this
plan. The window on the collapsed wall is dropped under every geometry choice.
A 3 mm wall keeps its window. Refusals for negative sizes and for 0 outside
walls, each size’s reason, and the log fields are pinned. Six mutations of the
change (skip removed, skip only without straightening, 5 mm threshold,
prepared fixed at false, tolerated always logged, server counting only
exact 0) each fail at least one test.npm run check clean; npm test 1,596 passed; the three sibling
parity suites pass (6, 22 and 4 tests); the 11 specs that drive RoomPlan
import pass on Chromium, Firefox and WebKit (117 runs). Four reviewers
(correctness, interchange contract, server and logging, evidence) checked the
change, with a skeptic re-checking each serious finding; the confirmed ones
are fixed. The rest of the browser suite runs in CI.Branch fix/roomplan-import-clone (PR #119). Build > Import RoomPlan failed
with “Failed to execute ‘structuredClone’ on ‘Window’: #
tests/browser/roomplan-import-attributes.spec.ts imports the sample both
ways and checks that the attributes reach the JSON export. It failed on the
previous build with the reported toast and passes on all three engines. The
older import specs all load tests/fixtures/handoff-roomplan.json, which has
no attributes./editor?import=CODE handoffs, which pass a plain value. None of the other
deep $state objects in the components reaches structuredClone: each is
display-only or passes on ids and primitives, and project data is read from
the stores. The Share dialog already snapshots its project (#112).npm run check clean; npm test 1,579 passed; the seven specs
that drive RoomPlan import (84 runs) pass on Chromium, Firefox and WebKit.
The rest of the browser suite runs in CI.Branch fix/editor-defects-3 (PR #112, stacked on #110). Merged and deployed October 16, with #107–#110: all five App Hosting rollouts succeeded, and the live handoff preflight returns 204 for the sibling origins. The 21 lower-ranked
findings left over from the October 15 hunt were re-checked by two verifiers
each (one reproducing, one refuting): 17 confirmed outright, the elevation
Escape defect by one verifier plus a reading of the code, the interchange
round-trip loss judged the documented v1 contract (documentation only), the
history-panel claim refuted, and the Portuguese leaks verified while fixing
(room-type names stay English by design: the name is also the room-type key
for fills, patterns, the Type dropdown and the sibling category). Fixed in
eight parallel worktrees, then reviewed by seven reviewers, whose 13 findings
were fixed in a second round:
wallXray.ts); colour canvas textures decode as sRGB (viewerTextures.ts);
time-of-day presets repaint the sky (three caches the background cube, so the
texture is disposed after a redraw); wall textures tile at a constant 200 cm
(mapWallUVs); straight-wall baseboards are cut at floor-level windows;
without WebGL the view explains why and the one-time 3D tip waits for a
working view; the orbit camera re-frames only on a new context
(sceneFraming.ts), not on edits or texture loads.snap() the unit tests cover.inlineEditorFocus.ts); the elevation view leaves Escape to
focused fields and open menus/panels (popupOwnsEscape) and its canvas takes
focus when pressed; busy Restore, Import, Share and library rename/delete
dialogs re-show after a forced close (reconcileNativeClose); Share no longer
fails with “could not be cloned”; the welcome overlay is a real modal.exportText.ts); long furniture captions are framed rather than clipped
(furnitureCaption.ts).Under “Is there a Wiki?”, an end user (Windows 11, Firefox, everything drawn by
hand) listed six problems on 2026-09-29; the October 15 work answered only the
documentation question. Each was reproduced with the modules of the version
live then (d68cadf) and fixed where still wrong:
exportVisibility.ts); everything still exports by default, and a floor
whose content is all hidden gets a notice. The status-bar “Automatic
dimensions” checkbox now writes the saved Settings > Dimensions switch.printPageLayout) now measures labels, notes, stair, measurement,
dimension and furniture captions at the sheet’s scale.navigator.storage.persist() once, and while storage is not persistent the
library shows a dismissible note recommending Download library backup; the
guide explains Firefox’s prompt, delete-on-close settings and profiles. The
app cannot bring lost projects back.044a797).Also fixed while verifying: the wall elevation now refreshes when the shown wall is edited in Properties (it kept the old length and heights until reopened, because the store edits walls in place).
Verification, October 16, local: npm test 1,577 tests in 163 files, npm run check clean,
production build. Chromium browser runs, each on a fresh build: the first
round’s 18 new or changed specs (32 tests) and 72 dependent specs (193 tests);
the review round’s 63 specs (165 of 166 tests; the elevation case failed on the
older refresh bug described above, and the four elevation specs then passed in
all three engines, 24 tests); the issue #101 round’s 35 specs (119 of 120; the
storage-note case relied on the user activation automation grants and now
models a person’s, 18 tests in all three engines). Blocked-storage,
save-conflict and library specs: 39 tests in all three engines, after a fix for
Firefox reporting an immediately rejected Web Locks callback as an uncaught
error. Full three-engine CI runs on the PR.
Known trade-offs and follow-ups (not fixed here):
getRoomPolygon re-traces faces per call; exports of very large plans are
about 1.8× slower (a 266-wall test plan: 176 → 325 ms per pass).ContextToolbar.svelte is dead code and the user guide (around lines
217–219) describes its flip buttons; FloorPlanCanvas still imports
duplicateDoor/duplicateWindow without using them.Branch fix/editor-defects-2 (stacked on the three branches below). A
37-agent hunt over the editor’s core flows, with two refuters per finding,
confirmed fifteen defects; all are fixed here with fixture-based regressions:
versionHistory.ts, tests/version-history.test.ts).roomDetection.ts, tests/room-detection-stubs.test.ts).doorSymbol.ts now gives SVG and DXF one shared, type-aware glyph and the
SVG output is byte-identical (tests/dxf-door-symbol.test.ts).roomplanImport.ts, tests/roomplan-import-labels.test.ts).projectOpening.ts).projectPackage.ts, BuildPanel.svelte); the paste
path accepts the same formats.hitTesting.ts, tests/wall-hit-testing.test.ts).mutate records nothing for a no-op edit and
endUndoGroup ignores timestamps (stores/project.ts,
tests/history-semantics.test.ts).placementArming.ts, wallTranslation.ts, marqueeSelection.ts).Verification, October 16, local: npm test 1,414 tests in 144 files,
npm run check clean, production build; chromium browser cases for the
touched areas: 73 passed in two batches (versions, project package, exports,
shortcuts, stitching, door join, PDF, walls, rooms, history, editor).
Sibling parity suites (6/22/4) and the cross-planner harness (4 per planner)
still pass. Firefox/WebKit not run locally.
The 21 lower-ranked candidates from the same hunt were verified and handled in the next batch (PR #112, the section above).
Branch fix/editor-small-bugs (stacked on the two branches above). Six
user-visible defects, each with a regression:
+ / = / - / _ (and numpad) now zoom the plan ×1.25 within
the canvas limits from handleGlobalShortcut; the shortcuts dialog had
advertised them with no handler. Never with Ctrl/Cmd, while typing or with a
modal open (tests/shortcuts.test.ts).alert() is gone; the DXF downloads and an
ExportNotice (“DWG saved as DXF”) explains it. exportDXF/exportDWG return
whether a file was downloaded (since PR #112, true or why nothing was
written: 'empty' or 'unsupported') (tests/exports.test.ts,
tests/browser/dimension-exports.spec.ts).alert() is a localized inline note.addWall, moveWallEndpoint and moveWallGeometryDuringDrag
store fresh { x, y } points, so magnet-snap flags (snappedToEndpoint,
snappedToWall, snappedWallId) no longer reach saves or Download JSON and
joined corners no longer share one object. Older saved projects that already
carry those keys are not rewritten; the keys are harmless and vanish when the
endpoint next moves (tests/wall-geometry-drag.test.ts).pdf.* keys, with exportPDF and
exportPDFWithFeedback taking a language (default en).ProjectPackageDialog.onimported
now receives the imported project’s library id
(tests/browser/project-package.spec.ts).Verification: npm test 1,339 tests in 135 files; npm run check clean;
production build; chromium browser cases for shortcuts, project package, PDF
and DXF exports pass (18). Firefox/WebKit not run locally.
Web branch fix/sibling-compat-and-open-issues, uncommitted at the time of
writing; nothing below is pushed or deployed. Dated record:
docs/reviews/2026-10-15-sibling-interchange-and-issues.md.
POST /api/handoffs (src/lib/server/handoffCors.ts, the
route): OPTIONS answers 204 and every answer carries
Access-Control-Allow-Origin for the family’s deployed origins only (apex
and app. hosts of openplan3d.com, openlandscape3d.com, openkitchen3d.com
and opencatio3d.com, plus openkitchen3d-cloud.web.app and
.firebaseapp.com, as an exact list; http://localhost / 127.0.0.1 only
with HANDOFF_CORS_ALLOW_LOCAL=true); never credentials; Retry-After
exposed. The
OpenLandscape3D and OpenCatio3D “Open in OpenPlan3D” buttons POST
cross-origin JSON; the live endpoint answered their preflight with 405 on
October 15, so they always fell back to a JSON download. Not deployed yet.src/lib/utils/interchange.ts, transferLink.ts):
projectToPreparedRoomPlan writes the family’s prepared RoomPlan JSON from
a web project; siblingImportUrl builds <origin>/#/import/v1/<payload>
links (unpadded base64url of raw-DEFLATE compact JSON, 262,144-character
cap; the kitchen origin is https://openkitchen3d-cloud.web.app until
app.openkitchen3d.com is live). Import keeps the openkitchen3d /
openlandscape3d / opencatio3d blocks, object attributes, defaults and
planNotes on project.interchange (validated by readProject, saved in
project JSON, not in the native package; Add to this plan merges them
too) and export writes them back without a webProject snapshot, with a
refreshed building and a ceiling default that follows the walls, and with
the family’s link-privacy rules on links, checked against each sibling’s
real fixtures (tests/fixtures/siblings/) and importers.
The Export menu gains Other planners → Open in OpenLandscape3D /
OpenKitchen3D / OpenCatio3D (new tab, plan in the link fragment, file
download when too large) and Download interchange JSON
(<Project-Name>-openplan3d.json).wallDimensionLabel.ts).angleSnapWithin, FloorPlanCanvas.svelte).roomplanMerge.ts,
stores/project.ts, BuildPanel.svelte). Door-to-door alignment: right-click
a door → Join to another door… → click the matching door; the scan’s wall
group turns and slides to the far side of that wall, one wall thickness
apart, with its furniture, as one history entry (scanAlignment.ts,
tests/scan-alignment.test.ts, tests/browser/scan-door-join.spec.ts).
Open: a gap control in the UI, merging the doubled wall into one, carrying
T-junction walls.docs/skills/openplan3d-plan-authoring/SKILL.md with two
examples checked by tests/skill-examples.test.ts. Issue #101 —
docs/user-guide.md, linked from the editor as User guide.openplan3d-www, uncommitted): the three planners are listed and
an FAQ entry explains the exchange.Verification, October 15, local, this working tree, after the adversarial
review fixes below: npm test passed 1,325 tests in 134 files (zero
failures); npm run check found 0 errors and 0 warnings; the production build
succeeded. Browser checks on that build, chromium only: 20 related specs
(67 tests) passed before the review fixes, and the 11 specs (38 tests) most affected by
them passed again afterwards, including the two new specs
tests/browser/roomplan-add-to-plan.spec.ts and
tests/browser/interchange-export.spec.ts. The nine new unit files: 147 tests
(handoff-cors 39, sibling-interchange 36, interchange 18,
endpoint-drag-snap 13, roomplan-merge 15, interchange-ui 8,
export-wall-measure-mode 6, transfer-link 6, skill-examples 6; counts
from npx vitest run <file>). The siblings’ own parity suites against this
checkout pass (OpenLandscape3D 6, OpenCatio3D 22, OpenKitchen3D Swift
fixtures 4), and a session-scratch harness that feeds this editor’s export to
each sibling’s real fromOpenPlan and decodePayload passes 4 cases per
planner, including a web file with an embedded webProject whose wall edits
made here arrive in OpenKitchen3D and OpenLandscape3D. Firefox, WebKit, full
browser CI and physical devices have not run on this tree.
Adversarial review (44 agents: four finders, two refuters per finding): 20
findings, 17 confirmed, all addressed in this tree — retained blocks no longer
re-emit webProject, refresh building and the ceiling default, and links
apply the family’s privacy rules; section anchors are interior points;
Add to this plan keeps the sibling block and attributes and resets the
editing context; walls rotate only as a closed group; the Export menu scrolls;
the CORS allow-list is exact with local origins behind
HANDOFF_CORS_ALLOW_LOCAL; the DXF export follows Measure walls; and the
documentation passages the review called out were corrected.
Open:
project.interchange; today it survives only in project JSON and library
backups.<block>.name; and OpenCatio3D
rebuilds its house from site.walls inside its block, so wall edits made
here do not reach its host walls until its importer prefers the top-level
arrays of a re-exported file.label and a per-room ceiling on import; export
derives the label from the room type or name and the ceiling default from
wall heights.Option A of the hosted connector design is implemented server-side and
disabled by default: POST /api/assistant-shares (validated, stripped
package, 7-day retention, code + hashed secret, own quota ledger),
DELETE /api/assistant-shares/CODE, and a stateless POST /mcp with four
read-only tools over src/lib/skills/, the TypeScript ports of the native
skill scripts with parity fixtures. See docs/assistant-shares.md for limits
and the deployment order (lifecycle rule first, then the flag). The web Export
menu has Share with Assistant (dialog, client service, en/pt strings, two
tests); the native action is openplan3d-ios PR #22. Enabled in production on
September 16 (PR #98) after the bucket lifecycle rules were applied; a live
curl check and a claude.ai chat with the custom connector (summary and photo
review tools, approval prompts, no re-asking for the code) both passed.
Open: ChatGPT check.
Web package export now writes the same statistics block the native app added
in openplan3d-ios PR #17: plan totals, per-level totals, per-room interior-face
areas and costs, recomputed on every export and never read back. A stale block
in an imported native package is replaced; baseline.json excludes it. See
src/lib/utils/planStatistics.ts and tests/planStatistics.test.ts; the two
exact plan-equality tests now compare everything except the derived block.
Implementation and validation are paused until access to the Asta 6 model is restored. All three repositories are committed, pushed and clean at this pause; no build, test, browser or native QA job is running. The overall goal remains active and all outstanding scope below remains open. Resume from STATUS.md.
Web PR #95 merged into main as 4bdc00d after the required build, unit/type
check, benchmarks and browser checks completed. The remote and local
codex/threejs-render-lab branch were removed; the web checkout is clean on
main and matches origin/main. The website checkout is also clean on main
at 4802b01 and matches its remote. Native development remains on its own
codex/local-floorplan-render branch. Work remains paused pending Asta 6 access.
Area Summary now separately displays interior area (room polygons minus the union of wall footprints) and explains that the existing breakdown uses wall centerlines. English/Portuguese text is included. Bounded scanline integration supports concavity, nested holes, overlapping/internal walls and faceted curves; invalid/over-budget cases display Unavailable. Floor-opening rooms are excluded.
The new calculation matches native’s 10.64 m² rectangle result. All 18 geometry
tests passed (35410, 796 ms); production build passed, and all three localized
phone browser workflows passed (8826, 20.1 seconds). See the area record for
algorithm details and logs. General native raster parity, room labels, category
schedules and exports remain open; this explicit comparison is not full alignment.
Paired native/web runtime baselines use identical 4 × 3 m wall centerlines and 20 cm wall thickness. Web resolves 12.00 m²; native enclosed-area and detection both return 10.64 m², exactly the analytical interior area. Difference: 1.36 m² (11.33% of centerline area), caused by boundary semantics rather than rounding.
Native simulator 75624 passed (exit 0, 0.027 seconds); all six web room tests
passed (404 ms). See area baseline and remaining work.
These current-behavior tests expose the mismatch and must be updated when a
common convention is implemented; they do not qualify area parity as complete.
Added grouped label-drag state coverage for saved and newly detected rooms.
Preview movement leaves stored metadata untouched; commit, two Undo/Redo cycles,
and geometry resolution with a stale moved preview retain exact metadata and
stable IDs without resurrecting the offset. All 16 history/resolution/nesting
tests passed in 11923 (exit 0, 769 ms), /tmp/web-room-label-state-history.log.
The original intermittent browser issue remains open: this narrows the tested state/resolver path, not browser event order, draw scheduling or viewport changes. No speculative runtime fix was made. See the nested-room reproduction review.
Repeated the actual package export, waited for the full expanded Save panel,
verified the changed filename in a separate accessibility read, and saved.
/Users/thelodgem1/Documents/native-ui-height-save-final.zip exists (4,911 bytes).
Its plan bytes exactly match the committed edited-height package fixture.
SHA-256: b174dc8efcd912079474fb0a0c9b686d126e79f4bf565988dde5e672922c9ef2.
This completes the scoped Catalyst selected-destination export check. Earlier
attempts remain inconclusive and do not establish a runtime defect. No code
change was needed. Physical-device handoff remains open, along with the broader
NEXT scope. Native evidence: docs/native-furniture-height-editor.md in the iOS repo.
Full simulator 64276 passed on native 2fc1dd9: 265 passes, two optional
integration skips, zero failures; XCTest 130.573 seconds, exit 0. Log:
/tmp/native-height-editor-full-simulator.log.
The native UI export generated an unchanged sandbox ZIP now committed as
tests/fixtures/native-ui-edited-heights-package.zip. All 39 package tests pass
in 62196 (1.68 seconds): imperial edit and explicit category reset import,
all 14 native furniture records return exactly apart from UUID case, and the
12 legacy items keep height omitted. Web projection uses existing eight-decimal
cm rounding; native return retains its original precision. The requested Save
destination file was not located, so that handoff detail remains open.
Isolated Catalyst UI on native 2fc1dd9 switched to Metric, entered 213.7 cm,
and reopened properties with that value retained. Use Category Height displayed
90 cm. Zero entry was rejected: saving retained explicit 0.9 m, width/depth
0.5 m, and the neighboring chair’s exact earlier height. See the native height
editor review for session/object IDs and limits.
Full iOS 26.5 simulator run 64276 is active in
/tmp/native-height-editor-full-simulator.log, using
/tmp/openplan3d-render-ios-build. Poll before competing builds; do not restart
because of quiet output. Remaining checks include other invalid input forms,
intermediate undo value, edited-height UI package return and physical devices.
Native 2fc1dd9 adds decimal height in the selected cm/in units and Use Category
Height, storing explicit reset values for package merging. Catalyst build
64930 passed. Actual isolated UI entered 48.5 in for a new chair, then Undo,
Redo and Save; saved height was 1.2318993347743592 m, width/depth still 0.5 m.
See openplan3d-ios/docs/native-furniture-height-editor.md for evidence and the
stale post-Undo accessibility action that produced an empty sheet (dismissed).
Metric entry, reset/invalid input, intermediate undo value and iOS/device checks
remain open; earlier full native suites predate this UI change.
Web height output was imported as a copy through isolated Catalyst UI, saved,
inspected in 3D, and exported through the native project-package Save dialog.
The actual native ZIP now lives at tests/fixtures/native-ui-height-package.zip.
Web service re-import/re-export preserves 3.125 m physical height, 125 cm at
Z scale 2.5, reflection, rotation, furniture extension metadata and attachment
bytes. All 38 package tests passed in 25778 (exit 0, 1.48 seconds). An initial
comparison failure was only UUID letter case, normalized as in existing tests.
See handoff evidence.
This closes the scoped Catalyst UI/package preservation check; physical-device handoff, height editing usability and richer furniture rendering remain open. The overall goal remains active. STATUS.md reflects this checkpoint.
CI 34735405519
completed successfully: build, 1,142 unit tests, type check, all 18 browser shards,
both benchmark profiles and aggregate check passed. Linux Firefox shard 3 passed
all 74 tests in 5.6 minutes, confirming the phone status-bar repair in the original
failing environment. Log: /tmp/openplan-statusbar-ci-firefox3-success.log.
This run covers implementation 3020a65; subsequent changes through 0fcae31
are tests/documentation.
Full Catalyst run 25783 also completed with exit 0 on native 8fdd1bf: 267
reported tests, 265 passed, two optional integration skips, zero failures;
XCTest 142.046 seconds. Log: /tmp/native-height-full-catalyst.log. Both current
native full suites now pass. The earlier active/pending entries below are history.
The overall goal remains open. Next: actual furniture-height native UI/package exchange, then the remaining investigation, fidelity, device and release scope. See STATUS.md for the maintained current state.
Full simulator run 72751 passed on native 8fdd1bf: 267 reported, 265 passed,
two optional worker skips, zero failures, exit 0; XCTest 93.611 seconds. Log:
/tmp/native-height-full-simulator.log. Full Catalyst run 25783 is active in
/tmp/native-height-full-catalyst.log; poll before competing native work in
/tmp/openplan3d-render-ui-build.
CI run 34735405519 build job 103665831685 passed all 1,142 web unit tests in
119 files, type check (zero diagnostics), and production build. Browser shards
and benchmarks remain active. Log: /tmp/openplan-statusbar-ci-build.log.
Passing fixture run 81629 generated /tmp/web-height-return-package.zip via
OPENPLAN_HEIGHT_RETURN_PATH. Its actual package records contain native height
3.125 m and web height 125 cm at Z scale 2.5. Use this verified web output for the
next native import/save/export check; actual cross-platform height qualification
is not yet complete. Log: /tmp/web-height-return-artifact.log.
Linux Firefox CI screenshot confirms a horizontal scrollbar covering the mobile
status text. Local unchanged Firefox reproduction 83240 passed (27 seconds),
so that pass alone did not qualify the Linux failure. Artifacts are retained in
/tmp/openplan-firefox-shard3-ci-artifacts, with the screenshot committed at
docs/reviews/assets/mobile-statusbar-firefox-before.png.
Web 5e09840 gives the phone status bar 48 px minimum height and centers the
controls. The existing click assertions remain; an additional check requires
buttons to fit their text line, and phone screenshots are attached. Build
35608 passed; all six desktop/phone cases across all engines passed in 85538
(exit 0, 1.8 minutes), /tmp/web-mobile-statusbar-browser.log. Focused Firefox
phone screenshot run 43494 also passed; the image was inspected and shows
unclipped controls. Log: /tmp/web-mobile-statusbar-visual.log.
See docs/reviews/2026-09-12-mobile-statusbar.md; Linux CI confirmation remains open.
Web 79633db projects optional native height to centimetres, exports physical
height including absolute Z scale, and divides out retained scale on native
height edits. Older encoders’ omitted height inherits the baseline. Unchanged
legacy native records keep height omitted; flat catalog symbols do not emit
invalid zero native height. The old category fixture now explicitly expects its
seven retained 94.625 cm web heights as 0.94625 m on export.
All 92 package/category tests passed in 4765; the first run exposed an invalid
zero-height test input and the stale fixture expectation, both corrected without
weakening comparisons. Type check 86377 passed with zero diagnostics. Build
74745 passed with exit 0, /tmp/web-package-height-production-build.log. Native 8fdd1bf
focused tests already passed; actual cross-platform height qualification remains
open. See the updated package contract and STATUS.md.
CI run 34734473227 is terminal: build, both benchmarks, and 17/18 browser
shards passed. Firefox shard 3 failed Portuguese layers at 390 px because the
bottom overlay intercepts the Grid button. Log:
/tmp/openplan-ci-firefox-shard3-failure.log. Reproduce/fix the obstruction;
do not treat it as another budget-only timeout.
Native 8fdd1bf retains measured furniture height through RoomPlan import,
saved documents, duplication, SceneKit preview, RoomPlan export, and package
merging, with positive/bounded validation and legacy category defaults.
Both simulator regressions passed in 40396 (exit 0; 0.022 and 0.057 seconds),
log /tmp/native-furniture-height-regression.log. Web package height projection/scale handling
and cross-platform qualification remain open; earlier full native passes predate
this implementation.
Raster browser 26950 completed with exit 0: all three engines passed the
unchanged moved-label export workflow, 2.2 minutes total. Log:
/tmp/web-raster-rounding-browser.log. CI run 34734473227 passed its build;
the browser shards/benchmarks remain in progress. STATUS.md reflects these
outcomes. Do not restart the terminal raster run.
CI run 34733632131 reached its 720-second suite limit in every engine:
Chromium 144 passed/249 unrun, WebKit 153 passed/240 unrun, Firefox 167 passed/
one failed/225 unrun. The browser matrix now has six shards per engine with
unique report artifact names. Per-case assertions and existing job/suite limits
are unchanged. Actual Playwright collection 40423 verified that the shards
cover all 393 cases per engine exactly once (1,179 total), with counts
71/61/74/58/64/65. Inventory: /tmp/openplan-browser-shard-inventory.json.
The workflow YAML parses and expands to 18 browser/shard jobs. Full CI execution
remains the qualification gate.
Firefox’s assertion failure was separate: the moved-label PNG width was 4095,
expected 4096. PNG/PDF canvas allocation now rounds up scaled dimensions before
applying the 4096 cap, avoiding integer truncation of floating-point dimensions.
All 27 export unit tests passed (53414, /tmp/web-raster-rounding-unit.log),
including exact capped/integer dimensions. Production build 42091 passed;
log /tmp/web-raster-rounding-production-build.log. The unchanged moved-label
export workflow runs across all three engines in 26950, log
/tmp/web-raster-rounding-browser.log; poll before another browser run.
Height-contract inspection exposed another reflection path: the retained-package
merger’s known furniture fields omitted mirrorX and mirrorY, so edits to
imported package furniture were not exported. Native 72906c0 includes both
fields. Regression 94702 passed (exit 0, 1.086 seconds): actual native and web
package import, save, and export retain edited true/false flags and explicit
removal, without changing angle, unknown metadata or retained web/baseline/mapping
documents. Log: /tmp/native-reflection-package-merge-regression.log. This scoped
pass is newer than the full native source baseline below. Height remains open.
CI run 34733632131 is now terminal: build and both benchmark profiles passed,
but all three browser jobs failed. Logs are being retrieved to
/tmp/openplan-ci-fixed-benchmark-browser-failures.log; diagnose these before
claiming full browser qualification.
Current PR #95 run 34733298723 built successfully but its benchmark jobs
failed before collecting measurements. Desktop job 103660094800 reports an
unsupported virtual: URL in Node’s ESM loader. Local collection 24804
reproduced that error (exit 1, zero tests). The Playwright benchmark glob also
matched the Vitest-only wall-hit.bench.ts; it now selects viewer.bench.ts.
This retains both desktop/phone viewport profiles and all three furnished-home
sizes. Corrected collection 87181 passed (exit 0): all six expected tests in
one file. Log: /tmp/openplan-benchmark-collection-after.log. CI run
34733632131 on fix 2b40fd4 then passed all three desktop cases (2.0 minutes,
job 103661083321) and all three phone-viewport cases (1.5 minutes, job
103661083299). Logs: /tmp/openplan-ci-benchmark-{desktop,phone}-fixed.log.
These are software-rendered measurements, not physical-phone performance gates.
Latest full web unit run 34034 is terminal (exit 1): 1,135 passed and three
five-second timeouts, 119 files/1,138 tests, 535.36 seconds. All nine tests in the
three affected files passed unchanged with one worker in 91966 (exit 0,
16.80 seconds), /tmp/web-reflection-unit-timeout-rerun.log. The full-run log is
/tmp/web-reflection-current-full-unit.log; no assertions or budgets changed.
Full browser 96834 is terminal (exit 1): 35 passed, five timeouts, 1,139 unrun,
52.3 minutes. Its runner-level error is the configured five-failure stop.
Traces are preserved in /tmp/openplan-reflection-full-browser-failures.
Area localization completed its assertions; both automatic-dimension cases
timed out at the initial Export click before importing geometry. Guide selection
timed out on its baseline export; measurement selection timed out on a later
JSON export. These do not establish a geometry defect. Current-source production
build 74413 passed (exit 0), /tmp/web-reflection-current-production-build.log.
Scoped reproduction 15604 then passed all five previously timed-out Chromium
cases unchanged (exit 0, 1.6 minutes), /tmp/web-reflection-five-timeout-repro.log.
The collection check confirmed exactly five cases, and artifacts are preserved
in /tmp/openplan-reflection-five-timeout-repro-passed. Full CI browser jobs
remain pending; these five passes do not qualify the 1,139 previously unrun cases. The previous browser build predates the reflection
validator scope correction.
Catalyst 68783 completed with exit 0 and TEST SUCCEEDED: 264 reported,
262 passed, two optional integration skips, zero failures (193.429 seconds).
Both full native destinations now pass on production source 3d076a9.
PlanDocument.fromRoomPlanJSON imports only furniture width/depth, discarding
dimensions[1]. PlanSceneBuilder and PlanDocument+Export then use category
default heights. Preserve measured height through the model, validation, native
preview/export, duplication and the web package bridge, with legacy-default and
round-trip coverage. This is separate from the qualified reflection work.
Follow-up inspection found that SVG furniture still applied translation/rotation
without reflection. PlanSVGExporter now appends local-axis reflection to both
outline and detail paths, preserving physical stroke widths. PNG/PDF already call
the corrected PlanRenderer. The new SVG test checks both path transforms for
X/Y/both reflection and exact unchanged output after removing only the reflection
transform. Native session 1018 passed (exit 0, test 0.013 seconds), log
/tmp/openplan3d-reflection-svg-tests.log. The full simulator FloorPlanTests
target passed in session 17153 (exit 0): 264 reported, 262 passed, two
opt-in worker integration skips, zero failures, 198.775 seconds XCTest duration.
Log: /tmp/openplan3d-reflection-full-native-tests.log. Full Mac Catalyst
qualification 68783 also passed with exit 0: 262 passed, two skips, zero
failures, 193.429 seconds. Log: /tmp/openplan3d-reflection-full-catalyst-tests.log.
See the newer checkpoint above for the terminal browser outcome.
Native 450b913 adds optional local-axis mirrorX/mirrorY, toggles reflection
without angle changes, retains it on duplication, and applies it in glyph/SceneKit
rendering and neutral exports (including corrected winding). Web package returns
preserve scale magnitudes, independent angle/axis choices and omitted flags from
older native encoders. RoomPlan export/import preserves equivalent reflected
orientation through the transform matrix. See docs/project-package-v1.md for
the exact contract; this is not a release or physical-device qualification.
Verified:
69010 and outward-face test 40145 passed.85762 passed all three reflection tests, including RoomPlan
round-trip. The xcresult summary confirms three passes, zero failures/skips
(/tmp/openplan3d-reflection-roundtrip-summary.json).75847, /tmp/web-package-reflection-tests-final.log).58649, /tmp/web-roomplan-reflection-tests.log).16540, 54415).19450 passed; its five formerly timed-out Chromium workflows all passed
in 67115 (12.8 minutes, /tmp/web-reflection-failed-workflows-rerun.log).
Traces: /tmp/openplan-reflection-browser-retry-artifacts. The timing changes
preserve every assertion; these are scoped passes on the package-reflection
build, which predates the RoomPlan source change.The asymmetric renderer test 12787 passed and both exported images were
visually inspected: the refrigerator handle changes sides while the body stays
fixed. Images and exact validation scope are in
docs/reviews/2026-09-12-furniture-reflection.md.
Full unit run 32001 had 1,135 passes and one stale exact-output expectation:
mirrored web fixtures now correctly export mirrorX: true. That expectation was
updated explicitly, and all 55 category cases passed in 58515; no runtime fix
or comparison relaxation was needed. Build 69534 passed.
Active work — poll these handles before competing runs:
96834: /tmp/web-roomplan-reflection-full-browser.log;
JSON report /tmp/web-roomplan-reflection-full-browser.json, five-failure stop.11694 passed and is terminal. Its fresh isolated app passed
live Mirror/Undo/Redo with visible handle reflection and persisted mirrorX: true
at the unchanged 15-degree angle. See the reflection review for exact saved data.The isolated app exported its actual reflected ZIP, now committed as
tests/fixtures/native-ui-reflection-package.zip. Web service import, exact native
furniture re-export comparison and a second web import passed with all 33 package
tests (43006, /tmp/web-native-ui-reflection-package-tests-final.log).
The web return was generated by passing fixture run 17631 and imported into
the native UI as an independent copy. Its preview retains the reflected handle;
complete saved furniture data matches the original (UUID case normalized). See
the reflection review for both session IDs. This scoped round-trip is verified;
current full browser qualification and physical-device testing remain open. The package validator
now scopes reflection flag types to furniture; all 32 package tests passed in
21070. This correction is newer than the build used by browser run 96834.
Native 43fd94f explicitly names Undo/Redo and contextual properties, left/right
15-degree rotation, duplication, mirroring and deletion, with desktop tooltips.
Tool buttons expose selected state. Catalyst build 33064 passed, log
/tmp/openplan3d-editor-action-labels-build.log. The isolated signed copy
/tmp/OpenPlan3D-Action-Labels-Sept12-QA.app exposed every new label in its
accessibility tree; changing Select to Wall updated selected traits. Adding a
chair exposed the contextual actions. Labeled rotation/properties/Undo/Redo/Done
were exercised, and the saved plan confirms a 15-degree rotation and 24-inch width.
See docs/reviews/2026-09-12-native-editor-action-labels.md for scope. Physical
VoiceOver/touch qualification remains open; device inventory found no devices.
Follow-up actions also exercised left rotation, duplication, Mirror, deletion and deletion Undo. Saved data confirms the duplicate and restored deletion. These checks exposed the former Mirror defect: it negated the angle instead of reflecting the shape. The reflection implementation and scoped UI/package/export verification above supersede that finding. Full regression and physical-device qualification remain open.
Browser collection at b65549e succeeds with 1,179 cases in 146 files: 393 per
engine. Inventory: /tmp/web-current-browser-inventory.json (session 82422,
exit 0, no collection errors). The historical 1,116-case qualification below
does not cover all current changes. The full unit suite passed all 1,132 tests
in 119 files (session 82392, exit 0, 211.46 seconds), log
/tmp/web-current-full-unit.log. Nested-room diagnostic validation passed all
three engines (session 68614, exit 0, 7.4 minutes); traces are preserved in
/tmp/openplan-nested-undo-coordinate-artifacts.
Pre-reflection browser qualification 5924 is terminal (exit 1): nine passed,
five timed out, 1,165 did not run (23.5 minutes). It stopped at the five-failure
limit; inspection confirmed the runner-level error reports that configured
five-failure stop rather than another application failure. Logs/report:
/tmp/web-current-full-browser.log, /tmp/web-current-full-browser.json.
Artifacts: /tmp/openplan-full-browser-pre-reflection-artifacts. Guide and
entourage Undo equality checks passed before timeout; AI-provider and Portuguese
alignment traces remain to inspect. No current full-browser pass is claimed.
The reflection package source is newer than this production build.
Local static GLB import now includes bounded container/resource/geometry/material validation, embedded PNG/JPEG decoding, original-byte/digest retention, project model definitions, source reuse and storage/history admission checks. The Objects panel provides preview, provenance, placement at the view center and unused-model removal. Saved references render through shared source leases with instance-owned GPU resources; leaving the viewer releases the last source owner. User model names now appear in canvas, Layers and Properties. The controlled GLB validator messages now have English/Portuguese translations, including variable schema identifiers; external diagnostics retain their fallback.
Verified: the textured loader fixture renders and cleans up across Chromium,
Firefox and WebKit (42690); placement/catalog tests and expanded lifetime tests
passed. The initial UI workflow passed in Chromium (95224) and WebKit (63246),
including exact source retention, placement Undo/Redo and confirmed save/reload.
Firefox’s earlier run timed out; current-build run 55495 subsequently passed
the complete import/placement/Undo/Redo/save-reload workflow. Legacy regression
57520 had 70 passes and two photo/history quota-test timeouts, not a full green run.
Cancellation/removal run 45654 passed cancellation, invalid-file rejection and
cancelled-removal equality before timing out. The split workflows now pass in Chromium, along with Portuguese controls and
full-viewer image lifetime, on the current names/messages/lazy-preview source.
Type checks for labels (79860) and messages (10317) passed with zero diagnostics.
Lazy-preview type check 85089 passed with zero errors/warnings and production
build 7222 passed. Chromium run 84773 passed all four tests, including exact
removal Undo/Redo, untranslated user names, translated source/removal errors,
shared decoding for two placed instances and image closure on leaving 3D.
A fixed-diagnostic inventory and translation tests passed (46495). The expanded
validator translations are newer than those browser runs; type check 12414 passed
with zero errors/warnings and production build 75751 passed. Their additional
Portuguese browser check passed in Chromium (10501), covering literal extension
identifiers and invalid geometry diagnostics.
Shared storage errors now refer to attachments and translate in the model dialog;
44 translation/service-message tests passed (13515). The preview also adds
labeled keyboard-operable rotation/zoom buttons and a translated image label.
Production build 22153 and fresh type check 73539 passed (zero type diagnostics).
Keyboard case 67983 timed out before preview after a 141-second Objects-tab action;
run 82817 passed every keyboard/image assertion with a larger whole-test budget.
Native simulator test 6249 passed, and the actual Swift output passed the three
web return tests (17274), preserving model references, original GLB bytes and
shared edits.
Native UI import, opening and unchanged save/export passed in the isolated Catalyst
QA app with exact retained GLB and sidecar bytes. UI geometry editing and physical
devices remain open. Expanded Firefox/WebKit run 76360 finished with 12/14 passes;
its Firefox import timeout and WebKit focus failure were subsequently addressed.
The explicit opener-focus fix passed check/build 6750 with zero type diagnostics.
Rerun 55462 passed import/save in Firefox and WebKit and cancellation/focus in all
three engines. Chromium import reached the preview just after its ten-second
assertion deadline; with a bounded 60-second preparation wait, rerun 9013 passed
the complete import/placement/Undo/Redo/save-reload workflow. All named desktop
custom-model workflows now have passing engine coverage; these are scoped runs,
not a new full-project regression pass. Native package count build 64966 and a
fresh isolated preview check passed, displaying “2 floors · 4 walls · 1 attachment
file”.
Remaining: broader browser qualification, external error/localization review,
accessibility/physical-device testing, and native UI return
qualification. See docs/reviews/2026-09-12-local-model-import.md for exact scope,
logs, source checkpoints and incomplete checks. This does not close broader NEXT work.
After moving the existing three-minute test budget before fixtures, the full
WebKit nested-room workflow passed, including label Undo, exports, save/reload,
and active/stacked slab checks. Session 38782 finished with exit 0. Current-build
session 65295 also finished with exit 0: two Chromium and two Firefox full
workflow repetitions passed (8.5 minutes total). Earlier attempts timed out
before Undo. These passes do not establish a fix for the intermittent failure.
The test now records label-local coordinates, canvas transforms and bounds around
Undo, retaining the exact restoration assertion. All-engine diagnostic validation
passed in session 68614 (exit 0, three passes in 7.4 minutes), log
/tmp/web-nested-undo-coordinate-diagnostics.log.
See
docs/reviews/2026-09-12-nested-room-undo-reproduction.md for exact evidence.
Native 56be55b correctly shows singular/plural frame counts and calibration
verbs. Three synthetic previews cover one/two frame pairs and zero/one complete
or legacy frames. A successful mixed-frame import produced a “2 frames” library
card; existing single-frame cards show “1 frame”. Catalyst build 58073 passed.
See docs/reviews/2026-09-12-scan-count-wording.md for fixture corrections and
exact evidence. Physical iOS UI and broader NEXT work remain open.
Native runtime 8ed0b20 builds successfully for Catalyst. In the isolated QA
app, edited Done, no-edit Done, and Close returned to review. Close preserved
the saved file hash; a forced save failure retained the editor and unchanged
file, and retry saved successfully. See
docs/reviews/2026-09-12-editor-menu-dismissal.md for exact scope and evidence.
Home → Draw a Plan also saved a rotated chair, returned to Home, and reopened
with the saved object. Simulator PlanRecoveryTests session 37936 completed
with exit 0: all 11 tests passed, zero failures. Log
/tmp/native-editor-local-dismiss-ios.log. Load-error Close also returned to review without changing the malformed file;
restoring its original bytes allowed Retry to reopen the editor. Physical UI,
canvas-gesture/long-run dismissal checks and the broader backlog remain open.
Initial Cancel and post-import Back to Library/Done now pass live Catalyst
checks. Two synthetic imports preserved every source hash. All 21 selected
dataset tests passed on both Catalyst and iPhone simulator, including the existing
194-frame/394-file real scan preservation check on each. Simulator session 50299
finished with exit 0; log /tmp/native-scan-local-dismiss-ios.log. See
docs/reviews/2026-09-12-full-scan-dismissal.md before resuming. Physical UI,
editor dismissal and broader NEXT work remain open.
Initial Cancel and post-import Back to Library/Done now dismiss through a shared
button inside the presented sheet. All three paths passed live isolated Catalyst
checks, including two independent real package imports. Catalyst build and 12
iPhone simulator package tests passed. See
docs/reviews/2026-09-12-import-sheet-dismissal.md. Full-scan import, editor
dismissal, physical-device UI and broader NEXT work remain open.
Export Options now requests dismissal from inside its own sheet before the
existing onDismiss callback starts queued work. Live isolated Catalyst checks
verified Cancel, package export/save, PDF export and save cancellation without
Escape. The Catalyst build and 12 iPhone simulator package tests passed. See
docs/reviews/2026-09-12-export-sheet-dismissal.md. Separate editor/import-sheet
delays, physical share-sheet checks and the broader backlog remain open.
Tracing images now support optional native floor ownership. Native rendering, scaling and fitting respect the owner; legacy images without an owner remain plan-wide. Package exchange preserves ownership through floor reordering and renumbering, including image-only floors, and retains original asset bytes.
Validation: 28 web package tests passed, Svelte check reported zero errors and
warnings, and 16 selected tests passed on both Mac Catalyst and iPhone 17 Pro
simulator. The production web build passed. Live isolated Catalyst checks
confirmed visibility, fitting and owner-specific menu controls while switching
floors. Actual native re-export/browser checks also passed: six cases across
Chromium, Firefox and WebKit verify owned and legacy image visibility, rotation,
returned floor ownership and exact PNG bytes. See
docs/reviews/2026-09-12-underlay-floor-validation.md. Earlier full-suite results
below apply to their recorded source revisions, not this change.
The broader NEXT backlog remains open.
Both full native suites passed on production source 5838f4b: 256 reported,
254 passed, two opt-in integration skips and zero failures on each of Mac
Catalyst and iPhone 17 Pro simulator. XCTest durations were 103.930 and 91.378
seconds. Both processes terminated with exit 0 (sessions 29871 and 26834;
/tmp/native-sept12-full-{catalyst,ios}.log). Skipped worker integrations lacked
external inputs and are not counted as passes.
Rotated tracing-image fitting also passed the unchanged-plan Catalyst visual
check. The cropped-underlay fitting issue is addressed and verified. No native
build or test process remains active. Full physical-device, release, billing and
broader NEXT requirements remain open. Native details and reproducible commands
are in openplan3d-ios/docs/native-september12-full-validation.md.
The prior native UI check showed the tall rotated tracing image cropped below
the walls. Native editor framing now unions the active plan geometry with the
rotated underlay rectangle, using the same decoded-image aspect ratio as drawing.
Image-only plans fit the actual image location, without adding default origin
bounds. Missing/invalid dimensions fall back to the original geometry bounds.
Measurement and export bounds remain unchanged. Initial loading reads image
dimensions before fitting; importing a fresh trace also fits the complete image.
New UnderlayFitTests project all four corners at five angles into a padded
viewport, preserve geometry bounds, and cover remote image-only plans and invalid
image dimensions. The first Catalyst attempt compiled the app but failed on an
ambiguous .infinity test literal; it is now explicitly CGFloat.infinity.
The corrected Catalyst run passed all 14 selected fit/package tests (session
41154, terminal exit 0; /tmp/native-underlay-fit-catalyst-fixed.log).
Visual verification of 5838f4b also passed using the same saved synthetic plan
in /tmp/OpenPlan3D-Underlay-Fit-QA.app. The running executable path was verified.
This new QA binary reused only the prior isolated rotated-underlay sandbox; the
installed development app was not used. On opening the editor, the full tall
image and all walls are visible, with the image’s bottom edge above the toolbar.
The prior build cropped the lower image in this same plan. Rotation and color
orientation remain correct. The QA app was closed after inspection.
All 14 selected tests also passed on the iPhone 17 Pro simulator (session
35632, terminal exit 0; /tmp/native-underlay-fit-ios.log). Native source
5838f4b now has selected Catalyst/simulator and Catalyst visual verification.
The broader NEXT/device/release work remains open.
The first web floor’s PNG/JPEG/GIF tracing image now maps to the native underlay
at arbitrary rotation using optional clockwise-radian angle metadata. The native
canvas rotates the image around its existing center. Save/reopen and package
return retain the angle and original image bytes. Omitted legacy angles remain
omitted; native angle changes and explicit resets update web rotation without
resetting opacity or locking. Both readers validate optional angles consistently.
Validation: 24 web package tests passed after four new cases failed on the old
implementation; web type checking reports zero errors/warnings and production
build passed. Native package tests passed on Mac Catalyst (11 tests, session
58398, exit 0) and iPhone 17 Pro simulator (12 tests including invalid-angle
and explicit-reset coverage, session 75363, exit 0). Rotated PNG/JPEG save,
reopen, import and re-export preserve exact asset bytes. Logs are
/tmp/web-underlay-rotation-{before,unit,check,build}.log and
/tmp/native-underlay-rotation-{catalyst,ios}.log.
Native visual inspection and live native export returned through all three browser engines now pass; see the UI report. The subsequent native fitting change includes the rotated image extent and is verified in both selected test suites and the same native QA plan. No build or test process remains active. Device, release and broader NEXT work remain open.
Reproduced a window extending above a fractional wall top after an elevation
drag: a 250.25 cm wall exported a 250.5 cm window top (Chromium session 50371,
exit 1, 5.8 seconds; /tmp/web-elevation-clearance-before.log). Rounding the
already-clamped sill height caused the 0.25 cm overrun. Elevation dragging now
rounds the requested sill first, then clamps it to the exact lower wall height
across the opening, retaining fractional limits on flat and sloped walls.
All 15 focused browser cases passed in two minutes (session 92103, terminal
exit 0; /tmp/web-elevation-clearance-browser.log): flat, rising and falling
wall limits plus Escape/Undo drag workflows in Chromium, Firefox and WebKit.
Exports verify the opening reaches but does not cross the wall top, other fields
remain exact, and Undo/Redo restores the full original/moved floor. Type checking
reports zero errors and warnings; the production build passed (session 28955,
exit 0; /tmp/web-elevation-clearance-{check,build}.log). No browser process is
active. Earlier full-suite qualifications remain historical; broader NEXT
requirements and the intermittent nested-room Undo investigation remain open.
A new Chromium reproduction confirms that pressing Undo while a room-label press is still held leaves the label displaced by 60 pixels after mouseup. The pending label offset was committed after history replay. The canvas now finishes its active pointer gesture before applying Undo/Redo shortcuts, so mouseup cannot overwrite the restored state. This is a separate reproduced failure; it does not establish the cause of the earlier intermittent nested-room Undo failure after a completed mouse release.
The new regression covers interrupted label dragging and Redo. Seven geometry
workflows also exercise Undo while pressed, exact exported geometry restoration,
Redo, and subsequent Undo/Redo. Existing mouse-release cases remain intact.
The pre-fix Chromium case failed in 15.6 seconds (session 35372, exit 1;
/tmp/web-label-mid-drag-undo-before.log). Type checking reports zero errors
and warnings; the production build passed (session 31291, exit 0).
The pre-change full unit checkpoint passed 1,037 tests in 99 files in 3.42 seconds
(session 92756, exit 0; /tmp/web-sept12-full-unit.log). The same full suite
also passed on df4dfa4: 1,037 tests in 99 files, 30.12 seconds (session 66523,
exit 0; /tmp/web-mid-drag-history-unit.log).
All 48 focused browser cases passed in 11 minutes (session 26654, terminal
exit 0; /tmp/web-mid-drag-history-browser.log): 16 per engine, covering both
normal release and Undo while held for seven geometry types and room labels.
The room-label case failed before the fix and now passes in all three engines.
The separate elevation reproduction also failed (session 49556, terminal
exit 1, 18 seconds; /tmp/web-elevation-mid-drag-before.log). After Undo and
mouseup, the exported window retained position 0.5666111677120893 and sill
height 110, rather than the original 0.5 and 90. The elevation capture-phase
listener now ends its drag group before the global history shortcut runs.
Focused text fields retain native Undo/Redo behavior. Type checking reports zero
errors and warnings and the production build passed (session 28836, exit 0).
All six elevation cases passed in 1.1 minutes (session 96723, terminal exit 0;
/tmp/web-elevation-mid-drag-browser.log), covering normal Escape exit and Undo
while held in Chromium, Firefox and WebKit. Exact exported geometry, Redo and
subsequent Undo/Redo assertions passed. No browser process remains active.
The 48 passing plan-canvas cases above apply to df4dfa4; the six elevation
cases apply to e6da660. The older 1,116-case qualification applies to runtime
03e0ae1; it is not a full-suite result for these new changes. Broader NEXT
requirements, including the earlier intermittent nested-room Undo issue, remain
open.
Full stage 4 finished successfully (session 22429, terminal exit 0):
546 passed in 1.5 hours; /tmp/web-orbit-fixed-full-browser-4.log.
Exact inventory matching across the four continuation stages and focused
verification logs confirms 1,116 unique current-runtime passes: 372 each in
Chromium, Firefox and WebKit. Repeated cases count only once. Production
remains 03e0ae1; this qualification combines the resumed runs and focused
reruns after test synchronization/budget fixes, rather than one uninterrupted
suite run. No browser qualification process remains active.
The earlier intermittent nested-room Undo failure remains unexplained despite passing subsequent coverage. Broader NEXT requirements remain incomplete.
Full orbit-runtime stage 2 is terminal (session 46921, exit 1): 282 passed,
one failed, 583 not run in 56.2 minutes. Log:
/tmp/web-orbit-fixed-full-browser-2.log. Including the 250 retained passes,
532 current-runtime cases passed. The Firefox bad-photo/quota recovery test
failed at item-details.spec.ts:224 while awaiting the second photo chooser.
The Add photo click completed about five seconds into the test, but no chooser
event arrived before the 60-second total timeout. The snapshot shows the
expected storage-full banner after editing the note. This is not evidence that
the test merely needs more time. Original trace and context are preserved in
/tmp/web-photo-chooser-failure. The unchanged Firefox reproduction passed in 39.2 seconds (session 43729,
exit 0; /tmp/web-photo-chooser-repro.log). The failed trace selected the old
Add photo coordinates while the note autosave could introduce the storage-error
banner; a layout race is a hypothesis, not a confirmed product defect. The test
now explicitly waits for the expected storage-full alert after editing the note
before clicking Add photo. This adds synchronization and preserves all original
photo, saved-byte, draft-export and retry assertions. Production is unchanged.
All three traced photo/quota cases passed in 2.6 minutes (session 80528,
terminal exit 0; /tmp/web-photo-quota-all-engines.log). This verifies the
synchronized workflow but does not prove the original layout-race hypothesis.
Exact inventory matching establishes 534 unique current-runtime passes;
Playwright confirms 582 remaining tests in 136 files. Full stage 3 runs in
session 63876, /tmp/web-orbit-fixed-full-browser-3.log, using
/tmp/web-orbit-fixed-remaining.txt. Poll that handle before competing runs.
Full qualification, the intermittent nested-room Undo issue and broader NEXT
requirements remain incomplete.
The unchanged Chromium idle reproduction failed again after orbit (session
88961, terminal exit 1; /tmp/web-viewer-idle-repro.log). An isolated run with
the installed OrbitControls needs 165 frames to settle a large-scene orbit:
2.75 seconds at 60 fps, but 41.25 seconds at 4 fps. The viewer now scales damping
by elapsed time, preserving the 60 Hz response and restoring the configured
factor for pointer-event updates. Its clock resets when orbit rendering sleeps
or walkthrough takes over. The browser idle assertion has not been relaxed.
All 26 focused damping/framing unit tests passed, including equal camera position
after one second at 4–120 fps and settling within five seconds at those rates.
Log: /tmp/web-orbit-damping-unit.log. Type checking completed with zero errors
and warnings (session 79798, exit 0; /tmp/web-orbit-damping-check.log).
The production build passed (session 33825, exit 0;
/tmp/web-orbit-damping-build.log). Production source is 03e0ae1.
All 18 idle-rendering, framing and texture-recovery cases passed across all three
engines in 2.3 minutes (session 54192, terminal exit 0;
/tmp/web-orbit-damping-browser.log). The original idle assertion is unchanged.
The current 1,116-case inventory is /tmp/web-orbit-fixed-inventory.log.
Full stage 1 is terminal (session 23185, exit 1): 229 passed, one failed,
868 not run, log /tmp/web-orbit-fixed-full-browser-1.log. Together with the
18 focused passes, 247 distinct cases passed on this runtime. The PDF source
case failed at pdf-3d-source.spec.ts:31: its 10-second canvas visibility wait
expired while the snapshot still showed “Loading 3D viewer…”. The trace contains
no console errors. The unchanged Chromium reproduction passed in 48.8 seconds (session 62767,
exit 0; /tmp/web-pdf-source-repro.log). Its trace records 13.27 seconds for
the successful canvas visibility check despite the configured 10-second wait.
This test now allows 60 seconds for lazy viewer startup and 180 seconds for the
whole multi-export workflow; all canvas-source, page-count, recovery, notice and
failed-download assertions remain unchanged. Production source is unchanged.
All three PDF cases passed (session 77637, terminal exit 0, two minutes;
/tmp/web-pdf-source-all-engines.log), including source selection, recovery,
page counts and failure notices. Exact inventory matching establishes 250 unique
current-runtime passes. Playwright confirms 866 remaining tests in 136 files.
Full stage 2 runs in session 46921, /tmp/web-orbit-fixed-full-browser-2.log,
using /tmp/web-orbit-fixed-remaining.txt. Poll that handle before competing runs.
This production change invalidates prior-runtime browser pass credit: the 374
passes documented below belong to the thumbnail-fixed runtime, not this fix.
Full browser qualification, the intermittent nested-room Undo investigation,
and broader NEXT requirements remain incomplete.
The library now uses only own thumbnail entries, preventing missing previews
for __proto__, constructor, and toString from becoming inherited values
and invalid image requests. Imported IDs and project data are unchanged.
All nine thumbnail cases passed across Chromium, Firefox and WebKit: absent
previews, saved previews, and preview-read failures, including ordinary IDs.
All 15 save-conflict cases also passed. Type checking reports zero errors and
warnings; the production build passed. Production source is c14fce7.
Logs: /tmp/web-reserved-thumbnail-check.log, /tmp/web-reserved-thumbnail-build.log,
/tmp/web-reserved-thumbnail-browser.log (eight Chromium passes), and
/tmp/web-reserved-thumbnail-browser-2.log (16 Firefox/WebKit passes). All those
sessions are terminal. The Firefox request assertion exempts only the exact
favicon URL declared by the document; thumbnail and unexpected-request checks
remain intact.
The updated inventory contains 1,116 cases. The full run passed 243 cases, then
furniture category previews reached a 10-second canvas-readiness deadline.
The snapshot had mounted 3D controls; the canvas assertion had not completed.
That wait now allows 60 seconds within a 180-second workflow, retaining category,
metadata and exact model-download checks. Session 44047 is terminal; log
/tmp/web-thumbnail-fixed-full-browser-1.log. With the focused checks, 267
distinct current-runtime passes match the inventory and are retained in
/tmp/web-thumbnail-fixed-passed.txt. Playwright confirms 849 remaining cases;
stage 2 ran in session 99630, log /tmp/web-thumbnail-fixed-full-browser-2.log.
It terminated after 11 passes (5.8 minutes), at the mobile damaged-import
case’s 10-second canvas deadline while the snapshot showed the loading viewer.
The wait now allows 60 seconds within 180 seconds overall; exact project,
library, undo/redo and extension checks remain. There are 278 distinct retained
passes; Playwright confirmed 838 remaining cases. Stage 3, session 76367,
is terminal: 15 passed, then nested-room label Undo failed at room-slabs.spec.ts:99.
The label moved 40 pixels as expected but remained there after Undo. This is
a behavioral failure, not a viewer-readiness timeout. There are 293 distinct
current-runtime passes. Log: /tmp/web-thumbnail-fixed-full-browser-3.log.
The unchanged Chromium reproduction passed the entire workflow in 2.2 minutes
(session 32789, terminal exit 0; /tmp/web-nested-undo-repro.log). The traced
all-engine run also passed unchanged: Chromium 2.2 minutes, Firefox 44.9 seconds,
and WebKit 35.5 seconds (session 59904, terminal exit 0;
/tmp/web-nested-undo-all-engines.log). Saved traces are in
/tmp/web-nested-undo-passing-traces. The original intermittent Undo failure
remains unexplained and open; passing reruns are not proof of a fix.
There are 296 unique current-runtime passes (the Chromium repeat counts once).
Playwright confirmed 820 remaining cases. Stage 4 (session 42770) is terminal:
its first case, room slabs across floor switches, exhausted its 60-second total
budget during the second 3D export. The first export and slab checks had passed.
Log: /tmp/web-thumbnail-fixed-full-browser-4.log. That multi-export workflow
now uses test.slow() (180 seconds), as does the neighboring nested-room case
via its explicit timeout. Geometry, persistence, export and page-error assertions
are retained. No runtime change was made. The same 820 cases, confirmed again
by Playwright, ran in stage 5, session 89669, now terminal after 48 passes.
Log: /tmp/web-thumbnail-fixed-full-browser-5.log. The wall-photo 3D recovery
case never observed an active connected WebGL context within its 15-second
idle check. The trace shows mounted viewer controls and only the deliberately
aborted texture request as a console error. The 3D cases now allow 60 seconds
for the first connected draw before the unchanged 15-second idle check, within
180 seconds overall. Exact retry count, pixel-change and no-input redraw checks
remain. All six focused 3D cases passed in 3.3 minutes across Chromium, Firefox
and WebKit (session 24549, terminal exit 0;
/tmp/web-texture-startup-browser.log). There are 350 unique current-runtime
passes. Stage 6 (session 43409) is terminal after 24 passes, bringing the
unique current-runtime count to 374. Log: /tmp/web-thumbnail-fixed-full-browser-6.log.
The viewer-idle case failed after orbit: one animation callback remained pending
through its 40-second idle check. Trace samples show callbacks continuing to
advance (20 to 169), about four frames per second, rather than a stalled process.
The viewer uses fixed per-frame orbit damping (0.08); investigation must determine
whether slow damping accounts for the failure. The idle assertion is unchanged.
The unchanged Chromium reproduction (session 88961) also failed; see the
orbit damping fix above. Log: /tmp/web-viewer-idle-repro.log. This historical
thumbnail-runtime full suite is stopped.
Inventory: /tmp/web-thumbnail-fixed-inventory.log; remaining list:
/tmp/web-thumbnail-fixed-remaining.txt. Do not mix prior-runtime passes into
this run. Poll the active handle before competing browser tests or rebuilds.
Full qualification and broader NEXT requirements remain incomplete.
Browser qualification against 87d3c86 stopped after 343 distinct passes.
The unknown-furniture editing test found that the canvas caption displayed the
internal catalog ID instead of the established “Unknown furniture” fallback.
The canvas now supplies a translated caption only for a known catalog entry,
allowing the renderer to retain its unknown-item fallback. This is a production
change: the prior 343 passes do not qualify the corrected runtime.
Type checking passed with zero errors/warnings and the production build passed. Logs:
/tmp/web-unknown-caption-check.log and /tmp/web-unknown-caption-build.log.
All nine unknown-furniture editing and known-caption localization cases passed
across Chromium, Firefox and WebKit (2.6 minutes), log
/tmp/web-unknown-caption-browser.log; session 91439 is terminal.
Full qualification against corrected runtime b3c7fa6 passed 104 cases, then
the floor-elevation test hit a 10-second 3D readiness wait while loading. That
assertion now allows 60 seconds, with all geometry/reload/stack checks retained.
That continuation passed 52 more cases, then mobile legacy-furniture recovery
reached its 10-second 3D readiness deadline while loading. Both viewport variants
now allow 60 seconds for readiness within a 180-second workflow; recovery-byte,
model-download and editing assertions are unchanged. The next continuation passed
81 cases, then mobile same-ID import recovery reached its 10-second canvas wait
while the 3D viewer was loading. That wait now allows 60 seconds within its
existing 180-second workflow; persistence and copy-isolation checks are unchanged.
Stage 4 ended after 40 passes (277 total on runtime b3c7fa6), then the
save-conflict recovery test reached a 10-second canvas wait while loading. Its
readiness wait now allows 60 seconds within a 180-second workflow; all conflict,
backup, copy and reload assertions remain. Session 8168 and all previous
continuations are terminal. Log: /tmp/web-caption-fixed-full-browser-4.log.
The 277 passes do not qualify the subsequent thumbnail fix.
Poll the live handle before competing browser tests or production rebuilds.
The previous browser session 27671 is terminal (97 passes in its last stage).
See the prior qualification report.
Broader NEXT requirements remain open.
Furniture results now react to the selected language and share catalog names
and category translations. Original English names/categories and catalog IDs
remain search aliases; matching stays accent-insensitive. Three focused unit
tests pass, check reports zero diagnostics, build passes, and three browser
engine cases pass (1.5 minutes). The browser flow finds Fogão through fogao,
finds Poltrona through Portuguese, English and ID queries, executes with Enter,
and confirms the exported new object still has catalog ID chair with unchanged
walls. Full interface/fluent-language review, physical accessibility and wider
NEXT requirements remain open.
Vector furniture labels now use the selected language from both the export menu and command palette. The exporters accept an explicit locale with English as the default for existing callers; the DWG-to-DXF fallback passes it through too. Unknown-item fallback text and source project data remain unchanged.
All 28 focused export/catalog tests pass. The new comparison checks exact SVG equality apart from translated captions and every DXF tag apart from captions and freshly allocated entity handles/references, plus unchanged source JSON. Final production check reports zero diagnostics and build passes. All three browser engines passed real Portuguese SVG/DXF downloads and before/after floor equality (1.4 minutes). Other export wording, native CAD-reader review, fluent Portuguese review, physical accessibility and the wider NEXT backlog remain open.
The 2D editor passes localized furniture names into the pure caption renderer. Changing language invalidates the draw scheduler once through a cleaned-up subscription, so a stationary plan updates without an edit or camera movement. Unknown catalog entries retain the renderer’s “Unknown furniture” fallback after the regression fix recorded above. Renderer callers that omit a caption retain their existing default labels.
Six focused unit tests pass, check reports zero diagnostics, build passes, and all three browser-engine cases pass (1.5 minutes): actual canvas fillText records show English→Portuguese→English changes while Settings stays open, with exact exported floor equality. SVG/DXF label localization, broader export review, physical accessibility and the wider NEXT backlog remain open.
Source 5980437 passed all 1,034 unit tests in 98 files (76.05 seconds).
The process finished successfully; no unit runner remains active. This covers the
catalog-name and history-label changes together with the existing unit suite.
The latest scoped browser evidence remains 24 catalog cases and nine Layers/3D/
photo cases; the earlier 1,098-case full browser audit applies to ddf7d90.
The naming audit found remaining direct catalog-name reads in
canvasRenderer.ts (2D furniture captions), export.ts (SVG labels), and
cadExport.ts (DXF labels). Extend those deliberately with locale-aware redraw
and export checks while preserving geometry and unknown/source identifiers.
This checkpoint does not close those gaps, fluent review, physical accessibility,
native/device work or release gates.
Layers, the 3D furniture picker and its placement hint now share the Portuguese catalog names used by object cards and properties. Unknown identifiers retain their original text. Three focused unit tests, production check (zero diagnostics) and build pass. All nine affected Chromium/Firefox/WebKit cases pass (5.0 minutes), covering translated Layers selection, finish persistence, exact photo bytes and 3D placement with catalog identity and full Undo/Redo export equality.
The first Chromium 3D run exhausted its 60-second deadline during the final Redo export after earlier assertions passed. Its bounded slow-test allowance is now 180 seconds; the rerun retained every assertion and passed. This is scoped browser evidence, not a fresh full-suite checkpoint. Fluent Portuguese review, remaining interface text, physical accessibility and wider NEXT work remain open.
Added display names for all 191 built-in and import-preview catalog IDs. Object cards, recent items, favorites, hover previews, properties headings and generated Undo descriptions now use them in Portuguese. Search accepts both original English names and accent-free Portuguese names. Catalog IDs, dimensions, saved data, drag payloads and unknown/custom identifiers stay unchanged.
Three focused unit tests pass, including exact catalog coverage and unchanged catalog/history data. Production check reports zero diagnostics, build passes, and all 24 scoped Chromium/Firefox/WebKit cases pass (9.8 minutes): category and bilingual search, keyboard favorites/recent placement, focus, and Portuguese properties persistence. The full browser suite was not rerun. Remaining naming surfaces such as Layers and the 3D catalog, fluent Portuguese review, physical accessibility and the wider NEXT requirements remain open.
Undo display now recognizes the 15 generated descriptions for supported door, window and column types and uses the existing localized catalog/type labels. The lookup enumerates exact known descriptions; unknown types, appended text, English descriptions and stored history remain unchanged. Eight focused history tests pass, including real door/window/column actions with exact project/history preservation assertions. Production check reports zero diagnostics and build passes. This adds no fresh browser-suite evidence. Furniture catalog descriptions, broader interface review and physical accessibility remain open.
The display-only Undo map now also translates eight existing descriptions for furniture stacking order, item details, adding/reusing photos, deleting retained attachments and locking/unlocking selection. Stored descriptions and unknown text remain unchanged. Seven focused history unit tests pass; production build and the explicit-production Svelte check pass with zero diagnostics. No browser suite was rerun for this string-map extension; the full browser checkpoint below remains evidence for its named earlier runtime. Dynamic catalog descriptions, full interface review and physical accessibility remain open.
All 1,098 distinct browser cases in 136 files now pass against runtime ddf7d90
after the shared history fixes: 366 cases per engine, with exact inventory
agreement. This was staged qualification after correcting one stale Portuguese
test expectation. All processes are terminal; no browser continuation is active.
The same runtime passed all 1,032 unit tests, check and build. See
the completed report.
The broader backlog, physical-device and release gates remain open.
Reproduced unchanged groups clearing Redo and pending groups inserting an old project into Undo after a replacement project loads. Unchanged groups now leave history intact; loading clears pending group state and descriptions. Nested changed groups still commit one action and discard the superseded redo branch. All 1,032 unit tests in 97 files pass (24.08s), check/build pass without diagnostics, and 24 Chromium/Firefox/WebKit drag cases pass (3.3 minutes): leaving elevation during a window drag plus stair, column, text, endpoint, parallel-wall, curve and room drags with Undo restoration. Physical-device, release and broader NEXT requirements remain open.
Three failing regressions reproduced shifted Redo descriptions after history jumps (including an existing redo tail), and invalid indices changing project state. Jumps now preserve each action’s state, description and timestamp and reject noninteger/out-of-range indices before touching history. Intermediate project states are not published. All 1,029 unit tests in 96 files pass (25.52s), check/build pass with zero diagnostics, and all 24 bilingual desktop/phone history browser cases pass (2.4 minutes). This is a fresh full unit checkpoint, not a full browser-suite or physical-device qualification; broader NEXT scope remains open.
Reproduced keyboard focus disappearing when a selected history entry is removed. History activation now focuses Close after the DOM update, with guards against a closed panel or focus already moved elsewhere. Check/build pass with zero diagnostics and 24 bilingual Chromium/Firefox/WebKit cases pass (2.4 minutes). Populated-panel checks now cover 1440/390px, light/dark text contrast, mobile overflow entry, Enter activation, Escape dismissal, restored opener focus and exact exported floor restoration. Existing empty-panel keyboard cases also pass. Physical touch/assistive-technology qualification and the full native/release and remaining interface backlog are still open.
Added Undo History to the phone overflow menu. Opening the panel focuses Close; Escape and Close return focus to the desktop toggle or persistent More actions button. The panel remains nonmodal. A desktop regression reproduced missing focus before the fix. Check/build pass with zero diagnostics, and 18 bilingual Chromium/Firefox/WebKit cases pass (2.1 minutes): empty-panel entry, dismissal, focus restoration and horizontal bounds at 1440/390px, plus the populated desktop contrast and exact keyboard Undo/export checks. Populated phone interactions, physical assistive technology and the wider native/release backlog remain open.
Darkened faint history metadata and added a readable dark-theme current-state color and explicit keyboard focus outlines. All six English/Portuguese browser cases pass across Chromium, Firefox and WebKit (1.7 minutes). They measure at least 4.5:1 for populated-panel text in settled light/dark themes and activate the floor-history action with Enter, preserving exact exported floor restoration. Check reports zero diagnostics and production build passes. This covers the populated desktop panel; empty-state, phone and physical assistive-technology qualification and the wider NEXT backlog remain open.
Undo History now translates 33 recognized built-in descriptions for English and Portuguese at display time. Unknown descriptions and stored history remain intact. Six focused unit tests, check (zero diagnostics), production build and all six English/Portuguese Chromium/Firefox/WebKit history cases pass. The browser run took 4.5 minutes with a bounded slow-test allowance after the initial 60-second deadline expired during the final export; all preservation assertions remain. Dynamic action descriptions, broader interface/catalog review and physical-device and release qualification remain open. This checkpoint does not close the backlog.
Updated September 9, 2026. This is the current backlog for the web app and iPhone companion. It supersedes the historical “next” sections in the original review and batch log. Priorities below are proposed order, not release dates or a claim of complete Planner 5D parity.
For the archived September 8–9 scan/render/video session, start with the session handoff. It records the final video and Blender files, public default-model deployment, open experimental PRs and native release work that remains.
Start the next session with the native new-computer handoff and copyable prompt. It records setup, tested state, local data that does not transfer through Git, and the next bounded implementation batch. The native repository is private and requires an authorized GitHub account.
Rendering direction: Blender Cycles on the Mac for finished renders; Three.js for interactive web previews/editing/walkthroughs. Share the geometry, camera and texture preparation pipeline. Start with the full scan dataset and calibration validator, then the Blender path; keep the existing native preview and web renderer during that work. Browser path tracing is optional future scope. See the rendering plan. Keep full scans, photo textures, render jobs and outputs local by default.
The native continuation adds a normative v1 contract, bounded streamed local scan import/export, corruption/path-traversal regression coverage and explicit calibration/coordinate metadata. Complete Mac Catalyst and iPhone simulator suites each pass 82 tests, including a transferred legacy real scan: all 394 original files / 194 frame pairs retain identical bytes and hashes through export, independent import and re-export. Private scan content stays out of Git. See the native validation report and contract.
All transferred frames remain legacy-incomplete. Physical capture with the new
metadata and measured reprojection are still pending. Native issue #8 remains
open for physical calibration, captured-camera rendering and photo texturing.
The native original-scan queue now includes status/cleanup/limits UI, verified
image previews and save dialogs, plus submit/list/cancel/remove/export local
commands. A separate pinned Blender worker supports continuous queue processing,
immutable RoomPlan preparation, cancellation and crash recovery. 91 native tests
ran on each platform with one optional skip and zero failures; 40 renderer checks
and three CLI checks passed, including real Blender and Swift/Python handoff.
See docs/native-render-queue-validation.md in the native checkout. Full-scan
exports also handle macOS ancestor path aliases correctly. Signed worker
packaging, edited-plan adapters and measured photo projection remain open. The next
camera batch now supports validated captured-camera scene attachment, rectangular
renders/native previews and explicit landmark-error reports. 51 renderer checks
passed, including seven real-Blender cases; synthetic numerical error was below
0.001 px and actual raster-marker error below 0.14 px. See native
docs/captured-camera-validation.md. Native captured-frame selection is now
implemented with source-photo thumbnails, unavailable-frame reasons, frozen frame
metadata and local-command support. 96 native tests ran on each platform with
one optional skip and zero failures; 53 renderer checks and four CLI checks passed.
The native-selected camera completed a Blender render and its verified preview.
See native docs/native-frame-selection-validation.md. Photo/render comparison
now adds verified source-photo and opacity-overlay previews while keeping PNG
exports unchanged. 99 native tests ran on each platform with one optional skip
and zero failures; a fresh native Blender job and all comparison modes passed
live Mac QA. See native docs/photo-render-comparison-validation.md. Measured
real-image calibration remains open; legacy metadata is never inferred. Local
frame-quality analysis now adds cancellable detail/exposure/rotation/tracking
hints and optional picker sorting. 103 native tests ran per platform with one
optional skip and zero failures, plus synthetic Mac UI validation. See native
docs/frame-quality-ranking-validation.md. Real-scan ranking validation and
integrated coverage review remain open. A developer visibility stage now samples
full-height geometry across captured cameras, reports occlusion and union coverage,
and records geometric view candidates. Nine real-Blender acceptance cases cover
visibility, concave geometry and bounds. See native docs/scene-coverage-validation.md;
native single-camera coverage summaries are now integrated with captured jobs
and receipt-verified preview review. 106 native tests ran per platform with
one optional skip and zero failures; renderer tests and a fresh native/Blender
UI job passed. See native docs/native-coverage-review-validation.md. A native spatial
sample map now adds verified locations, state/surface filters, rotation controls
and a camera marker. 109 native tests ran per platform with one optional skip
and zero failures, plus renderer and live Mac map checks. See native
docs/coverage-sample-map-validation.md. Native combined review now compares
2–16 verified captured frames of the same unchanged scan, with union/overlap counts,
unique contributions and a map with all selected camera markers. 112 native
tests ran per platform with one optional skip and zero failures; live two-camera
results matched independent counts. See native docs/combined-coverage-validation.md.
Quality-aware suggestions now combine verified source-photo hints with coverage
gains and offer a reviewable smaller selection. 115 native tests ran per platform
with one optional skip and zero failures, plus live ordering/cap/apply checks.
See native docs/coverage-view-suggestions-validation.md. Direct developer batch
analysis now prepares RoomPlan once without rendering and preserves every camera
state in a bounded report. All 72 renderer cases passed across host/Blender execution;
synthetic batch output matched both prior single-camera reports. See native
docs/direct-coverage-batch-validation.md. Monitored coverage-only queue jobs now
add immutable multi-frame inputs, worker limits, cancellation and verified receipt
recovery. The full 76-case renderer run and final five-case coverage-job suite passed.
See native docs/coverage-only-queue-validation.md. Native distinct receipts and
verified batch summary/map review now work in mixed queues. 120 native tests ran
per platform with one optional skip and zero failures, plus live Mac checks.
See native docs/native-coverage-batch-review-validation.md. Native multi-frame
submission now freezes cameras from one scan and records verified photo identities.
123 native tests ran per platform with one optional skip and zero failures;
a native → Blender → native job passed with unchanged photos. See native
docs/native-coverage-submission-validation.md. Batch subset review now adds
coverage-loss counts, recalculated contributions, selected maps and per-camera
detail. 126 native tests ran per platform with one optional skip and zero failures,
plus live select/clear/restore checks. See native docs/coverage-batch-subsets-validation.md.
Local-command coverage submission now uses the native verification path with
existing list/cancel/remove support. 128 native tests ran per platform with one
optional skip and zero failures; six CLI tests passed, plus live command/worker/UI
validation. See native docs/coverage-command-validation.md. Quality-aware batch
suggestions now verify recorded photos, report coverage omissions and apply a
review-only subset. 131 native tests ran per platform with one optional skip
and zero failures, plus live ordering/cap/apply checks. See native
docs/coverage-batch-quality-validation.md. Per-camera review now measures local
photo detail/exposure at visible wall and floor samples, with explicit edge/proxy
exclusions. 135 native tests ran per platform with one optional skip and zero
failures, plus live synthetic patch review. See native
docs/surface-photo-evaluation-validation.md. Batch suggestions now weight each
measured local patch and keep geometric coverage counts independent; absent patch
scores never inherit global photo quality. 137 native tests ran per platform with
one optional skip and zero failures, plus live ordering/limit/apply checks. See
native docs/local-patch-suggestions-validation.md. New batches now preserve validated
surface normals, and native photo review displays viewing angle and source pixel
density. 140 native tests passed per platform with one optional skip; the pinned
Blender renderer suite completed 78 host tests with 11 skips and no failures. See
native docs/surface-projection-geometry-validation.md. Batch suggestions now combine
local photo quality, incidence cosine and relative
projected resolution, with explicit photo-only mode for older batches. 142 native
tests ran per platform with one optional skip and zero failures; see native
docs/geometric-camera-suggestions-validation.md. A separate developer photo-projection command now bakes a calibrated sRGB PNG into
wall/floor UV islands with per-texel occlusion checks, neutral fallback, UV padding,
coverage mask, packed Blender scene and preview. 84 renderer host tests passed
with 13 delegated skips and no failures, including actual Blender photo bakes and
renders. See native docs/photo-projection-prototype-validation.md. Native queue and
photo-conversion integration, masks, seam handling,
physical ranking validation, continuous heatmaps and measured calibration remain open.
The prototype now combines up to eight calibrated photos using per-texel local
quality, viewing angle and projected resolution, with source identities and a texel
provenance map. 88 renderer host tests passed with 15 delegated skips and no
failures, including reversed-order atlas equivalence and actual Blender bakes.
See native docs/multi-photo-projection-validation.md. Explicit per-photo keep masks
now exclude marked regions with exact dimensions, a two-pixel guard band and recorded
mask identities. 91 renderer host tests passed with 16 delegated skips and no
failures, plus a visual exclusion check. See native
docs/photo-exclusion-masks-validation.md. Automatic semantic detection, mask-authoring
UI, seam blending, native integration and physical validation remain open.
Optional overlap normalization now estimates bounded linear RGB gains from valid
unmasked overlap, with correction/rejection reports and unchanged coverage/source
labels. 94 renderer host tests passed with 16 delegated skips and no failures,
including actual corrected previews and reversed-order atlas equivalence. See native
docs/photo-normalization-validation.md. Physical color validation, global alignment,
seam handling and native integration remain open. Optional UV-island overlap blending
now uses linear light and coverage feathering, preserves masks/neutral fallback and
records all contributing sources. 98 renderer host tests passed with 16 delegated
skips and no failures, including normalization, mask preservation and reversed-order
atlas equivalence. See native docs/overlap-blending-validation.md. Seam handling
across islands, native integration and physical validation remain open.
Rendering stays local: Blender Cycles for finished renders and Three.js for web
previews with shared preparation. This batch changes no web runtime, Firebase
storage/quotas, project-package format or rendering engine.
Undo History now exposes a translated named region and current-step marker to assistive technology, and uses semantic timestamps formatted for the selected language. Zero-error/warning application checking, the production build and six English/Portuguese browser cases pass across three engines (3.8 minutes). Tests add a floor, navigate back through history, compare original floor exports and verify panel close/expanded state. This is scoped verification after the previous full checkpoint; action-description translation and physical accessibility review remain open.
Canvas accessible-name qualification is complete: all 141 affected cases, 47 per engine, have passing evidence with exact inventory agreement. The final 116-case continuation passed in 28.5 minutes; no browser process from this batch remains active. These are staged results across the documented builds, not a fresh run of the entire browser suite. The current source also passes a fresh full unit run: 1,025 tests in 94 files, 125.13 seconds. Application checking has zero errors/warnings and the production build passed. See the consolidated canvas accessibility report. The pending-run paragraphs below retain historical checkpoints. Broader catalog and interface review, physical accessibility/performance, native and release requirements remain open.
Firefox canvas qualification checkpoint: all 47 affected Firefox cases now
have passing evidence, matching Chromium’s completed 47. Exact inventory comparison
finds 97/141 passing identities (47 Chromium, 47 Firefox, three WebKit), leaving
44 WebKit cases at this checkpoint. The same continuation remains active in
/tmp/web-canvas-label-browser-2.log. Resume it before another build or browser
suite. Full affected-file qualification is still pending.
Chromium canvas qualification checkpoint: all 47 affected Chromium cases now
have passing evidence across the initial run, focused room-keyboard rerun and
continuation. Exact identity comparison finds 51/141 total cases passed (47
Chromium, two Firefox, two WebKit), leaving 90. The continuation is still active
in /tmp/web-canvas-label-browser-2.log and has moved to Firefox. Keep this
process running; do not rebuild or start another browser suite until it finishes
or fails. These are staged results, not a single fresh run or completed overall
qualification.
Room-keyboard verification now passes: all six desktop/phone cases across
Chromium, Firefox and WebKit passed (4.4 minutes) on the rebuilt panel-cleanup
source. Combined with the initial canvas run, 25 distinct cases have passing
evidence. An exact inventory comparison leaves 116 cases; their continuation is
now running in /tmp/web-canvas-label-browser-2.log, selected by
/tmp/web-canvas-label-remaining.txt. Resume this process before rebuilding or
starting another browser suite. This staged qualification is still incomplete.
Canvas qualification continuation: the original 141-case run stopped after
19 Chromium passes when the room-keyboard regression exceeded its 60-second
budget during repeated exports. A focused rerun reproduced the timeout. The test
now has Playwright’s bounded slow-test allowance, retaining all assertions. The
first desktop Chromium case passes in 1.4 minutes; the six-case focused run is
still active in /tmp/web-canvas-room-keyboard-bounded.log. Resume it before
rebuilding or starting another suite. The remaining affected-file qualification
is not complete; the earlier paragraphs below are historical checkpoints.
UndoHistoryPanel and SettingsDialog now release four store subscriptions when destroyed, preventing closed editor instances from continuing to receive updates. Svelte reports zero errors/warnings and the production build passes. The focused browser run above uses this rebuilt source. Direct memory profiling and completion of the affected browser inventory remain open.
Canvas accessible-name verification is in progress. The drawing canvas now
uses the selected language for its accessible name; English remains unchanged.
Seventeen affected bilingual browser test files now accept either language, and
the desktop/phone canvas-hint cases explicitly require the Portuguese name.
Five dictionary tests, zero-diagnostic Svelte checking and the production build
pass. The 141-case affected-file browser run is still active, with its first ten
Chromium cases passing at this checkpoint. Do not treat this as completed browser
qualification. Resume the existing process and inspect
/tmp/web-canvas-label-browser.log before starting another suite or rebuilding.
Remaining version-history diagnostics now translate missing history, changed or foreign snapshots, unreadable versions and the unchanged-current-plan outcome. Forty-seven focused unit/dictionary checks, zero Svelte diagnostics and the production build pass. Nine history browser cases pass across three engines (1.2 minutes), verifying rejected foreign-project restoration, unchanged project and history records, recovery download and before/after floor exports. The first run exposed a test baseline that omitted the legitimate Session start snapshot; the corrected test waits for it and verifies exact preservation thereafter. This is scoped verification; other localization and physical/native/release gates remain open.
The recovery-error gaps listed below now have display translations, and the
version-history panel applies the shared translator. Forty-six focused unit and
dictionary checks pass, Svelte reports zero diagnostics, and the production build
passes. Six history browser cases pass across three engines (1.4 minutes), checking
translated damaged-history guidance, exact backup text, canceled deletion and
successful version restoration. This is scoped verification after the full unit
checkpoint below. Missing-history and failed-restore outcome messages in
src/lib/stores/versionHistory.ts remain to be translated; broader gates stay open.
Fresh full unit checkpoint: all 1,021 tests in 94 files pass on 5ca61a4
with npx vitest run --maxWorkers=1 (106.30 seconds). This verifies the accumulated
validation/localization batches together. The existing successful check/build and
15 transfer browser cases also apply to this source; the full browser audit remains
the earlier staged checkpoint, not a fresh full browser run.
The next recovery-message gaps are now identified in source: blocked database
upgrades, unreadable legacy libraries and recovery-ID allocation in
src/lib/services/localDatabase.ts; save-copy ID allocation and saved-project ID
mismatch in src/lib/services/datastore.ts; restored-ID allocation and recovery
archive preservation in src/lib/services/libraryRestore.ts; item-detail/retained
state errors in src/lib/utils/itemDetails.ts; and unreadable/oversized history
in src/lib/utils/snapshotStorage.ts. Their error strings are absent from the
current dictionaries and still need display translation and recovery verification.
These remaining gaps do not change the broader physical/native/release scope.
General project-validation explanations now translate while preserving exact field paths and unknown details. Forty-three focused unit/dictionary checks, zero Svelte diagnostics, the production build and 15 transfer browser cases pass across three engines (2.5 minutes). Mixed-backup verification includes a Portuguese field error and exact invalid-project bytes in the recovery archive. This scoped post-audit batch does not close other service diagnostics or physical/native/release requirements.
The native package bridge’s geometry/reference, invalid identity-map and duplicate identity diagnostics now translate into Portuguese. Thirty-six focused unit and dictionary checks pass; Svelte reports zero diagnostics and the production build passes. All 15 transfer browser cases pass across three engines (2.6 minutes), including rejection of a negative wall height with unchanged saved records and successful subsequent import. This is scoped post-audit verification; general project-validator field messages and physical/native/release work remain open.
Package preview notices and service-level attachment/retained-data diagnostics now translate at display time, preserving filenames and unknown details. Thirty-three focused unit/dictionary checks, zero Svelte diagnostics, the production build and 15 transfer browser cases pass across three engines (2.6 minutes). Tests verify missing-attachment rejection, both preview notices, exact original-package download and retention of unpreviewable image bytes after import. This is scoped post-audit verification; deeper native-plan validation and broader gates remain.
Package diagnostics now translate the invalid-package prefix, known ZIP/JSON validation, size limits, unsupported manifests and damaged filenames. Unknown details remain intact. Thirty focused unit/dictionary checks, zero Svelte diagnostics, the production build and 15 transfer browser cases pass across three engines (2.4 minutes). Rejected packages leave records unchanged and a subsequent valid package imports successfully. This is scoped post-audit verification; native-plan/attachment-specific validation and package preview notices remain open.
Backup preview warnings now translate known damaged-project/version counts, missing attachments, recovery archives, unreadable history, mismatched IDs and unsupported thumbnails. Twenty-three focused unit/dictionary checks, zero Svelte diagnostics, the production build and 12 transfer browser cases pass across three engines (1.9 minutes). A mixed-backup restore verifies original downloads and exact damaged records in the stored recovery archive. This scoped post-audit batch leaves general project-validator diagnostics and package-specific messages open.
Library-backup validation now translates malformed JSON, unsupported versions, invalid saved-text sections, empty-backup errors and duplicate-key diagnostics. Literal key contents remain unchanged. Fifteen service-message unit cases and five dictionary checks pass, along with zero Svelte diagnostics, the production build and nine transfer browser cases across three engines (1.7 minutes). Rejected uploads retain their original downloads and leave project/history records intact. This is scoped post-audit verification; package-specific validation and preview warnings remain open. See the localization integration record for evidence.
Restore-library and project-package dialogs now translate known storage failures together with their retry guidance at display time. Unknown diagnostic details remain intact. Six focused unit cases, zero-error/warning Svelte checking, the production build and 12 browser cases pass across all three engines (3.4 minutes). Quota-injected transfer checks verify unchanged project/history bytes, retained original downloads and successful retry. This is a scoped runtime change after the browser audit checkpoint below; other service validation messages and warnings remain to be translated.
Browser audit checkpoint: every case in the expanded inventory has passing evidence: 1,059 distinct cases, 353 per engine. The final inventory comparison found no missing or extra cases. This aggregates staged runs and focused follow-ups across the documented runtime versions; it is not one uninterrupted full-suite result. See the audit report for provenance and limits. That audit checkpoint also has 987 passing unit tests, zero Svelte diagnostics and a successful production build. Physical-device, native, performance, usability and release/cost gates remain open. The paragraphs below retain earlier checkpoints and their then-pending counts.
The ninth audit passed 156 WebKit cases before a test-only IndexedDB observer raced app hydration and created an empty database. The helper now aborts schema creation, rejects failed reads and closes connections; transfer tests wait for the library’s ready state. Nine helper/transfer cases pass across all engines (1.7 minutes), including a direct pre-initialization regression. No runtime code changed. Combined evidence covers 1,033 of the expanded 1,059 browser cases, leaving 26 audit cases; physical-device and release gates remain open.
The eighth browser audit added 12 WebKit passes before the phone Layers test clicked a status control while its horizontal scrolling layer had stale hit-test positions. A focused reproduction failed on a different toggle. The test now scrolls each control into view, waits two frames and verifies its click point hits the control before clicking. All six desktop/phone cases pass (2.6 minutes), plus three repeated phone WebKit cases (1.0 minute). No runtime code changed. Combined audit evidence covers 873 of 1,056 cases, with 183 remaining.
All 21 project-opening cases pass across three engines on the translated production build (6.6 minutes). The multi-stage same-ID import/reload/3D case now uses Playwright’s bounded slow-test budget after traces showed the default minute expiring during renderer/screenshot work; all assertions and screenshots remain. Combined audit evidence now covers 860 distinct cases in the expanded 1,056-case inventory. The next continuation has 196 cases remaining; physical-device and release gates remain open.
Project-service translation now has a successful production build and nine passing browser cases across Chromium, Firefox and WebKit (3.0 minutes): desktop/phone quota-blocked opening, reactive language changes, original-byte preservation, backup/retry/import and existing save-status localization. The phone test now opens Settings from More actions. Broader opening qualification remains pending after one diagnostic-screenshot timeout in the first combined run.
The seventh audit stopped after 577 additional passes on WebKit’s native form Undo grouping. A plain HTML form reproduces its grouped note/numeric-draft Undo. The keyboard regression now saves/verifies/reloads numeric edits before testing native text history, explicitly checks historyUndo/historyRedo and exact Redo text, and retains the final geometry/save/reload assertions. All six desktop/ phone cases pass across three engines (3.7 minutes). Accumulated original-audit coverage is 846 of 1,050 cases; 204 remain. The seventh run is finished.
Known project-opening and storage diagnostics now translate at display time in
the import alert, save banner, library actions and editor load recovery. English
service diagnostics and unknown detail strings remain intact. Four focused unit
cases pass and Svelte checking reports zero errors/warnings. Production build
and browser verification of this new localization batch are pending until the
active seventh audit finishes; that audit still serves the 78d3ad0 runtime.
The separate project-service-localization.spec.ts now collects six desktop/
phone-width cases across the three engines for quota-blocked import/New Project,
reactive error language, original-byte preservation, JSON backup, retry and
successful later import. Collection is verified; execution is still pending.
The newly added cases are outside the active audit’s original 1,050-case inventory.
Full unit verification of this source now passes 987 tests in 94 files with
one worker and unchanged timeouts. The initial parallel run had two five-second
timeouts; both passed in the full serial retry. Production/browser verification
of the translation change remains pending behind the active WebKit audit.
The seventh browser audit has now completed Firefox as well: accumulated audit and focused follow-up evidence verifies 350 Chromium and 350 Firefox cases. WebKit is running, with no failure yet in this continuation. The full 1,050-case qualification and physical-device gates remain open.
Browser audit at d6e33d1: the seventh continuation completed the remaining
Chromium cases. The accumulated audit and focused follow-ups now verify all
350 Chromium cases in the current 1,050-case inventory. Firefox and WebKit
qualification remains in progress; the seventh run is still active. This is a
Chromium checkpoint, not completion of the full browser or physical-device gates.
Browser audit at 1dfd862: another 48 Chromium cases passed before the Portuguese
print test expected the obsolete English room count. Its expectation now matches
the Portuguese UI. All three print cases pass (34.2 seconds), retaining paper
ratio, scale rejection, PDF content and translated print-caption checks. No runtime
code changed. Accumulated qualification covers 267 distinct cases; full browser
qualification remains unfinished.
Browser audit at b9452d0: the phone group-drag case targeted a stair covered by
the Properties sheet. It now uses Fit selection and asserts the drag starts on
the canvas. All six desktop/phone-width group cases pass across three engines
(16.3 seconds), preserving movement, locked-item and Undo assertions. No runtime
code changed. Accumulated qualification covers 216 distinct cases; the complete
browser suite remains unfinished.
Browser audit at 31f9ba4: another 47 Chromium cases passed before a Portuguese
phone-layout modal test searched for an obsolete English panel label. Its selector
now follows the current locale. All 12 EN/PT desktop/phone-width modal cases pass
across the three engines (20.9 seconds), retaining cancellation, focus and saved
data checks. The accumulated audit covers 210 distinct cases; full qualification
remains unfinished. No editor code changed.
Browser audit continuation at ef2feff: 27 more Chromium cases passed before a
curve-handle drag coordinate failure. The geometry-drag test now lets panel
resize/fit redraws settle before converting world coordinates into pointer
positions. All 21 geometry-drag cases pass across the three engines (46.1 seconds),
including exact Undo/Redo for stairs, columns, text, wall endpoints, parallel walls,
curves and rooms. No editor code changed. The accumulated audit/follow-up evidence
now covers 154 distinct cases; the complete 1,050-case suite remains unfinished.
Browser audit continuation at c47207e: 47 more Chromium cases passed before a
save/reload timing failure in the sloped-wall test (102 distinct audit passes so
far). Slope/elevation tests now await completed saves, including the hidden status
text on phone layouts, and await an imported copy before selecting its floor.
All nine slope/elevation cases pass across the three engines (1.5 minutes).
This changes test synchronization only; the full browser audit remains unfinished.
Broader browser audit at 78d3ad0: the 1,050-case run stopped after 55 Chromium
passes on a room-label coordinate check following a floor switch. The test now
waits for the queued initial fit/redraw before clicking a rendered label, retaining
its exact name and cross-floor checks. Three cases pass, followed by nine repeated
cases across all three engines (37.2 seconds). No runtime change was made.
The complete browser audit remains unfinished; catalog manifest validation passed.
Touch cancellation now releases the mouse pipeline without synthesizing clicks or double-taps. Drags beyond 10 screen pixels also break the tap chain, including a drag that returns to its start; ordinary double-taps remain supported. Nine browser cases pass across all three engines (26.6 seconds), covering synthetic touch classification, pinch/pan idle behavior and delayed tracing-image floor isolation. Check/build pass with zero Svelte diagnostics. Actual device gestures remain open.
Follow-up curve hit-testing performance: a conservative bounds check now skips distant curves before exact projection. Local 400-wall median lookup cost fell from roughly 0.75ms to 0.002–0.006ms; see the reproducible measurement and its scope limits. All 983 unit tests pass (3.68 seconds), check/build pass, and 21 browser cases pass across all three engines (1.4 minutes), including curve click/drop, automatic dimension fit and canvas idle/wakeup behavior. The duplicate visibility label now distinguishes automatic dimensions from saved dimension annotations in EN/PT.
Latest full web unit checkpoint: 983 tests across 93 files passed with continuous curved-wall hit testing and opening positioning (3.82 seconds). Exact-path projection replaces the 20/40-point sampling used for selection and placement. Fifteen browser cases pass across all three engines (54.1 seconds), covering curved opening click/drop, Undo/Redo and curved split persistence. Check/build pass with zero Svelte diagnostics. Physical gesture/performance qualification remains.
Earlier full web unit checkpoint: 976 tests across 92 files passed with curved door/window drag placement (3.84 seconds). Drops now use the closest point on the quadratic wall path instead of the endpoint chord, retaining the endpoint margin and drop radius. Nine browser cases pass across all three engines (21.9 seconds), covering both opening kinds, Undo/Redo and existing catalog interactions. Check/build pass with zero Svelte diagnostics. Physical touch remains unqualified.
Earlier full web unit checkpoint: 967 tests across 91 files passed with furnished room layout corrections (3.82 seconds). Default 400×300cm templates now keep catalog footprints inside the walls without furniture overlap and apply specified rotations. Nine browser cases pass across all three engines (20.1 seconds), with click/drag placement, bedroom rotations, Undo/Redo and save/reopen coverage. Check/build pass with zero Svelte diagnostics. Circulation, physical-device use and arbitrary room-size layout adaptation remain unqualified.
Earlier full web unit checkpoint: 960 tests across 90 files passed with furniture context-menu mirror and stacking fixes (3.92 seconds). Check/build pass with zero Svelte diagnostics; six browser cases across Chromium, Firefox and WebKit pass (21.8 seconds). Mirroring preserves signed scale, and front/back changes now use Undo history without clearing Redo for unchanged order.
Rotated furniture placement now groups creation and its initial angle into one Undo entry. Nine browser cases pass across all three engines for zero/30-degree and wall-snapped placement, exact removal on Undo, restoration on Redo and save/reopen (33.6 seconds). The wall case checks that snapping overrides the preview angle and keeps the sofa flush with the wall surface. Check/build passed for the implementation; physical gestures remain separate qualification cases.
Furnished room-template dragging now places the selected template at the drop point. Previously the draggable cards emitted a type the canvas did not handle. Nine browser cases pass across all three engines (19.2 seconds), covering desktop drag placement/Undo/Redo/save-reopen and existing desktop/phone-width click placement. Check/build pass with zero Svelte diagnostics. Physical touch dragging remains unqualified.
Earlier full web unit checkpoint: 958 tests across 90 files passed with curved wall splitting (3.65 seconds). Check/build and 12 wall-action browser cases pass across Chromium, Firefox and WebKit (54.4 seconds), including curved split controls, opening/room/group preservation, Undo/Redo and save/reopen. Device/release gates remain.
Follow-up curved-split mesh qualification passes in all three browser engines (three cases, 17.1 seconds). After a toolbar split, exported meshes retain clear sampled door/window apertures, solid wall at the split and trim along the curve, including stacked floors and active-floor changes. This checks exported geometry; pixel-level rendering and physical-device qualification remain open.
Earlier full web unit checkpoint: 955 tests across 90 files passed at f54d739
(3.63 seconds). Three sloped-wall browser cases pass across Chromium, Firefox and
WebKit (22.3 seconds), covering height edits, opening preservation, Undo/Redo,
elevation, save/reload and stacked-3D entry. No runtime changes in this checkpoint;
physical-device and broader release gates remain open.
Earlier full web unit checkpoint: 952 tests across 90 files passed at 9620cb2
(3.60 seconds). Six crossing/curved-room browser cases pass across Chromium,
Firefox and WebKit (25.0 seconds), checking exported 3D slab geometry through
stacking and active-floor changes. No runtime changes in this checkpoint;
physical-device, native area agreement and release gates remain open.
Earlier full web unit checkpoint: 949 tests across 89 files passed with the unchanged-room update guard (3.62 seconds). Check/build and six room-menu browser cases at 1440px/390px across three engines pass (26.8 seconds), including Redo preservation after accepting an unchanged room name. Full release gates remain.
Earlier full web unit checkpoint: 945 tests across 88 files passed with the room-deletion metadata fix (3.20 seconds). Check/build and six room keyboard-menu cases across three browser engines at 1440px/390px pass (24.6 seconds), including label reset, full deletion and exact Undo. Device/release gates remain open.
Earlier full web unit checkpoint: 943 tests across 87 files passed at 6a59c00
(5.66 seconds). Nine existing English browser cases pass across Chromium,
Firefox and WebKit (1.4 minutes): modal keyboard isolation at 1440px/390px,
stored/exported floor preservation, and room-label positioning, reset, drag and
Undo. This integrates the recent context-menu and rename focus changes; full
browser/device/release qualification remains open.
Earlier full web unit checkpoint: 943 tests across 87 files passed with the opening-safe wall split change (5.06 seconds). Check/build and nine contextual wall-action browser cases across Chromium, Firefox and WebKit also pass (52.9 seconds). Splits crossing opening interiors are rejected without mutation; exact opening-edge splits remain supported. Full browser/device/release gates remain.
Earlier full web unit checkpoint: 939 tests across 87 files passed at 85f42dc
(4.48 seconds). Six existing English wall-dimension cases pass across three
engines at desktop/narrow widths (2.3 minutes), retaining connected edits,
opening values, invalid-input recovery, imperial precision and persistence.
See the localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 939 tests across 87 files passed at 6026e66
(21.12 seconds). Six existing camera resource cases pass across all engines at
1440px and 390px (2.6 minutes), covering repeated preview/capture/reposition and
2D/3D transitions. See the
localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 939 tests across 87 files passed at 97e7b2b
(22.56 seconds). This includes the floor-transition calibration regression. Six
existing English sloped-wall and modal/elevation/3D/print browser cases pass across
all engines (1.8 minutes). See the
localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 82aa35f
(20.05 seconds). Six attachment failure/selection-change cases pass across all
engines (1.1 minutes), checking saved-data preservation, exportable drafts, retry
and cancellation of an obsolete photo decode. See the
localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 4cc7a2c
(24.53 seconds). Six existing English wall-dimension cases pass across all engines
(2.6 minutes), checking desktop/compact connected edits, invalid drafts, imperial
conversion, full-precision undo, persistence and 3D viewer opening. See the
localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 31b7486
(20.85 seconds). Eighteen existing English background-image browser cases pass
across all engines (1.1 minutes), covering initial framing, delayed-image priority
and failed-image fallback at desktop/narrow widths. See the
localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at b671dfc
(4.74 seconds). Fifteen existing English browser integration cases pass across all
engines (3.4 minutes), covering desktop/narrow furniture tint, finish, reload and
resource reuse, plus nested/disconnected floor slabs and active-floor switches.
See the localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at dae6421
(runtime source 950fc36, 7.02 seconds). Twelve symbol-export browser cases pass
across all engines (1.2 minutes), covering PNG/SVG/PDF image readiness, framing,
embedding and failed-image export handling.
See the localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 8b8e707
after Build tools, opening catalogs and room-choice localization (5.03 seconds).
Twelve English browser cases pass across all engines (1.1 minutes), covering
catalog/recent favorites and RoomPlan/template modal cancellation and focus.
See the localization integration record.
Full-browser and physical-device qualification remain open.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 6efa2a7
after Layers and canvas-control changes (3.41 seconds). Nine English browser cases
also pass across all engines (1.3 minutes): large-plan zoom at both widths and
sloped-wall edits, reversal, elevation, reload and stacked-view navigation.
See the localization integration record.
This does not qualify a full browser suite, physical device or deployment; the
previous intermittent WebKit grid-click observation remains unresolved.
Earlier full web unit checkpoint: 938 tests across 87 files passed at 7809b06
after export, view, floor and save-toolbar localization (11.89 seconds). Nine
English browser cases also pass across all engines (1.2 minutes), covering
failed-save/deployment recovery and floor-camera preservation at both widths.
See the localization integration record.
This does not qualify a full browser suite, physical device or deployment.
Earlier full web unit checkpoint: 938 tests across 87 files passed at e12cff4
after printed-sheet and typed-notice localization. The English failed-save/reload
flow also passed across all three engines (three cases). See the
localization integration record.
This does not qualify a full browser suite, physical device or deployment.
Earlier full web unit checkpoint: 936 tests across 87 files passed at 31e263d
after the initial localization batches. The existing English palette/modal-field
flow also passed in all three engines at desktop/phone widths (six cases). See the
localization integration record.
This is not a full-browser, physical-device or deployment qualification.
Earlier full web unit checkpoint: 931 tests across 86 files passed at faaa9cd
on September 10, 2026, after texture and furniture download recovery changes.
This is not a full-browser or deployed-release claim; see the
integration record.
The user’s new priority is a universal iPhone/iPad/Mac app with local commands and eventual photo-based RoomPlan rendering. The native implementation plan is tracked in native issue #8 and the detailed roadmap. Start with Catalyst and local saved-plan commands, then a portable full scan dataset with calibrated photographs, a separate local Blender worker, and measured photo projection/texturing. Native source is in the currently private companion repository; this is separate from web delivery or an App Store release. Keep scan photos, intermediate assets and rendering off Firebase by default. The existing 64 MiB project ZIP is not the full scan dataset format.
The furniture category batch for #63 is implemented in both repositories. Package/RoomPlan imports share category rules, native display aliases recognize web IDs, unknown categories remain identifiable, and imported stairs have a procedural preview. Source categories, IDs, fractional dimensions and metadata survive actual native return packages. See the batch report and GitHub PR checks for merge/release status and final browser CI results.
The browser compatibility batch for #65 adds the full CI suite to Chromium, Firefox and WebKit, fixes canvas shortcuts intercepting field editing, and preserves furniture dimensions while replacing empty/invalid drafts. See the browser report and PR checks for final engine results and merge/release status.
Local validation: 654 web unit tests; native dataset work passes 82 XCTest tests on Mac and 82 on the iPhone simulator. Production web build and audit pass; type checks report zero errors and zero Svelte warnings. Desktop and phone-width browser checks cover labels, editing, persistence and 3D. Native source availability remains separate from TestFlight/App Store release.
Already delivered: storage safety and recovery; connected editing and numeric dimensions; named-room exports and physical PDF scale; dependency remediation; floor elevations and sloped walls; direct AI provider configuration; safe imports and project switching; local tab conflict recovery; IndexedDB migration; full library backup/restore; two-way local iPhone/web packages; editable item notes/photos/costs and pooled attachment history; furniture appearance fixes; category continuity; field keyboard editing and browser-engine CI; camera preview and 3D resource cleanup; repeatable furnished-home benchmarks and preservation of 3D views during metadata edits; responsive top-down camera framing; onboarding hints that stay within resized viewports; idle 3D animation cleanup measured in native Safari; walkthrough timing, held-input recovery and stationary rendering cleanup; 2D drawing on demand with explicit display/image wakeups; modal keyboard protection; keyboard-accessible library actions with explicit, recoverable rename/delete dialogs. See the library actions report and PR checks for final browser CI and deployment verification. Earlier batches and pause hashes are recorded in the dated review log and git history.
The deployment check for #81 bypasses stale size/mtime validators when reading the version file. Native Safari confirmed the stale cached response and the fixed editor’s unconditional request. The same polling limits, immutable asset caching and save-before-reload recovery remain. Real HTTP-cache regressions cover equal-size version replacements and recovery. See the report and #83 for final CI, native Safari and deployment verification.
The measured resource batch for #67 repairs blank reopened camera previews, releases renderer contexts and replaced scene textures, and disposes/reapplies wall highlights through rebuilds. The pre-fix browser measurements confirmed retained contexts and texture growth. See the resource report and PR checks for final validation and merge/release status.
The furnished-home batch for #69 adds deterministic small/medium/large fixtures, desktop/DPR-2 phone-viewport CI measurements, and a scene snapshot that avoids rebuilding on project names and item notes/costs/photos. Geometry, finishes, history and area-unit changes still refresh. See the benchmark report for results and measurement limits; CI software rendering is not a device budget.
The top-down framing batch (#71) fits both screen axes, reserves vertical overlay space, clears pending orbit motion and includes distant/stacked geometry in the visible depth range. Corner projection and browser pixel checks cover portrait, desktop and landscape layouts. See the framing report and PR checks for validation and release status.
The onboarding hint batch (#73) tracks viewport changes and measured hint bounds, keeps the dismissal button visible on short screens, and cleans up animation/timer callbacks. Resizing does not restart the eight-second timeout; manual and automatic dismissal still retain seen-tip behavior. See the hint report and PR checks for browser validation and release status.
The idle-rendering batch (#75) replaces continuous orbit polling with requested frames that stop after damping. Native Safari on M4 Max recorded 1,500 idle callbacks before the change and zero after it in matched 25-second intervals; a post-orbit repeat also returned to zero. See the report and sanitized metrics. The regression checks controls, scene changes, placement previews and teardown in all three engines. It also fixes the desktop Help button covering Lighting Controls. These results establish idle behavior, not general FPS or battery-life targets.
The walkthrough timing batch (#77) uses elapsed animation time and consistent acceleration/coasting, bounds stall catch-up, and clears input on blur, visibility changes and mode transitions. Field arrows and both Shift keys have independent behavior. Controlled tests compare equal-duration movement/look at 30/60/120 Hz and preserve floor-relative eye height. See the timing report and PR checks for final browser, native Safari and deployment verification.
The stationary walkthrough batch (#80) stops frame requests after input/coasting settles and wakes for keyboard, mouse, eye-height and scene changes. Native Safari recorded zero callbacks and rendering frames in medium/large 25-second stationary samples, down from 1,500 each. See the report and numerical measurements for provenance, limits and final PR/CI verification.
The 2D drawing batch (#84) replaces idle dirty-flag polling with coalesced redraw requests. Local display, camera/minimap controls and image completions wake the canvas; late underlays cannot replace another floor’s image. See the report and PR #85 for native measurements, browser regressions and final deployment status.
Next, measure active orbit, stacking and editing on representative desktop/phone hardware and agree frame-time and memory targets. Use those results to choose shared geometry, object-level visual updates or mobile quality controls. Extend desktop Safari checks to actual iPhone/iPad touch devices. The initial small-home native Safari calibration and initial medium/large stationary walkthrough samples are complete; repeated active-navigation measurements and physical phones remain. Keep category contract fixtures in both repositories synchronized when extending the catalog.
The legacy preview batch (#86) refreshes identifiable old chair fallbacks on opening saved projects, JSON and history copies. Retained native categories determine presentation; edited geometry, explicit replacements, photos and unknown fields stay intact. Reading leaves raw library/history recovery bytes untouched, and normal saves retain the existing category marker. Unsupported or ambiguous retained data remains recoverable; RoomPlan chairs without a retained source are not guessed. See the report and PR checks for final browser and deployment verification.
The modal keyboard batch (#88) prevents editing shortcuts from changing a selected object behind an open dialog. Native dialogs provide focus and background inertness; keyboard guards also cover window/document listeners, elevation Escape and 3D input. Command actions execute after their palette closes. Area Summary safely includes imported room categories it does not recognize and releases its subscriptions when closed. See the report and PR checks for final browser, Safari and deployment verification.
#30 is closed. Its client migration and Storage rules gates were completed on October 16, 2026:
POST /api/handoffs. Physical-device QA, including Files/AirDrop package
exchange and real LiDAR/AR capture, is tracked in
openplan3d-ios#47.storage.rules
identical to the deployed rules so later deployments cannot reopen the bypass.One gate remains:
Preserve the low-cost design: ordinary editing, history, backups, photos and full project-package exchange stay local. Reuse bundled, cacheable catalog assets and unchanged temporary shares; keep downloads lazy. The endpoint currently bounds captures to 1 MiB, 1,000 reservations or 250 MiB per UTC day, and 10 reservations per minute. Failed writes retain reservations. These limits do not cap downloads or total spending. Keep the audited one-day inbox lifecycle and seven-day soft delete unless new measurements justify a reviewed change. Avoid adding a database, durable cloud copies, sync or media uploads without a cost model and enforceable quotas. See the cost audit and cost/browser report.
These are follow-up work areas, not claims that every item is a reproduced bug.
Area/geometry agreement: define whether area is measured at interior wall faces or another boundary, reconcile native raster-based areas with web polygons, and test room split/merge identity and schedules. Matching area totals are not yet an established cross-platform guarantee.
Straight and curved wall splitting now updates saved room references while retaining room identity, names and finishes. Unit cases include two rooms using different portions of a long wall; browser cases verify export, Undo/Redo and save/reopen. Dividing/merging rooms themselves remains separate work.
Splitting a grouped wall now keeps both child segments in the original group. Unit coverage checks group copying and exact Undo/Redo; three browser cases check membership, geometry and save/reopen (14.1 seconds). Group gesture behavior on physical devices remains to be qualified.
Curved walls now split through exact quadratic subdivision. Opening clearance uses each child’s curved path, and controls/heights/positions are remapped without changing the underlying curve. The existing fixed facet count per wall can yield a finer rendered approximation and slightly different estimated area after subdivision. Physical gestures and pixel-level visual qualification remain.
FEATURES.md now provides implementation scope linked to regression coverage;
COMPARISON_REVIEW.md is explicitly historical. README links the matrix and
package import/export contract and uses the catalog source instead of a stale
item count. CONTRIBUTING.md, fixture-oriented issue/PR templates and the release
evidence checklist are now available. Continue maintaining these with behavior
changes; historical findings are not current release status.main against origin/main; reread
open GitHub issues and #30 for release updates. Start a focused codex/… branch
from current main after checking the browser batch merge status.NODE_ENV=production npm run check,
NODE_ENV=production npm test and NODE_ENV=production npm run build.
Finish check before starting build; both regenerate SvelteKit artifacts.
Production browser workflows run in GitHub CI with cloud uploads/analytics
disabled. Use the approved browser-control tools for local interactive QA.openPlan3d.xcodeproj, scheme FloorPlan, Debug simulator
tests with CODE_SIGNING_ALLOWED=NO. Select an available simulator; rerun
actual native return-package fixtures when the contract changes. Complete
physical-device release checks separately.Native photo preparation now verifies completed coverage-job source bindings and
converts JPEG/HEIC/PNG into sRGB projection inputs, preserving dimensions and raw
camera precision with separate original/derived hashes. 145 native tests per
platform passed with one optional skip and zero failures. A native-generated
fixture also completed a pinned Blender bake; 99 renderer host tests passed with
16 delegated skips and zero failures. See the native repository’s
docs/native-projection-photo-validation.md. Native export UI, supervised photo
projection jobs and physical/held-out validation remain open.
Native coverage review can now export selected cameras as a bounded, streaming
projection ZIP. The Blender prototype accepts the extracted package, verifies all
file identities and records original/derived provenance. A two-camera package
saved through the isolated Mac QA app completed a real bake. 145 native tests per
platform and 103 renderer host tests passed with expected skips and zero failures.
See the native
repository’s docs/native-projection-export-validation.md. Supervised projection
jobs, native result review/masks, inter-island seams and physical validation remain.
The native photo prototype now samples diagonal triangle texels at their covered
area’s centroid, removing 144 unresolved edge samples in the exported QA scan.
Neutral pixels in a fixed top-wall band fell from 539 to 3 while strict visibility
and exclusion checks remain. 104 renderer host tests passed with 17 delegated
skips and zero failures. See the native repository’s
docs/triangle-boundary-sampling-validation.md. Broader seams, coarse atlas
coverage boundaries and physical fidelity remain unverified.
An independent photo-result verifier now binds native-package bakes to expected
input identities/settings and validates artifact hashes, PNGs and map/count
consistency. It emits a compact receipt for future supervision without opening the
Blender scene. See the native repository’s docs/projection-result-verification.md.
109 renderer host tests passed with 17 delegated skips and zero failures.
Native projection queue/result review and physical fidelity remain open.
Photo projection now runs through the shared serial queue with frozen packages,
resource limits, cancellation, orphan-worker exit and independently verified
publication/recovery. The native queue lists and cancels the distinct job kind.
146 native tests per platform and 114 renderer host tests passed with expected
skips and zero failures. See the native repository’s
docs/projection-queue-validation.md. Native submission/result review, supervised
masks, seams and physical validation remain open.
Native coverage review now submits photo projection jobs directly, with selected
cameras, texture detail and optional exposure matching/blending. A live Mac UI
submission completed in the worker without ZIP export or CLI submission.
146 native tests per platform and 115 renderer host tests passed with expected
skips and zero failures. See the native repository’s
docs/native-projection-submission-validation.md. Native result/map review, masks,
seams and physical validation remain open.
Native completed projections now have verified preview, atlas, coverage and colored
source/contributor map review with frame legends. The reader binds frozen inputs,
receipts and artifacts and recounts map pixels before displaying results.
151 native tests per platform passed with one optional skip and zero failures.
All five image choices and scrolling legends passed synthetic Mac UI review.
Renderer code is unchanged at 115 host tests with 17 delegated skips. See the native
repository’s docs/native-projection-review-validation.md. Native masks/export,
preview-space coverage/confidence overlays, seams and physical validation remain.
Projection preview review now includes a frozen source-photo comparison and opacity overlay. Native readers reverify the preview camera’s photo before use; 151 tests per platform pass with one optional skip and zero failures. Synthetic Mac UI checks cover source, projection and 0/50/100% opacity. This is a contributing camera comparison; held-out measurements and coverage/confidence overlays remain.
New projection jobs now render a verified camera-space coverage pass, displayed
in native review with an opacity slider. White denotes observed atlas samples;
black denotes unobserved samples or proxies; background is transparent. The saved
appearance scene is preserved. 152 native tests per platform passed with one
optional skip; 116 renderer host tests passed with 17 delegated skips, all with
zero failures. Live Mac overlay endpoints passed. See the native repository’s
docs/camera-coverage-overlay-validation.md. Confidence estimates, native masks,
seams and physical/held-out accuracy remain open.
Supervised projection jobs now accept camera-bound CLI masks, freeze their bytes,
reserve their disk space and verify them before execution and recovery. Native
review verifies the mask bindings and labels masked source cameras. 153 native
tests per platform passed with one optional skip and zero failures. Renderer
validation covered 118 host tests with 17 delegated skips; two environment timeouts
and a corrected recovery assertion passed isolated rechecks with unchanged limits.
Live Mac review confirmed the excluded camera contributed zero samples. See the
native repository’s docs/supervised-projection-masks-validation.md. Native mask
authoring/submission, confidence, seams and physical accuracy remain open.
Native projection submission now supports per-camera PNG mask selection,
validation, thumbnails, replacement/removal and frozen job submission. Updated
workers advertise mask support; older connected workers are rejected for masked
jobs. 154 native tests per platform passed with one optional skip and zero
failures; 24 worker lifecycle/capability tests passed. A native-created masked
job completed in Blender and reopened in review. See the native repository’s
docs/native-mask-submission-validation.md. Brush/polygon mask authoring,
confidence, seams and physical accuracy remain open.
Native brush mask drawing now supports Exclude/Restore, adjustable diameter,
Undo/Reset and editing selected masks over verified source photos. 156 native
tests per platform executed with one optional skip and zero failures; focused
checks passed again after canonicalizing generated PNG metadata. A native-drawn
mask passed the independent worker decoder, completed in Blender and reopened in
native review with 2,398 projected samples. See the native repository’s
docs/native-mask-drawing-validation.md. Polygon tools, imported PNG metadata
consistency, confidence, seams and physical validation remain open.
Native imported and directly submitted masks now normalize to worker-compatible
binary RGBA PNGs, removing unsupported metadata while preserving pixel positions.
Rotated and animated masks are rejected. 158 native tests per platform passed,
with one optional skip and zero failures, plus independent worker decoding of
the prepared fixture. See the native repository’s
docs/native-mask-compatibility-validation.md. Polygon authoring, confidence,
seams and physical/release validation remain open.
Native polygon mask authoring now supports click/tap vertices, fill previews,
Exclude/Restore, point undo/clear and mixed brush/polygon edits. 161 tests per
native platform passed, with one optional skip and zero failures. Independent
worker decoding confirmed source coordinates; a native job completed in Blender
with 2,336 projected samples and reopened in review. See the native repository’s
docs/native-polygon-mask-validation.md. Zoom/pan, automatic semantic masks,
confidence, seams and physical/release validation remain open.
Native mask editing now supports 1×–8× zoom, Move mode and Fit while preserving
source-pixel alignment for brushes, saved masks and unfinished polygon outlines.
164 tests per native platform passed, one optional skip and zero failures;
independent worker decoding and live Mac checks covered coordinate alignment and
saving/reopening. See the native repository’s docs/native-mask-zoom-validation.md.
Full-size image memory measurements, physical gestures/accessibility, automatic
semantic masks, confidence, seams and physical/release validation remain open.
Native mask editing now checks the entire work budget before decoding or
rasterizing. Oversized-polygon rejection dropped from 69.642s to 0.009s on Mac
and 56.061s to 0.003s on iOS in local debug runs. 165 tests per native platform
passed, one optional skip and zero failures; representative output PNGs remain
byte-identical and pass worker decoding. See the native repository’s
docs/native-mask-work-preflight-validation.md. Accepted-history performance,
maximum-image memory, physical input/accessibility, confidence, seams and release
validation remain open.
Verified projection diagnostics now measure color jumps at source-camera changes
within UV islands, separately from same-source image edges. Reports include
linear RGB metrics, frame identities and bounded atlas-coordinate examples;
missing boundaries produce unavailable metrics. 20 renderer tests ran with four
delegated skips and zero failures. The native fixture lacks eligible source-change
edges, so an overlap fixture is still required for seam comparisons. See the native
repository’s docs/projection-source-boundary-validation.md. Seam correction,
inter-island measurements, confidence and physical validation remain open.
A calibrated synthetic overlap fixture now creates camera-source boundaries using
the actual Blender bake. Four-texel blending reduced mean linear RGB difference
across 42 boundary pairs from 0.497558 to 0, preserving 812 observed texels and
source labels. PNG reports and captured-camera previews were checked. 13 tests
passed, including four in pinned Blender, with no skips/failures. See the native
repository’s docs/overlap-seam-fixture-validation.md. Textured overlap, partial
FOV/mask edges, inter-island seams, confidence and physical validation remain open.
The synthetic overlap study now includes known texture, exposure mismatch and
partial/shared masks. Normalization reduced blended reference intensity error
from 0.072231 to 0.002218 and boundary gradient residual from 0.018639 to 0.003201,
with 781 observed texels preserved. Raw boundary contrast rose with restored
texture, so it cannot serve as a standalone quality score. 13 tests passed,
seven inside pinned Blender, with no skips/failures. See the native repository’s
docs/textured-overlap-validation.md. Real photos, FOV/inter-island seams,
confidence and physical/release qualification remain open.
The overlap fixture now validates narrow camera fields of view against analytic
wall/frustum intersections. Normalization preserves 641 observed samples while
reducing reference error from 0.071994 to 0.001168; combined FOV/masks preserve
615 samples and reduce error from 0.071347 to 0.001244. 17 tests passed, ten
inside pinned Blender, with no skips/failures. Independent PNG checks reproduced
all four study reports. See the native repository’s
docs/field-of-view-overlap-validation.md. Real photos, inter-island seams,
non-planar/occlusion cases, confidence and physical/release validation remain open.
Topology-based island diagnostics now pair observed samples across actual shared
coplanar mesh edges in separate UV tiles. The study excludes creases, overlapping
or disconnected geometry and missing coverage, and exports reproducible topology.
Normalization reduces the fixture’s shared-edge reference residual from 0.007316
to 0.001659 with unchanged coverage. 16 tests passed, eleven inside pinned
Blender, no skips/failures. See the native repository’s
docs/shared-island-boundary-validation.md. Seam correction, cross-mesh matching,
native topology/review integration, confidence and physical validation remain open.
An isolated /render-lab route now supports local GLB loading, interactive
preview and progressive GPU path tracing with spatial noise reduction. A private
Blender photo-study scene was exercised locally; its model and media remain out
of this repository. See setup, verified results and limits.
The render-lab batch passed 653 unit tests and a production build. The subsequent
keyboard cleanup passes 654 unit tests with zero type errors or Svelte warnings.
Desktop camera presets, mode switching, pause and PNG export were exercised.
This remains a test branch: finish material baking, browser/device qualification
and shared scene integration before replacing
any viewer. Physical-device capture and measured reprojection remain pending.
The native editor now exports its current floor as portable neutral scene JSON
for the existing Blender CLI worker. Wall openings and transforms reuse the
native preview; furniture is represented by boxes. Five targeted tests passed on
each of Mac Catalyst and the iOS simulator, with no failures or skips. A native
fixture rendered successfully in pinned Blender and its input/output hashes and
PNG were checked. See openplan3d-ios/docs/edited-plan-render-export-validation.md.
Source-to-edit camera alignment remains open. This does not qualify physical export interaction or real-photo accuracy.
The follow-up slab correction now bounds native preview/export floors around full transformed geometry, including rotated furniture and wall thickness. Labels no longer enlarge furniture-only slabs. Both regressions failed before the fix; seven edited-scene tests pass on each native platform after it, with zero failures or skips. A verified Blender render confirms full furniture support. These remain rectangular envelopes; room-outline and structural slab fidelity are still open.
The Mac editor now also provides Render This Floor… for direct native queue
submission of current edits, including unsaved changes. The normal worker,
budgets, receipt verification, native result preview and PNG save flow apply.
An isolated UI check rendered an unsaved table and saved the verified PNG while
saved plan/original room hashes stayed unchanged. Native-to-Python-to-native
verification also rejected deliberate frozen-input tampering. See
openplan3d-ios/docs/native-edited-floor-queue-validation.md for exact validation.
Queue/geometry runs covered 33 tests per native platform with zero failures and one opt-in skip each. Final focused runs passed all five edited-floor queue tests on both platforms without skips, including disk/job budgets and verified external worker output with tamper rejection.
The 3D viewer now offers Export Blender Scene, converting displayed active or stacked floor meshes into the shared metre/Y-up scene contract. It reuses viewer geometry, transforms and loaded furniture; neutral export omits window panes, ceilings, labels and camera helpers. Mesh/vertex/triangle/file budgets reject unsupported inputs with a visible message. See validation and usage.
660 unit tests, five browser checks, Svelte checks and the production build passed. Desktop/phone exports in Chromium and WebKit were byte-identical. Both actual active/stacked downloads rendered in pinned Blender, with verified scene and PNG hashes and visually checked results. Editable backup dimensions stayed unchanged. Browser testing caught and fixed an overlap with Undo History.
Next: direct browser-to-native queue handoff, material baking and further inactive-floor fidelity, camera alignment and physical/release qualification. This is a neutral snapshot adapter, not a completed replacement for either viewer.
Curved walls now cut door/window apertures in active and stacked floors. Active trim and glazing follow the curve; baseboards clear doorways. Saved dimensions remain unchanged. Exported-triangle testing also caught and fixed a rounding sliver that blocked a window at a facet join.
665 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. Actual active/stacked downloads rendered in pinned Blender with independently verified scene/PNG hashes and visually checked results. See validation. Facet joins, inactive slab and trim fidelity, native curves, material baking and queue handoff remain open.
Active and inactive floors now use shared closed 5 cm slabs under resolved room polygons. Stacked slabs preserve concave recesses and gaps between disconnected rooms instead of filling the wall bounding rectangle. Unenclosed walls no longer invent an upper-floor slab. Source dimensions and room finishes remain intact.
667 unit tests, eight Chromium/WebKit browser checks, Svelte checks and the production build passed. Browser rays verify support and clear gaps through active-floor switches. The actual stacked export rendered in Blender with independently verified scene/PNG hashes and a visually checked result. See validation. Curved room boundaries, stair/courtyard voids, editable slab thickness, wall-face offsets and native slab fidelity remain open; the export’s separate rectangular ground support is unchanged.
Room detection and polygons now follow the viewer’s 16-facet quadratic wall path, so fills, floor finishes and slabs reach the curved boundary. Derived area uses that same centreline polygon. Source wall IDs retain room names and finishes; T-junction room splits and a room enclosed by a curve plus a straight wall are covered. Source dimensions remain unchanged.
671 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. The actual stacked download rendered in Blender with independently verified scene/PNG hashes and visual inspection. See validation. Exact analytic/interior-face areas, crossing-wall topology, courtyard/stair voids, slab authoring and native curved-room fidelity remain open.
Room detection and polygon reconstruction now split nonparallel wall crossings in the derived graph, including intersections between curved wall facets and straight dividers. Source walls remain intact. Overhanging crossing dividers resolve four separate rooms and matching slabs, with names/finishes retained by source boundary IDs.
674 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. Actual exported slab geometry passed both-floor switching checks; the stacked download rendered in Blender with verified scene/PNG hashes and visual inspection. See validation. Collinear overlaps, duplicate walls, ambiguous boundary-ID sets, courtyard/stair voids and native topology/area agreement remain open.
Coincident split segments now form one derived graph edge retaining every source wall ID. Partial collinear overlaps and duplicate dividers preserve room polygons and source boundary identity across input order changes. Saved room metadata matches expanded aliases, so adding a duplicate/partial overlap retains names and finishes; ambiguous saved matches are left unmatched. Editable walls and viewer wall meshes remain intact.
677 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. Browser exports preserve four room slabs on both floors across active-floor switches. See validation. Repairing overlapping wall solids, ambiguous face identity, courtyard/stair voids, native topology/area agreement and device performance qualification remain open.
Active and stacked 3D builders now resolve room metadata and polygons together, reusing one ephemeral graph per floor. The large furnished fixture previously rebuilt the same floor graphs 51 times per stack; it now builds them three times. No persistent cache is introduced, and in-place edits remain fresh.
Local Node measurements reduced median large-fixture room computation from 2.402 ms to 0.393 ms, about 84%, with matching output. 679 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. See method, limits and raw samples. This does not establish browser FPS or device budgets; representative active navigation, full scene construction, memory and phone measurements remain open.
The editor now shares detected room polygons across fills, labels, hit tests, rename placement and furniture-room dimensions. Floor/geometry changes refresh the polygons and room IDs together; unchanged redraws do not rebuild each room’s graph. 680 unit tests, ten Chromium/WebKit browser checks, Svelte checks and production build passed. See validation.
A browser regression exposed a separate usability issue: selecting a room can open the properties panel and move the canvas between the two clicks of a direct double-click, causing the adjacent room to receive it in WebKit. Selection then rename at the updated position works and is covered. The native mouse layout interaction is addressed in the subsequent double-click batch; active-editing/device measurements and broader performance qualification remain open.
Native Select-mode double-clicks now keep the first press’s floor coordinates and skip second-press reselection after a sidebar moves the canvas. Room rename, text editing and wall splitting use that original point. Other drawing modes, ruler behavior and synthetic touch paths retain their existing handling.
The direct Room 1,1 double-click that previously opened Room 2,1 in WebKit now passes without the preselection workaround. 680 unit tests, eight Chromium/WebKit browser checks, Svelte checks and production build passed. See validation. Physical double-tap/device qualification and broader active-editing performance work remain open.
The 2D renderer, hit testing and inline rename now share the saved label anchor. Dragging visibly moves the text; room geometry and dimension annotations stay fixed. An explicit Reset Label Position context-menu command replaces an invisible reset hit region. Clicks no longer commit offsets from stale pointer coordinates; actual drags use screen deltas and an independent undo group, preserving separate reset/drag undo steps even when performed quickly.
680 unit tests, ten Chromium/WebKit browser checks, Svelte checks and production build passed. Tests verify saved offsets, reset, dragging, rename placement, undo, direct rename across floors, idle/wakeup and simulated touch. See validation. Physical touch/long-press, other export/viewer label conventions and active-editing budgets remain open.
PNG, SVG, PDF and DXF now preserve saved room label positions. Framed exports include name/area text bounds even outside the walls. PNG and the PDF plan raster retain up to 2x scale with a 4096-pixel longest-side cap, preventing distant labels from causing unbounded canvas allocations.
682 unit tests, two Chromium/WebKit browser checks, Svelte checks and production build passed. Actual downloads verify SVG text bounds, PNG dimensions, DXF coordinates and PDF serialization; the PNG was visually inspected. See validation and limits. PDF page-layout qualification, curved geometry/export fidelity, native label conventions and physical-device downloads remain open.
PNG/PDF/SVG now draw quadratic wall curves; DXF uses the viewer’s 16 facets. Export bounds include exact quadratic extrema and wall thickness, preventing curves or thick strokes from clipping. Dimensions use path length and sit outside curved strokes. Source geometry remains unchanged.
685 unit tests, four Chromium/WebKit browser checks, Svelte checks and production build passed. Pixel tests verify the curved stroke and empty former chord; SVG/DXF checks cover geometry and PDF downloads serialize. The final PNG was visually inspected, including the dimension placement. See validation. Curved opening symbols/gaps, DXF facet joins, PDF layout and native/device export fidelity remain open.
PNG/PDF/SVG now clear the curved wall interval beneath doors and windows; symbols span the actual quadratic jambs and clip at curved wall ends. DXF cuts opening intervals from straight and faceted wall outlines. Saved dimensions are unchanged.
688 unit tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. The exported PNG was visually inspected. See validation. DXF joins, PDF page layout, all-symbol overlap cases and native/physical-device export qualification remain open.
DXF now emits a continuous wall outline between openings instead of separate rectangles for every curve facet. Offset-line intersections preserve thickness; bounded miter/bevel joins avoid sharp-bend spikes. Door/window intervals still split outlines, and saved geometry is unchanged.
691 unit tests, four Chromium/WebKit export checks, Svelte checks and production build passed. See validation. Joining separate source walls, self-overlap unions, third-party CAD qualification, PDF layout and native/device export parity remain open.
Room schedules now wrap text to column widths and add pages before reaching the footer. New pages repeat headings; totals and summary stay together. This fixes large schedules running off-page and names silently truncated at 30 characters.
Two Chromium/WebKit export checks, eight export unit tests, Svelte checks and production build passed. The full unit run had one unrelated timeout, which passed separately. All four schedule pages in the 32-room PDF were rendered and visually reviewed; text/geometry inspection verified complete names and footer clearance. See validation. Long title-block metadata, Unicode fonts, extreme-plan readability, optional 3D layout and physical printing remain open.
PDF export now selects the explicitly marked main ThreeViewer canvas, skips lost WebGL contexts, and no longer probes unrelated canvases or substitutes the last 2D canvas. This prevents misleading optional 3D pages.
Four Chromium/WebKit checks, Svelte checks and production build passed. 694 unit tests were covered across the full run and the corrected document-stub rerun. The exported 3D page was rendered and visually checked. See validation. Blank-frame detection, serialization failure handling, extreme aspect ratios and broader PDF/native fidelity remain open.
If optional 3D capture or image encoding fails, PDF export now removes unfinished pages before saving the completed plan and schedule. Previously, failures after page creation could leave a blank perspective page.
13 export/real-jsPDF tests, two Chromium/WebKit download checks, Svelte checks and production build passed. Tests cover tainted capture and invalid PNG data, including failure after the optional page was added. See validation. Required plan-image failures, blank-frame detection, error messaging and native parity remain open.
Toolbar and command-palette PDF exports now share dismissible feedback for required preparation failures, empty floors and omitted optional 3D captures. Complete exports clear stale notices. Failed preparation no longer escapes the entry point as an unhandled error; the message offers retry/JSON-copy guidance.
15 targeted tests, two Chromium/WebKit browser checks, Svelte checks and production build passed. Both export entry points were tested with a forced plan-image failure and no download. See validation. Blank-frame detection, remaining PDF typography/layout, native feedback and non-PDF error handling remain open.
Toolbar 3D PNG export now waits for the main viewer’s first rendered frame, encodes only that canvas, and reports failure. Readiness and encoding use separate bounded waits; duplicate requests are disabled. Automatic 2D switches restore 2D after success or failure, and stale project/floor downloads are prevented.
Eight capture tests, two Chromium/WebKit checks, Svelte checks and production build passed. The actual PNG was visually reviewed. See validation. Full asynchronous asset readiness, blank-frame checks and broader device/camera qualification remain open.
Project PNG export now uses the full-floor renderer without querying an on-screen canvas. Empty floors never substitute another viewport. Null blobs, thrown errors and encoding timeouts are reported through the shared notice from both toolbar and command palette; complete exports clear prior notices.
21 targeted tests, six Chromium/WebKit browser checks, Svelte checks and production build passed. Browser coverage includes 2D export while 3D is open, failed encoding through both entry points, and curved wall/opening regressions. See validation. All-object bounds/typography, asset readiness, physical downloads and native parity remain open.
PNG, SVG and PDF now frame furniture using rotated symbol dimensions and stroke width, replacing fixed extents or missing furniture bounds. Oversized items outside the walls remain visible; saved furniture and raster caps are unchanged.
14 final geometry/export tests, three real-jsPDF tests, six Chromium/WebKit checks, Svelte checks and production build passed. The actual PNG was visually reviewed. See validation. Long symbol labels, other object categories, title text and native parity remain open.
PNG/PDF now render saved text annotations; SVG preserves multiline spacing; DXF includes rotated lines on true-color layers for hex colors. Framed formats include rotated measured text bounds, preserving notes outside the wall envelope.
18 targeted tests, six Chromium/WebKit checks, Svelte checks and production build passed. The exported PNG was visually reviewed. See validation. Universal font coverage, title-block text, dimension annotations, other objects and native parity remain open.
PNG/PDF now draw dimension callouts; PNG/SVG/PDF bounds include offset geometry and labels. SVG default labels use selected units; DXF includes callout primitives. Zero offsets now survive rendering and hit testing. Canvas zoom scaling and long-label gaps are corrected without changing saved annotations.
21 targeted tests, six text/curve browser regressions, two final dimension browser checks, Svelte checks and build passed. The exported PNG was visually reviewed. See validation. Standalone measurements, all-object bounds, universal fonts, title layout, physical checks and native parity remain open.
PNG/PDF and DXF now include saved measurement lines, endpoints and labels. SVG includes endpoint dots and uses selected units. Framed bounds include measurement geometry and labels outside the wall envelope.
19 targeted tests, four Chromium/WebKit checks, Svelte checks and production build passed. Metric and imperial labels are covered; the PNG was visually reviewed. See validation. All-object framing, fonts, physical/device qualification and native parity remain open.
Shared length formatters now round total inches before splitting feet/remainders,
preventing labels such as 1'12". Precise tenths carry correctly as well; negative
values use a single leading sign without negative zero. Metric formatting and
saved geometry are unchanged.
738 unit tests across 59 files, Svelte checks and production build passed. Export checks include carried imperial measurement labels in PNG/PDF/SVG/DXF. See validation. Input parsing, physical scale/device checks and native unit parity remain open.
Wall length now accepts explicit metric and feet/inches units, with bare values
using the displayed cm/in units. The parser correctly reads 12" as inches and
rejects trailing junk/nonfinite values. Invalid drafts preserve saved geometry
and undo history; unchanged rounded displays retain full stored precision.
757 unit tests across 60 files, four Chromium/WebKit desktop/phone-width checks, Svelte checks and production build passed. See validation. Fractional notation, other numeric property fields, native unit parity and physical device checks remain open.
Clearing a stair dimension previously saved zero. Stair and column fields now preserve saved geometry for invalid drafts, honor existing riser/column ranges, accept fractional dimensions and retain precision on unchanged blur. Column limits convert to the selected units.
Four final structural and four wall/opening regression checks passed in Chromium/WebKit at desktop/phone widths; Svelte checks and build passed. See validation. Other property editors, native parity and physical device qualification remain open.
Entourage width/rotation, furniture/column rotation, text font size/position/ rotation and background-image rotation now preserve saved values for invalid numeric drafts. Fractional values remain visible and editable; entourage minimum width converts to inches correctly.
Four Chromium/WebKit desktop/phone-width checks, Svelte checks and build passed. See validation. The phone check also exposed a remaining usability issue: fit-to-view can place content behind the open properties sheet; panning reveals it. Fit framing, physical touch qualification and native parity remain open.
Fit-to-view now measures the visible canvas above an overlapping properties sheet and centers the plan there. The prior phone panning workaround is removed from the regression: the painted note is above the sheet and directly selectable. Desktop sidebars continue using their existing layout space.
Eight Chromium/WebKit desktop/phone-width checks, Svelte checks and build passed; the phone screenshot was visually reviewed. See validation. All-object fit bounds, automatic reframing on selection and physical touch qualification remain open.
Fit now uses shared geometry bounds for walls, furniture (including unknown catalog fallbacks), rotated stairs/columns, entourage, measurement/dimension geometry, text and loaded tracing images. Floors without walls can be fitted; current-floor content takes precedence over the floor-below fallback.
763 unit tests, four final note-only browser checks and four property-editor regressions passed, along with Svelte checks and build. See validation. Caption/room-label bounds, minimap parity, initial automatic fitting and extreme zoom limits remain open.
Initial framing now uses shared content bounds instead of requiring walls. It waits for tracing-image dimensions on image-only floors and coalesces pending attempts. After the initial fit, subsequent edits retain the camera.
12 Chromium/WebKit desktop/phone-width checks, Svelte checks and build passed. See validation. Per-floor/selection reframing, caption/room-label bounds, minimap parity and extreme zoom limits remain open.
A browser regression found that Delete Room removed the walls/openings while leaving a saved room record referencing deleted walls. The store now removes the room boundary, attached openings and saved room metadata in one undo operation. Two unit cases cover saved/detected rooms, unrelated data preservation and exact Undo/Redo. All 945 unit tests and six desktop/narrow browser cases pass. Keyboard label reset also preserves other room fields and restores its offset with Undo. Full browser, shared-boundary deletion semantics and physical/release gates remain.
Room deletion now retains boundary walls used by another saved or detected room, including attached openings. Two failing unit cases reproduced shared-wall loss; all four room-deletion unit cases now pass, with exact Undo/Redo and unrelated data preservation. Check/build and six browser cases at 1440px/390px across three engines pass (17.0 seconds). In a four-room grid, deleting a corner removes only its two exclusive walls, preserving three neighbors, shared openings and all furniture. Imported duplicate/coincident wall topology and physical/release gates remain open.
Unchanged saved-room updates, empty patches and nonexistent room IDs now return before a snapshot or save notification. Label offsets compare their coordinates, not object identity. First-time metadata persistence for detected rooms remains. The regression verifies that unchanged fields add no Undo step and preserve Redo. All 949 unit tests, check/build and six desktop/narrow browser cases pass. Other item types and remaining NEXT scope are not qualified by this room-specific fix.
Furniture updates now skip missing targets, empty patches, unchanged scalar values and equal position/scale coordinates before snapshotting. Fifty furniture unit cases pass, including the reproduced Undo/Redo regression and existing geometry interactions. Check/build and three Portuguese browser cases pass (13.1 seconds), verifying redo after reapplying a color, dimensions, rotation/mirroring, appearance reset and retained item details. Other item types and physical qualification remain.
Equivalent start/end/curve coordinates now skip wall snapshots. Explicit uniform height updates still flatten a slope when the requested height equals the stored maximum; that is a geometry change, not an unchanged edit. Two regressions reproduced endpoint-history loss and skipped flattening. All 43 focused wall unit cases, check/build and three Portuguese wall-control browser cases pass (16.3 seconds). Physical interaction and the broader NEXT scope remain open.
The desktop minimap now uses shared content bounds for drawing and navigation, includes omitted objects/annotations as navigation markers and lines, and handles unknown furniture footprints. Object-only floors no longer show the empty-plan hint. Phone minimap visibility is unchanged.
764 unit tests, eight content browser checks and six canvas-idle regressions passed, along with Svelte checks and build. See validation. Caption/room-label bounds, extreme zoom limits, floor-switch framing and physical device checks remain open.
First visits now frame each floor; returning restores its camera center and zoom for the current editor session. Views are keyed by project/floor and remain transient UI state. Same-floor edits retain the camera and stale tracing-image loads remain rejected.
Fourteen Chromium/WebKit desktop/phone-width checks, Svelte checks and build passed. See validation. Caption/room-label bounds, extreme zoom limits, selection framing and physical device checks remain open.
Fit can now go below 10%, with a per-floor lower limit at one-quarter of its fitted scale. Canvas and mobile overflow controls share that limit. Rulers extend their spacing at small scales, percentages remain nonzero, and a screen margin keeps short labels clear of rulers. Phone zoom controls now sit beside Tools.
764 unit tests, four final large-plan browser checks and eight floor/property regressions passed; Svelte checks and final build passed. The phone screenshot was reviewed. See validation. Caption/ room-label bounds, selection framing and physical device checks remain open.
An image-only floor now waits for its image before completing initial framing, instead of prematurely fitting the visible floor below. Failed loading releases the wait and permits that fallback; stale image callbacks remain rejected.
Eighteen Chromium/WebKit desktop/phone-width checks, Svelte checks and build passed. See validation. Caption/ room-label bounds, selection framing and physical device checks remain open.
Fit includes the combined room-name/area text at its saved offset, using actual screen font measurements and a bounded scale search. Hidden labels are excluded; geometry and offsets remain unchanged. Shared minimap world bounds include labels.
The 765-test suite, eight final bounds tests, four final label browser checks and eight large-plan/property regressions passed. Final Svelte checks/build passed, and the phone screenshot was reviewed. See validation. Overwide single-line labels, other annotation captions, selection framing and physical checks remain open.
Fit now measures text annotations at their rendered screen font size, including the eight-pixel minimum, multiline spacing and rotation, before converting bounds to world coordinates. The default export bounds behavior is preserved.
766 unit tests, ten Chromium/WebKit checks, Svelte checks and build passed. Longer note ink and export regressions are covered; the phone screenshot was reviewed. See validation. Overwide text, measurement/dimension captions, selection framing and physical checks remain open.
Fit measures saved measurement and dimension captions at their rendered screen font sizes and includes endpoint dots, arrows and leader extensions. Hidden measurement/annotation layers are excluded from these bounds and minimap lines. Saved geometry and labels remain unchanged.
767 unit tests, twelve Chromium/WebKit checks, Svelte checks and build passed. Desktop/phone caption framing, hidden measurements, object-only minimap behavior and export regressions are covered; the phone screenshot was reviewed. See validation. Overwide text, automatic wall/internal room dimension labels, selection framing and physical checks remain open.
Fit includes automatic wall caption ink, offset lines/ticks and internal room width/depth captions at their rendered screen sizes. Curves and clear-span wall insets use the renderer’s geometry helpers. Both possible wall dimension sides are bounded; hidden dimensions are excluded. Internal room text now explicitly sets its alignment when room-name labels are hidden.
The 767-test suite, 11 final bounds tests, eight framing checks and four final wall-editing checks passed. Svelte checks/build passed; phone QA was reviewed. The wall regression’s lazy 3D readiness assertion now allows 30 seconds after initial Chromium timeouts. See validation. Overwide text, selection framing, selected-opening distance annotations and physical device qualification remain open, alongside the broader backlog.
PNG, PDF, SVG and DXF now accept furniture, notes, measurements and dimension annotations without requiring walls. Their existing rendering/bounds logic frames that content, including distant and rotated objects. Empty/blank/degenerate-only floors remain excluded, with updated PNG/PDF messages. DWG’s DXF fallback shares the eligibility rule.
769 existing unit tests plus the new eligibility test, ten Chromium/WebKit checks, Svelte checks and build passed. All four object-only PDF pages were rendered and visually reviewed. See validation. Tracing images, stairs, columns and entourage still need consistent export coverage; broader layout, physical-scale and native parity work remains open.
Round and rotated square columns now appear in PNG/PDF/SVG/DXF exports, with diagonal markers and bounds that include stroke width. Column-only floors qualify for export. Raster outputs reuse the editor renderer; SVG retains fill color; DXF uses circle/closed-outline primitives on a COLUMNS layer.
The 770-test suite, two new geometry tests, twelve Chromium/WebKit checks, Svelte checks and build passed. Both column PDF pages were rendered and visually reviewed. See validation. Tracing images, stairs, entourage, CAD fill styling, physical-scale and native parity work remain open.
Fit and single-stair selection outlines now include L-shaped outer runs and U-shaped landings. Hit testing uses the filled footprint, excludes empty corners and the U void, and uses a circle for spiral stairs. Fit includes stair caption ink; straight UP/DN arrows now remain inside their footprint.
772 existing unit tests, four new geometry tests, eight final footprint browser checks, four structural regressions and two phone QA checks passed. Svelte checks/build passed, and unobstructed phone screenshots were reviewed. See validation. Stair exports, multi-selection framing, below-floor ghosts, L/U direction indicators, physical checks and native parity remain open.
L-shaped and U-shaped arrows now reverse with Direction. Spiral arrows reverse their arc and endpoint, and their heads now point along the travel direction. Existing saved geometry and undo/redo semantics are preserved.
780 unit tests, eight Chromium/WebKit behavior checks, four final phone QA checks, Svelte checks and build passed. L/U/spiral phone screenshots were reviewed. See validation. Stair exports, below-floor ghosts, multi-selection framing, physical checks and native parity remain open.
The reference floor now uses the shared stair renderer at reduced opacity, preserving shapes, treads, labels and direction arrows. Empty-floor fallback Fit includes only the walls/stairs actually displayed; invisible lower-floor objects no longer distort its bounds. Reference stairs remain non-interactive.
780 unit tests, twelve delayed-image regressions, four final Chromium/WebKit reference checks, Svelte checks and build passed. Desktop/phone screenshots were reviewed. See validation. Stair exports, multi-selection framing, physical checks and broader native/geometry parity remain open.
Group boxes now enclose curved walls, rotated furniture/columns and actual stair footprints. Their drag region uses the same bounds. Removing a duplicate history snapshot at pointer-up makes a single Undo restore a group drag.
782 unit tests, eight stair regressions, four final Chromium/WebKit group-drag checks, final Svelte checks and build passed. The fitted phone screenshot was reviewed. See validation. Dedicated Fit Selection, opening symbol extents, entourage/annotation group operations, stair exports and physical/native qualification remain open.
Wall endpoint/parallel/curve, room, stair, column, text and group drags now own one undo group, starting after actual pointer movement and ending on release. This removes duplicate final snapshots, preserves the pre-drag state for rooms, and groups curve mutations. Selection clicks do not start a geometry undo group.
782 unit tests, checks for seven individual drag types in both browser engines, desktop/phone group regressions, Svelte checks and build passed. The final stair rerun waits for painted Fit to avoid a framing race. See validation. Opening/guide/entourage history is addressed below; physical gestures and broader export/native work remain open.
Doors, windows, guides, entourage movement and entourage resizing now use one undo group per drag, starting after three screen pixels of movement. Selection clicks no longer take entourage snapshots, and opening/guide movement no longer adds a snapshot for every update.
782 unit tests, all five drag paths in Chromium and WebKit, Svelte checks and production build passed. The final door rerun compares against the imported plan and waits for canvas layout to settle. See validation. Physical gestures, opening symbol bounds, entourage group operations and export/native parity remain open.
Group bounds now include rotated built-in and custom entourage, marquee selection includes entourage, and group dragging moves unlocked entourage with the other objects. Locked furniture and entourage stay stationary. Selection and Fit share the entourage bounds calculation.
783 unit tests, four final Chromium/WebKit desktop/phone cases, Svelte checks and build passed. The browser checks cover marquee selection, equal movement, locks and one-step Undo; phone screenshot review passed. See validation. Other entourage group operations, annotation selection, opening bounds, physical gestures and export/native parity remain open.
Canvas Duplicate now copies all selected supported objects, including stairs, columns and entourage, in one history action. Copied curved walls carry translated control points and correctly remapped openings; complete saved groups are copied. The contextual toolbar now supports stairs, columns and entourage. Deleting objects cleans saved group references, including openings removed with a wall.
The 785-test suite and three final focused unit cases passed; all four final Chromium/WebKit mixed/entourage Duplicate–Undo–Redo–Delete–Undo checks, Svelte checks and build passed. See validation. Clipboard copy/paste remains on older ID-based paths. Alignment/distribution, annotation selection, opening bounds, physical gestures and export/native parity remain open.
Copy/Paste now uses captured geometry instead of looking up live source IDs. Walls, openings, furniture, stairs, columns, entourage and saved groups remain pasteable after source edits/deletion. Successive pastes use fresh IDs and offsets, with one Undo per paste. Openings require a valid destination wall. The clipboard persists between floors and clears on project changes to keep custom assets valid.
788 unit tests, four Duplicate regressions, four final Chromium/WebKit clipboard workflows, Svelte checks and build passed. The final browser rerun verifies the plain-store snapshot fix. See validation. Cross-project custom assets, annotation clipboard support, alignment/distribution, opening bounds, physical gestures and export/native parity remain open.
Alignment now supports furniture, stairs, columns and entourage using rotated plan bounds, including asymmetric stairs and custom symbols. Shared furniture bounds account for scale. Locked objects remain stationary; distribution spaces visual centers between fixed endpoints and locked anchors. No-op operations add no Undo entry, and unavailable toolbar actions are disabled.
798 unit tests, ten final alignment unit cases, eight Chromium/WebKit desktop/ phone toolbar workflows, Svelte checks and build passed. See validation. Wall/opening and annotation alignment, cross-project clipboard assets, opening bounds, physical gestures and export/native parity remain open.
Ctrl/Cmd+L now locks selected furniture and entourage together when any is unlocked, or unlocks them together when all are locked, with one Undo step. Unselected/unsupported objects are unaffected. The R shortcut ignores locked furniture without adding a history entry.
799 unit tests, the final focused fixture check, four Chromium/WebKit desktop/ phone shortcut workflows, Svelte checks and build passed. See validation. Broader rotation/group transforms, annotation selection, opening bounds, physical gestures and export/native parity remain open.
R now rotates selected furniture, stairs, columns and entourage by 15 degrees around their movable bounds center, updating positions and orientations together. Locked objects stay fixed and do not affect the pivot. A single movable object rotates in place; the operation has one Undo step and skips no-op history.
805 unit tests, eight Chromium/WebKit rotation/lock workflows, Svelte checks and build passed. See validation. Wall/opening and annotation transforms, cross-project assets, opening bounds, physical gestures and export/native parity remain open.
Fit and multi-selection bounds now include opening symbols at their wall tangent, covering door swings, pocket tracks, folding/garage details, bay/casement windows and sliding arrows. Screen-sized details participate in zoom refinement. Bounds are conservative, allowing some spare space on unused swing sides.
818 unit tests, eight Chromium/WebKit desktop/phone framing checks, Svelte checks and build passed. Phone screenshot review passed. See validation. Fit Selection remains next; selected-opening dimension labels, annotation group operations, physical gestures and export/native parity remain open.
The zoom toolbar now offers Fit Selection (Shift+F), using selected geometry and zoom-aware bounds while F fits the full plan. Opening hosts do not enlarge an opening-only selection; selected rooms include their walls even with hidden labels. Zoom controls sit above an overlapping phone properties sheet. Choosing a Layers item clears the old multi-selection so unrelated geometry is not included.
821 unit tests, three final focused cases, twelve final Chromium/WebKit framing checks, Svelte checks and build passed. Phone screenshot review passed. See validation. Selected-opening dimension labels, annotation group editing, physical gestures and export/native parity remain open.
Guide, measurement, dimension and text clicks clear stale group/room and auxiliary selection. Layers selection activates the corresponding canvas target, enabling correct deletion. Escape clears auxiliary and room selection so a later Delete cannot act on an invisible old target. Deletion also clears primary selection.
Sixteen Chromium/WebKit selection and drag/Undo checks, Svelte checks and build passed. See validation. Annotation group transforms and clipboard support, selected-opening labels, physical gestures and export/native parity remain open.
Copy/Paste now supports notes, measurements and dimension annotations, including canvas-local selection. Copies retain styling, rotation, labels and offsets while translating coordinates and preserving measured lengths. Annotation removal uses shared group-reference cleanup, and each paste remains one Undo step.
822 unit tests, six Chromium/WebKit annotation clipboard workflows, Svelte checks and build passed. See validation. Annotation multi-selection/group bounds and transforms, cross-project assets, selected-opening labels, physical gestures and export/native parity remain open.
Layers now lists text notes with normalized multiline labels and an empty-note fallback, making distant notes selectable for framing and editing. Shared selection clears old group/room targets. Notes retain existing always-visible rendering; the category can be collapsed.
Eight Chromium/WebKit selection workflows, Svelte checks and build passed. The distant-note case verifies Fit Selection, Escape, isolated Delete and exact Undo restoration. See validation. Annotation group editing, note visibility, physical qualification and export/native parity remain open.
Notes, measurements and dimensions now participate in marquee and Select All, with zoom-aware caption bounds. Dragging a selected annotation moves the whole group by one snapped delta, preserving lengths and styling in one Undo step. Group Delete handles an annotation primary target, and Deselect All clears stale auxiliary selection. Keyboard and context-menu Select All share one path.
18 final Chromium/WebKit annotation workflows, four existing mixed-object checks, Svelte checks and build passed. The unit suite passed 822 of 823 cases; the one image-storage timeout passed with its 19-case file on an isolated retry. Phone screenshot review passed. See validation. Saved-group re-selection, Shift-click annotations, rotation/alignment, note visibility, physical qualification and export/native parity remain open.
Canvas clicks on notes, measurements and dimensions now reopen saved groups and can immediately drag the group. Ctrl/Cmd-click isolates an annotation; Shift-click adds or removes one member and keeps primary selection consistent. Deleting the last remaining selected member clears selection fully.
All 16 Chromium/WebKit workflows, Svelte checks and build passed. Desktop and phone cases reopen groups through every annotation type, check group drag and Undo, toggle members, and verify isolated deletion against exported data. See validation. Annotation rotation/ alignment, note visibility, wider object modifier consistency, physical checks and export/native parity remain open.
R now rotates notes, measurements and dimensions, including canvas-local selection. Single notes rotate in place; single dimensions rotate around their endpoint midpoint. Mixed selections share a zoom-independent geometric pivot, preserving annotation metadata and lengths while respecting object locks. The entire rotation remains one Undo step.
All 827 unit tests, 12 Chromium/WebKit rotation workflows, Svelte checks and build passed. Phone screenshot review passed. The multi-megabyte quota test now has a targeted 15-second timeout after repeated five-second timeouts; its assertions and application storage limits are unchanged. See validation. Annotation alignment, visibility, wall/opening rotation, physical qualification and export/native parity remain open.
Shift-click now toggles furniture, columns, stairs, entourage, doors, windows and walls before the Shift-pan handler intercepts the press. Objects share the annotation toggle path, keeping the remaining primary selection consistent. Shift-drag on empty canvas retains panning.
18 Chromium/WebKit workflows, Svelte checks and build passed. The furniture fixture was corrected to use a rendered catalog chair and passed a focused rerun; the other sixteen checks passed in the main run. Exact floor exports verify no geometry changes during selection/panning, isolated deletion, and Undo restoration. See validation. Ctrl/Cmd group isolation, unknown-catalog rendering/hit-testing, annotation alignment/visibility, physical qualification and export/native parity remain open.
Ctrl/Cmd-click now isolates every supported object type from saved groups. Group drag bounds and selection handles yield to the modifier, and all object selection paths pass it consistently. Selection alone preserves saved groups.
All 18 Chromium/WebKit workflows, Svelte checks and build passed. Each object type is isolated with its group selected and deselected; exact exports verify unchanged selection state data, isolated deletion and full Undo restoration. See validation. Saved-object group drag initiation, unknown-catalog rendering/hit-testing, annotation alignment and visibility, physical qualification and export/native parity remain open.
The first press on a deselected saved-group member now starts the shared group drag instead of moving just the clicked object. Furniture, columns, stairs, entourage, straight walls, and openings grouped with their host walls are covered. Locked members remain fixed and the complete drag stays one Undo step.
All 18 Chromium/WebKit workflows, Svelte checks and build passed. Exact exports verify common movement of objects, grouped notes/dimensions, unchanged locked members and opening positions, full Undo/Redo, and retained Ctrl/Cmd isolation. See validation. Opening-only group movement, curved-wall group translation, unknown-catalog rendering/hit-testing, annotation alignment/visibility, physical qualification and export/native parity remain open.
Group dragging now translates curved-wall control points with both endpoints, preserving curve shape. A dedicated geometry update applies the move together without normalizing unrelated height fields, and validates points before mutation. The existing shared drag retains one Undo/Redo step.
All 829 unit tests, eight Chromium/WebKit workflows, Svelte checks and build passed. Opposite curves, attached openings, grouped annotations, locked members and an unselected curved wall are covered with exact exported-floor comparisons; straight-wall dragging and Ctrl/Cmd isolation also pass. See validation. Opening-only group movement, unknown-catalog handling, annotation alignment/visibility, physical qualification and export/native parity remain open.
Furniture with an unavailable catalog entry now renders a generic labeled symbol and supports selection, movement, rotation, resizing and deletion. Drawing, hit testing, bounds, minimap and property defaults agree on saved dimensions or the shared 50 cm fallback. Mirrored furniture captions remain readable.
All 834 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed. Exact exports verify preserved catalog identity and metadata plus Undo/Redo; phone screenshots cover saved mirrored dimensions and omitted defaults. See validation. Unknown-catalog 3D/export parity, opening-only group movement, annotation alignment/visibility, physical qualification and broader native parity remain open.
PNG/PDF/SVG/DXF furniture footprints now preserve nonuniform scale and share the editor’s 50 cm missing-catalog fallback. Missing entries receive a readable label. DXF applies its vertical-axis inversion after rotating each corner, correcting previously reversed rectangle orientation. Saved project data is unchanged.
All 836 unit tests, four Chromium/WebKit workflows, Svelte checks and build passed. Tests check all four formats, exact DXF corners, saved/default sizes, export bounds and real downloads; scaled PNG visual review passed. See validation. Unknown-catalog 3D rendering, detailed furniture export symbols, opening-only group movement, annotation alignment/visibility, physical qualification and native parity remain open.
Saved furniture with unavailable catalog entries now reaches the existing box fallback in the main viewer. Saved dimensions/color or shared 50 cm/gray defaults are preserved, with rotation, signed plan scale and height-scale magnitude applied. Known 2D-only symbols remain excluded and saved data remains unchanged.
All 839 unit tests, Svelte checks and build passed. Four new Chromium/WebKit desktop/phone rendering workflows and six existing resource workflows have passing results; screenshots confirm both saved-size and default-size objects are visible. The Chromium textured-resource test needed longer startup/total allowances for software rendering; the final full loop passed unchanged resource assertions. See validation for those reruns. Detailed furniture export symbols, opening-only group movement, annotation alignment/ visibility, physical qualification and broader native parity remain open.
Notes, measurements and offset dimensions now participate in all eight alignment/ distribution operations using measured bounds at a fixed world scale. Endpoint vectors and annotation metadata remain intact, locked object anchors retain their existing behavior, and each operation is one Undo/Redo step with repeated no-ops excluded from history.
All 16 final Chromium/WebKit workflows, Svelte checks and build passed. Unit coverage adds all eight annotation operations; the full run passed 845/847 tests and both unrelated timeout files passed in isolation. Phone visual review passed. See validation. Walls/openings alignment, annotation visibility and multiline property editing, opening-only group movement, detailed furniture exports, physical qualification and native parity remain open.
The note text field now uses a resizable textarea, preserving imported line breaks and blank lines and supporting Enter during editing. Exact exported data verifies text-only changes, unchanged note metadata and Undo/Redo; typing shortcut letters in the field does not trigger plan actions. All 15 focused unit tests, six Chromium/WebKit workflows, Svelte checks and build passed. Phone visual review passed. See validation. Annotation visibility, walls/openings alignment, opening-only group movement, detailed exports, physical qualification and native parity remain open.
Text notes now have visibility controls, and all three annotation types respect hidden state in drawing, hit testing, marquee and Select All. Hiding clears their active selection; explicit selection in Layers reveals the category. Hidden notes are omitted from Fit/minimap while saved data and exports remain intact.
All 848 unit tests, 14 Chromium/WebKit workflows, Svelte checks and build passed. Exact exports verify hidden-item protection from Delete/Select All, Undo and reveal; existing auxiliary selection checks also pass. See validation. Walls/openings alignment, opening-only group movement, detailed furniture exports, physical qualification and broader native parity remain open.
Walls now participate in all eight alignment/distribution operations, using bounds that include thickness and quadratic curve extrema. Each move translates both endpoints and the curve control point together, preserving height profiles and metadata. Attached openings follow their host without changing normalized data. Only selected walls move; unselected neighboring walls remain unchanged.
All 856 unit tests, 16 Chromium/WebKit workflows, Svelte checks and build passed. Tests verify all operations, curve shape, hosted openings, no-op history and exact Undo/Redo; phone visual review passed. See validation. Independent opening alignment, opening-only group movement, detailed furniture exports, physical qualification and broader native parity remain open.
Saved door/window groups now move on the first drag even when their host walls are outside the selection. Each original center is translated then projected onto its own host using existing wall constraints. Openings on selected hosts are not moved twice. Host geometry and opening metadata remain intact with one Undo/Redo step. Curved or differently oriented hosts constrain movement independently.
All 858 unit tests, 12 Chromium/WebKit workflows, Svelte checks and build passed. Two additional phone screenshot workflows passed; fitted visual review passed. See validation. Independent opening alignment, detailed furniture exports, physical qualification and native parity remain open.
PNG/PDF exports now use the editor’s catalog-specific furniture renderer, preserving symbol details, color, rotation and signed scale with readable mirrored captions. Unknown entries keep their fallback symbol, with no selection handles or project mutation. All 859 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed. Gallery visual review and oversized furniture-bound regressions passed. See validation. Detailed SVG/DXF furniture symbols, independent opening alignment, physical qualification and broader native parity remain open.
SVG exports now reuse the furniture icon registry through a vector drawing adapter, preserving paths, curves, ellipses and symbol text without embedded raster images. Geometry retains rotation and signed scale; captions remain readable outside the mirror transform. All 861 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed. Every catalog entry is covered by adapter tests; decoded SVG gallery visual review and export-bound regressions passed. See validation. Detailed DXF furniture symbols, independent opening alignment, physical qualification and broader native parity remain open.
DXF exports now reuse catalog furniture geometry as editable lines, quadratic splines and exact rational elliptical arcs, preserving scale, mirroring and rotation. Duplicate fill/stroke outlines are removed; unknown entries retain their rectangular footprint. Output uses the existing monochrome furniture layer. All 864 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed; downloaded DXF geometry visual review and export-bound regressions passed. See validation. Independent opening alignment, physical qualification and broader native parity remain open.
Openings now support all eight alignment/distribution operations along their host walls, using complete symbol bounds and curved-wall tangents. The closest reachable position is used when a target cannot be met; hosts are preserved, and selecting a host carries its openings once. Mixed object selections and one-step Undo/Redo are covered. The full 874-test unit suite and expanded 37-test alignment suite passed, along with 16 Chromium/WebKit workflows, Svelte checks and build. Phone controls were visually reviewed. See validation for constraints and evidence. Physical qualification and broader native parity remain open.
PNG/PDF plans now draw all four stair shapes with the shared editor renderer, including treads, rotation, direction arrows and captions. Stair-only plans are accepted, and shared footprint/caption bounds prevent clipping. All 879 unit tests, four Chromium/WebKit workflows, Svelte checks and build passed. Downloaded PNG and PDF plan-image visual reviews passed. See validation. SVG/DXF stair symbols, physical qualification and broader native parity remain open.
SVG now shares the editor’s stair renderer through the vector symbol adapter, retaining all four shapes, treads, directional arrows, labels and rotation as editable geometry. Stair-only SVG plans and complete caption/footprint framing are supported. All 884 unit tests, four Chromium/WebKit workflows, Svelte checks and build passed; downloaded SVG visual review and furniture regressions passed. See validation. DXF stairs, physical qualification and broader native parity remain open.
DXF now shares the stair renderer through a CAD symbol adapter, retaining all four shapes, treads, arrows, labels and rotation as native lines, exact curve splines and text on a STAIRS layer. Stair-only plans now work across all four plan formats. All 889 unit tests, four Chromium/WebKit workflows, Svelte checks and build passed; downloaded CAD geometry visual review and furniture regressions passed. See validation. Physical qualification, stair voids and broader native parity remain open.
SVG exports now include built-in entourage as editable paths and custom symbols as embedded images, preserving aspect ratio, rotation and opacity. Entourage-only plans are accepted and rotated symbols are fully framed. All 890 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed; decoded SVG image/pixel checks, visual review and stair/furniture regressions passed. See validation. PNG/PDF entourage framing and image readiness, DXF entourage, physical qualification and native parity remain open.
PNG now frames entourage-only plans and waits for custom images before drawing. Failures/timeouts report through export feedback; same-ID image replacements refresh the cache. Export snapshots retain their own prepared images during loading. All 894 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed, plus the final focused suite. Delayed-image and downloaded PNG visual checks passed. See validation. PDF entourage framing/readiness, DXF entourage, physical qualification and native parity remain open.
PDF now frames entourage-only plans and waits for custom images. The export keeps a plan/image snapshot and captures its optional main 3D view before waiting, preventing later viewport substitution. Async failures report through both export entry points without placeholder downloads. The full 894-test unit suite and expanded 32-test PDF/export suite passed, along with eight Chromium/WebKit workflows, final Svelte checks and build. PDF plan-image visual review passed. See validation. DXF entourage, physical qualification and broader native parity remain open.
DXF now exports built-in entourage as native lines and exact curve splines on an ENTOURAGE layer, preserving position, width/aspect and rotation. Built-in entourage-only plans are accepted; invisible symbols are omitted. All 910 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed. Downloaded CAD visual review and furniture/stair regressions passed. See validation. Custom raster DXF entourage, partial CAD transparency, physical qualification and native parity remain open.
Built-in entourage now writes saved partial opacity as native DXF transparency attributes on its own entities. Other geometry is unaffected, and repeated serialization does not duplicate tags. All 911 unit tests, six Chromium/WebKit workflows, Svelte checks and build passed; downloaded opacity preview and furniture/stair regressions passed. See validation. Custom raster DXF entourage, physical CAD/plot qualification and broader native parity remain open.
Floor settings now support a positive slab thickness per floor, with a 5 cm legacy default, reset, input validation and undo/redo. Active and stacked 3D use the saved depth below the existing surface. JSON save/reimport and actual exported mesh bounds passed in Chromium and WebKit; visual review passed. All 913 unit tests, Svelte checks and production build passed. See validation, including the initial Chromium screenshot timeout and passing rerun. Stair/courtyard openings, wall-face offsets, native slab authoring/rendering and physical qualification remain open.
Per-floor slab depth now crosses project packages in centimetres/metres, preserving legacy web/native defaults. Native decoding and package merging retain/validate the value; selected-floor render exports and single-storey previews use the saved depth below zero. Native Catalyst build and 16 tests passed, as did Svelte checks, web build and two Chromium/WebKit workflows. The 915-test web run had one stress test timeout during native compilation; its six-test file passed in isolation. See validation. Native controls, room-shaped slabs, stair/courtyard openings, stacked native preview/elevations and physical qualification remain open.
The native Statistics & Defaults sheet now edits each floor’s slab depth, with
fractional unit-aware input, validation, default reset and snapshot undo. Final
Catalyst build and 17 tests passed. Native UI/visual checks verified imperial
editing, invalid recovery, reset and undo. See
openplan3d-ios/docs/native-slab-controls-validation.md. Native room-shaped slabs,
openings, stacked preview/elevations and physical-device qualification remain open.
Native edited-plan previews now separate populated floors by saved elevation, using legacy 3 m level spacing and per-floor slab depths. Elevations cross the package bridge with unit conversion and reset/validation support. Single-floor portable exports remain at local zero. All 916 web tests, Svelte checks/build, two Chromium/WebKit package workflows and 21 native Catalyst tests passed. Native three-storey import/visual review passed. See validation. Native elevation controls, room-shaped slabs, openings, curved/sloped walls and device qualification remain open.
The native defaults sheet now edits per-floor elevations with fractional/negative
input, validation, default reset and undo. Catalyst build and 22 tests passed;
native UI/visual checks verified edits, invalid recovery, resets and undo. See
openplan3d-ios/docs/native-elevation-controls-validation.md. Native room-shaped
slabs, openings, curved/sloped walls and physical-device qualification remain open.
Native previews/render exports now use closed straight-wall room faces, preserving
concave recesses and disconnected-room gaps with saved slab depth/elevation.
Crossings, T-junctions and duplicate edges are handled; open walls/furniture alone
create no slab. Catalyst build and 25 tests plus native visual QA passed. See
openplan3d-ios/docs/native-room-slabs-validation.md. Courtyard/stair voids,
interior-face area agreement, curves, large-graph performance and device gates
remain open.
Native slab boundary detection now filters wall pairs and uses spatial vertex
lookup. A local 2,000-wall/500-room fixture improved from 2.25 s to 0.069 s while
retaining all rooms. Catalyst build and 28 tests passed, including dense rotated
grids and snapping-boundary coverage. See
openplan3d-ios/docs/native-room-graph-performance.md. Worst-case memory and
physical-device profiling remain open alongside voids and area parity.
Native preview/export now enforce graph and slab triangle budgets, with a
localized preview error for excessive geometry. Dangling wall trees are removed
from boundary tracing while preserving the original walls in the scene.
Catalyst build and 32 selected tests passed, including pathological limits,
branching partitions, and a 1,500-segment detached tree. See native
docs/native-room-graph-limits-validation.md and
docs/native-partition-preprocessing-validation.md. This does not complete
nested boundaries, courtyard/stair holes, area parity, or device profiling.
Nested room rings now partition active/stacked slab geometry and active finishes without overlapping the child room. Strict containment excludes touching and crossing rings; portable scene exports retain the hole geometry. All 919 unit tests, four Chromium/WebKit slab workflows, Svelte diagnostics, and build passed. See validation. Explicit courtyard/stair voids, nested 2D fills/area accounting, native nesting, and wall-face areas remain open.
Room detection now subtracts immediate nested footprints before rounding; area labels and summaries no longer double-count contained rooms. Area Summary uses current resolved geometry instead of stale saved areas, preserving room metadata. All 921 unit tests, 18 Chromium/WebKit slab/modal workflows, Svelte checks, and build passed. See validation. Nested 2D fills, explicit courtyard/stair openings, native nesting, interior-face areas, and physical measurement qualification remain open.
Parent room colors and floor patterns now clip around immediate nested rooms in the 2D editor. PNG/PDF plan images and SVG use compound paths, avoiding blended overlap colors. All 921 unit tests, four Chromium/WebKit slab workflows with downloaded-image color checks, Svelte checks, and build passed. See validation. Explicit openings, nested label placement/selection, native nesting, and physical qualification remain open.
Room and furniture-guide hit testing now selects the smallest enclosing footprint instead of the first room. Closest labels win, with innermost-room tie-breaking for coincident anchors; moved labels retain their saved positions. All 922 unit tests, six Chromium/WebKit label/slab workflows, Svelte checks, and build passed. See validation. Automatic label placement, explicit openings, native nesting, and physical gestures remain open.
Default label anchors now lie inside concave/nested visible floors, shared across 2D/3D, hit testing, Fit, and image/vector/CAD export labels. Saved offsets retain their coordinate convention; dragging starts at the displayed anchor and undo restores automatic placement. All 923 unit tests, Svelte checks/build, eight browser regressions and two corrected Chromium/WebKit nested workflows passed. See validation for the coordinate-test failure and rerun. Full text collision handling, stale Room Properties areas, explicit openings, native nesting, and physical qualification remain open.
Room Properties now resolves selected geometry instead of displaying stale saved areas, retaining boundary-matched names and finishes. The nested fixture verifies 20/12/4 m² despite saved 999 m² values. All 923 unit tests, Svelte checks/build, five initial browser workflows, and one isolated Chromium timeout rerun passed. See validation. Explicit openings, native nesting, full text collisions, and physical qualification remain open.
Room Properties now provides Open to floor below. The saved room flag removes its active/stacked slab and finish, leaves the surrounding nested hole intact, and excludes its area. Editor and image/vector fills leave the opening empty. Undo/redo, JSON reimport, and web metadata preservation through native package returns are covered. All 926 unit tests, checks/build, and four final Chromium/ WebKit workflows passed. See validation, including the onboarding-test race correction. Arbitrary/stair-derived holes, inter-floor ceiling coordination, native opening authoring/rendering, and physical qualification remain open.
Native documents and package merges now retain optional boundary wall IDs and floor-opening intent, including explicit resets. Web package imports use those boundaries to distinguish nested rooms with coincident label centers; exports reuse existing wall identities. All 927 web unit tests, Svelte checks/build, and 33 selected Catalyst tests passed. Four Chromium/WebKit package and nested-slab workflows also passed. See validation. Native opening geometry/authoring, arbitrary holes, and ceiling coordination remain open.
Native edited-plan previews and portable render exports now exclude nested room
rings and honor floor-opening metadata. Catalyst build and all 36 selected tests
passed, including watertight net volumes, floor coverage, multiple holes, input
reversal, stacked floor isolation, and exported openings. See
openplan3d-ios/docs/native-nested-slabs-validation.md. Native opening controls
and 2D/area display, arbitrary cuts, ceiling coordination, and device QA remain open.
Native Room Properties now edits floor-opening intent. Native room labels and
statistics exclude opening area; canvas/SVG omit the marked room’s color fill.
Catalyst build and all 39 selected tests passed. See
openplan3d-ios/docs/native-opening-controls-validation.md. Native boundary-aware
raster association, duplicate-label totals, measurement parity, arbitrary cuts,
ceiling coordination, and device/UI QA remain open.
Native rooms with explicit boundary IDs retain area/fill association when labels
move or share centers. Auto-Label recognizes those regions, and raster bounds
now depend only on walls. Catalyst build and all 42 selected tests passed. See
openplan3d-ios/docs/native-room-regions-validation.md. Automatic native boundary
assignment, duplicate-label totals, batch performance, measurement parity,
arbitrary cuts, ceiling coordination, and device/UI QA remain open.
Native canvas, statistics, Auto-Label, and SVG reuse floor association work across
labels. Catalyst build and all 43 selected tests passed; a local 16-room fixture
returned identical seeds in 0.315 s batched versus 5.915 s individually. See
openplan3d-ios/docs/native-room-batch-validation.md. Per-region raster costs,
automatic boundary assignment, duplicate-label totals, larger/device profiling,
and measurement parity remain open.
Native statistics now count each labelled raster region once per floor and
exclude regions marked open even when duplicate labels disagree. Catalyst build
and all 45 selected tests passed. See
openplan3d-ios/docs/native-duplicate-area-validation.md. Automatic boundary
assignment, duplicate-label presentation, batch raster work, measurement parity,
arbitrary cuts, ceiling coordination, and device QA remain open.
New native room labels now receive boundary IDs when their walls reconstruct an
unambiguous room; moving associated labels preserves those IDs. Catalyst build
and all 47 selected tests passed, including nested/shared-wall rooms and JSON
retention. See openplan3d-ios/docs/native-boundary-assignment-validation.md.
Deliberate reassociation, ambiguous graphs, duplicate-label presentation, batch
raster work, measurement parity, arbitrary cuts, and device QA remain open.
Native Room Properties can deliberately reassign a label to the room beneath
its position, preserving metadata and retaining the old association on failure.
Catalyst build and all 49 selected tests passed. See
openplan3d-ios/docs/native-room-reassignment-validation.md. Ambiguous graphs,
duplicate-label presentation, batch raster work, measurement parity, arbitrary
cuts, ceiling coordination, and device/UI QA remain open.
Interactive Catalyst QA verified opening/reassignment controls, Undo/Redo, and
save/reopen. Stale selection-bar area and duplicate preview/PDF summary totals
were corrected; final UI showed 0 ft² for the selected opening and 319 ft² for
the saved QA plan. All 49 selected tests passed. See
openplan3d-ios/docs/native-room-ui-validation.md. Duplicate-label overlap/fills,
PDF visual checks, ambiguous graphs, raster batching, and device QA remain open.
Native canvas/SVG paint each labelled region once and omit it when any duplicate
marks it open. All 51 selected tests passed; interactive QA confirmed the inner
opening lost its stale green fill while retaining the outer fill and area total.
See openplan3d-ios/docs/native-duplicate-fills-validation.md. Duplicate text
overlap, raster batching, measurement parity, ambiguous graphs, and device QA
remain open.
Native canvas/SVG separate colliding room labels without changing saved anchors.
Editor selection follows displayed bounds, and drags retain their anchor offset.
All 53 selected Catalyst tests passed; native visual QA confirmed three formerly
overlapping labels are readable. See
openplan3d-ios/docs/native-label-layout-validation.md. Viewport edges, dense
layouts, other-object collisions, and pointer/touch/device qualification remain open.
Native canvas/SVG label placement moves fitting labels inward at viewport edges
and tries upward stacking near the bottom. Selection shares those bounds. All 54
selected Catalyst tests passed. See
openplan3d-ios/docs/native-label-edges-validation.md. Dense/oversized labels,
toolbar/other-object collisions, and physical gesture qualification remain open.
Native labels use nearby columns when vertical stacking cannot fit. The bounded
search preserves text sizes and saved anchors. All 55 selected Catalyst tests
passed, including a nine-label multi-column case. See
openplan3d-ios/docs/native-dense-label-validation.md. Over-capacity layouts,
oversized text, toolbar/other-object occlusion, and physical gestures remain open.
Statistics no longer uses a whole-document area fallback for unresolved rooms.
An unenclosed upper room cannot borrow ground-floor area at the same position;
ordinary unresolved rooms remain unknown and openings contribute zero. All 56
selected Catalyst tests passed. See openplan3d-ios/docs/native-statistics-floor-isolation-validation.md.
Raster batching, measurement parity, ambiguous graphs, PDF visual checks, and
physical-device qualification remain open.
Statistics now rasterizes once per floor and visits each requested component
once, reusing results for duplicate labels and unresolved exterior regions.
All 57 selected Catalyst tests passed, retaining duplicate/opening and floor
isolation behavior. See openplan3d-ios/docs/native-statistics-batch-validation.md.
Renderer fill batching, device performance qualification, measurement parity,
PDF visual checks, and other NEXT work remain open.
Canvas/SVG fill ownership now shares Statistics’ per-floor batch measurement
helper, preserving first-colored-label ownership and opening suppression. All
58 selected Catalyst tests passed, including distinct rooms with duplicate
labels and an opening. See openplan3d-ios/docs/native-fill-owner-batch-validation.md.
Fill-rectangle batching, renderer area queries, device performance qualification,
measurement parity, PDF visual checks, and other NEXT work remain open.
Room label, property, and selection area queries now share a bounded batch
cache keyed by walls and room associations. All 59 selected Catalyst tests
passed, including cache refresh after label, level, boundary, and wall edits
and opening toggles. See openplan3d-ios/docs/native-renderer-area-batch-validation.md.
Fill-rectangle batching, device performance qualification, measurement parity,
PDF visual checks, and remaining NEXT work are still open.
Single-seed area and fill queries now reject nonfinite and overflowing grid
coordinates before integer conversion. All 60 selected Catalyst tests passed,
including invalid inputs across single and batch queries and valid interiors.
See openplan3d-ios/docs/native-raster-seed-validation.md. Fill-rectangle batching, device
performance qualification, measurement parity, PDF visual checks, and remaining
NEXT work are still open.
The complete Catalyst FloorPlanTests target passed after the room/model/renderer
changes: 220 tests executed, two optional worker-integration skips, zero failures.
The available private-scan byte-preservation test also passed. See openplan3d-ios/docs/native-room-full-integration-validation.md.
Fresh simulator, interactive and physical-device checks, worker integration,
signed release validation, and the broader NEXT requirements remain open.
The complete FloorPlanTests target passed on iPhone 17 Pro / iOS 26.5 simulator:
220 tests executed, two optional worker-integration skips, zero failures. This
complements the Catalyst run after the room/model/renderer changes. See openplan3d-ios/docs/native-room-ios-integration-validation.md.
Physical gestures/capture, measured calibration, device performance, signed
release validation, and the broader NEXT requirements remain open.
Renderer room fills now share a 10 cm wall grid per floor and a bounded cache
of rectangles. Batch output matches individual fills and opening toggles refresh
correctly. The full Catalyst target passed: 221 tests, two optional skips, zero
failures. See openplan3d-ios/docs/native-fill-grid-batch-validation.md. SVG batching, shared
component/run indexing, device performance, area parity, PDF visual checks, and
remaining NEXT work stay open.
SVG label areas and fill ownership share batch region measurements, and room
color paths share one fill grid per floor. Existing geometry and presentation
conventions remain. Catalyst build and all 25 selected tests passed. See
openplan3d-ios/docs/native-svg-batch-validation.md. Shared fill-component indexing,
device performance, area parity, PDF visual checks, and other NEXT work remain.
The 16-room fixture now verifies exact batch/individual area and fill output.
A local Catalyst run measured areas at 0.075 s batch / 0.867 s individual and
fills at 0.175 s / 0.525 s. All ten room-region tests passed. See openplan3d-ios/docs/native-room-batch-measurement.md.
These are single-run debug measurements, not device budgets. Shared fill indexing,
active-editing/device measurements, area parity, PDF QA, and other NEXT work remain.
Batch color fills now share component traversal and one row scan, preserving
ordered rectangles and rejecting exterior regions. The full Catalyst target
passed: 221 tests, two optional skips, zero failures. A local 16-room run measured
0.024 s batch / 0.300 s individual with exact output equality. See openplan3d-ios/docs/native-fill-component-validation.md.
Device latency/memory targets, area parity, PDF visual checks, and remaining
NEXT requirements stay open.
Each floor PDF page now reports its own room count and area. Twelve opening
presentation tests passed, including PDF text and rendered-wall assertions.
Both pages of a metric Letter fixture were visually reviewed: ground fill/area
and upper opening with zero label area are correct. See openplan3d-ios/docs/native-pdf-floor-summary-validation.md.
Long titles, paper-size/large-plan coverage, device memory/share sheets, print
scale, and broader NEXT requirements remain open.
PDF notes, up-to-two-line titles, and statistics now occupy separate footer rows.
Thirteen targeted tests passed; Letter, A4, and A3 long-title fixtures were
rendered and visually checked without footer overlap. See openplan3d-ios/docs/native-pdf-title-layout-validation.md.
Large-plan/device memory, physical sharing, actual print scale, and broader NEXT
requirements remain open.
PNG/PDF plan bitmaps now cap both axes at 2080 pixels, preventing extreme
portrait aspect ratios from allocating unbounded-height images. PDF scale uses
the actual fitted transform. All 14 targeted tests passed, including an actual
4-by-3000 m PNG export at 920 by 2080 pixels. See openplan3d-ios/docs/native-export-canvas-bounds-validation.md.
Total device memory, physical printing/sharing, and other NEXT work remain open.
PNG/PDF/SVG share UTF-8-bounded filename sanitization with a hash suffix for
truncated titles, preserving full document/export content. Fifteen targeted
tests passed, including actual long CJK/emoji PNG and SVG exports. See
openplan3d-ios/docs/native-export-filenames-validation.md. Physical sharing/storage recovery,
device memory, and broader NEXT requirements remain open.
PDF page-render failure now fails the export rather than skipping a page.
Staging plus atomic publication preserves previous same-name exports on failure;
PNG writes are atomic too. All 16 targeted tests passed, including failed-render
byte preservation and successful retry. See openplan3d-ios/docs/native-export-atomic-validation.md.
Disk/quota faults, device sharing, full/iOS revalidation, and other NEXT work remain.
The complete iPhone 17 Pro / iOS 26.5 simulator target passed: 226 tests, two
optional worker-integration skips, zero failures. One title assertion was fixed
to tolerate an observed line break; visual review confirmed the title was intact.
See openplan3d-ios/docs/native-ios-export-integration-validation.md. Physical sharing, quota faults,
device memory, printing/release qualification, and remaining NEXT work stay open.
Actual PDF/PNG/SVG destination-directory collisions now have regression coverage:
export fails, preserves existing contents, cleans PDF staging, and succeeds on
retry after the obstruction is removed. All 17 targeted Catalyst tests passed.
See openplan3d-ios/docs/native-export-write-failure-validation.md. Full-disk/quota/permission faults,
process crashes, physical share sheets, and other NEXT work remain open.
Statistics, raster regions, canvas/SVG strokes, and native 3D walls now honor
document height/thickness defaults while preserving per-wall overrides. Renderer
caches refresh when defaults change. The full Catalyst target passed: 228 tests,
two optional skips, zero failures. See openplan3d-ios/docs/native-wall-defaults-validation.md.
Exterior-wall inference, UI/iOS checks, area parity, device measurement, and other
NEXT requirements remain open.
Properties and Elevation now display effective wall defaults. Edited-render
export validates those same dimensions and retains explicit override precedence.
Catalyst build and all 30 selected tests passed. See openplan3d-ios/docs/native-wall-default-consumers-validation.md.
Interactive UI/iOS qualification, exterior inference, area parity, and remaining
NEXT requirements stay open.
Rasterization validates effective thickness and caps candidate-cell stamping
work before integer conversion, returning unknown regions on invalid/over-budget
input. Full Catalyst validation passed: 230 tests, two optional skips, zero
failures. See openplan3d-ios/docs/native-raster-thickness-budget-validation.md. Device latency, resource-limit
UI feedback, area parity, and remaining NEXT requirements stay open.
Statistics marks incomplete totals as partial and no longer assumes all missing
areas mean outside labels. PDF footers identify measured subtotals and unmeasured
counts. Twenty targeted tests and a focused rerun passed; the partial PDF footer
was visually checked. See openplan3d-ios/docs/native-partial-area-feedback-validation.md. Exact failure
diagnostics, interactive/device checks, area parity, and other NEXT work remain.
The full iPhone 17 Pro / iOS 26.5 simulator target passed after wall defaults,
raster guards, export write-failure checks, and partial-area feedback: 231 tests,
two optional skips, zero failures. See openplan3d-ios/docs/native-ios-wall-area-integration-validation.md.
Interactive/device qualification, quota faults, performance, printing, releases,
and remaining NEXT requirements stay open.
The preview marks partial or unavailable area and reports unmeasured-room
counts. Saved-plan reload coverage verifies partial, all-unresolved, and
resolved-plus-opening states. Catalyst build and all 31 selected tests passed.
See openplan3d-ios/docs/native-preview-partial-area-validation.md. Interactive/device qualification,
failure diagnostics, area parity, and remaining NEXT requirements stay open.
A separate stored-ZIP QA fixture imported successfully. Catalyst preview showed
355 ft² with one unmeasured room; Statistics displayed partial living/gross totals
and the full warning without clipping. See openplan3d-ios/docs/native-partial-area-ui-validation.md.
Device/touch checks, all-unresolved interactive layout, diagnostics, area parity,
and remaining NEXT requirements stay open.
Room Properties explains unresolved measurements without assuming an outside
label. Statistics describes distinct measured regions, opening exclusions, and
the approximate half-wall-footprint formula. Catalyst build passed; see openplan3d-ios/docs/native-area-help-validation.md.
Interactive/device qualification, exact diagnostics, area parity, and remaining
NEXT requirements stay open.
The full iPhone 17 Pro / iOS 26.5 simulator target passed after preview notices
and area-help corrections: 232 tests, two optional skips, zero failures. The
partial/unavailable/complete saved-plan regression passed. See openplan3d-ios/docs/native-ios-preview-integration-validation.md.
Interactive/device checks, area parity, performance, release gates, and remaining
NEXT requirements stay open.
Paired 4 × 3 m rectangle fixtures measured three wall thicknesses. At 20 cm,
web reports 12.00 m², native 10.31 m², and exact interior faces give 10.64 m².
This isolates centerline/interior convention differences and native raster error.
Eleven native region tests and the web characterization test passed. See openplan3d-ios/docs/area-convention-baseline.md.
Convention adoption, general geometry fixtures, numerical correction, and the
remaining NEXT requirements stay open.
Wall cells now use footprint containment instead of filled square stamps. All
three rectangle references match exact interior areas; rotated/translated rooms
pass a bounded-error regression. Full Catalyst: 233 tests, two skips, zero failures;
then all 12 region tests passed. See openplan3d-ios/docs/native-raster-footprint-validation.md.
Fresh iOS/UI checks, runtime budgets, general geometry and web area parity remain
open alongside the other NEXT requirements.
Full iPhone 17 Pro / iOS 26.5 validation passed after the wall-footprint fix:
234 tests, two optional skips, zero failures, including rotated/translated
room references. See openplan3d-ios/docs/native-ios-raster-footprint-validation.md. Interactive/device checks, general
error bounds, performance budgets, area parity and remaining NEXT work stay open.
Concave L, shared-partition and nested-room measurements match independent
interior-face references; an inner floor opening correctly removes only its area.
All 14 Catalyst region tests passed. See openplan3d-ios/docs/native-interior-shapes-validation.md. General geometry,
area parity, device/performance checks and remaining NEXT work stay open.
A candidate-loop shortcut passed 14 region tests but showed no reliable timing
improvement in the debug sample; it was removed. See openplan3d-ios/docs/native-raster-reuse-experiment.md.
Repeated release measurements/profiling are needed before another optimization;
production geometry and the remaining NEXT requirements are unchanged.
Three optimized M1 Max runs passed the 16-room batch/individual agreement check.
Median batch times: association 12.94 ms, area 6.54 ms, fills 1.85 ms. See openplan3d-ios/docs/native-release-room-measurement.md.
Local test signing required hardened runtime disabled by command-line override;
release packaging, active-editor/device budgets and other NEXT work remain open.
Split/merge coverage verifies stale-boundary detection, deliberate reassociation,
ID/note preservation, JSON reload, snapshot restoration and deduplicated merged
area. All 15 Catalyst region tests passed. See openplan3d-ios/docs/native-split-merge-identity-validation.md. Interactive
undo/usability, schedule behavior, area parity and other NEXT work remain open.
SVG and full-detail plan labels explicitly show unavailable area; low-zoom
labels stay compact and floor openings show zero. All 21 Catalyst presentation
tests passed, including clearing the notice after resolution. See openplan3d-ios/docs/native-unresolved-labels-validation.md.
Dense-plan visual/device checks, exact diagnostics and remaining NEXT work stay open.
Initial editor fit now reserves vertical space for floating tools. A previously
hidden unmeasured-room label is visible above the toolbar in fresh Catalyst QA;
22 presentation tests passed. See openplan3d-ios/docs/native-editor-toolbar-fit-validation.md. Read-only preview framing,
adaptive toolbar sizes, device checks and remaining NEXT work stay open.
Read-only Plan now fits between measured top/bottom safe-area content. Fresh
Catalyst QA showed the complete fixture and unavailable-area label above the
card; switching to 3D remained normal. Build passed; see openplan3d-ios/docs/native-preview-safearea-validation.md.
Device, Dynamic Type, full-screen and remaining NEXT checks stay open.
The full iPhone 17 Pro / iOS 26.5 target passed after geometry references,
split/merge, unavailable labels and preview/editor framing: 239 tests, two
optional skips, zero failures. See openplan3d-ios/docs/native-ios-label-framing-validation.md. Physical-device/layout
qualification, area parity, release gates and remaining NEXT work stay open.
Document bounds include rotated furniture footprints, correcting center-only
fitting of furniture-only plans and outlying objects. Full Catalyst validation:
240 tests, two skips, zero failures. See openplan3d-ios/docs/native-furniture-fit-validation.md. Visual/iOS/device
checks, decorative/text extents and remaining NEXT requirements stay open.
The rotated-table fixture now exercises actual PNG export. The focused test
passed and visual inspection showed the entire table/chair details with margins.
See openplan3d-ios/docs/native-furniture-export-validation.md. Broader categories/formats, device checks and remaining
NEXT requirements stay open.
Full iPhone 17 Pro / iOS 26.5 validation passed after rotated furniture bounds
and actual PNG export coverage: 240 tests, two optional skips, zero failures.
See openplan3d-ios/docs/native-ios-furniture-fit-validation.md. Broader visuals, physical-device checks, release gates
and remaining NEXT work stay open.
Editor fit uses measured top/bottom control stacks, including context actions,
with a stable viewport during a drag. Catalyst build passed; see openplan3d-ios/docs/native-adaptive-toolbar-framing.md.
Expanded-control/cancelled-gesture/Dynamic Type/iOS interactive checks and the
remaining NEXT requirements stay open.
Gesture-state reset now restores a cancelled pending edit/pan and releases its
frozen fitting viewport. Normal completion keeps its undo behavior. Catalyst
build passed; see openplan3d-ios/docs/native-drag-cancellation.md. Runtime gesture ordering, interruption,
pinch/device checks and remaining NEXT requirements stay open.
Multiple running builds shared one bundle identifier, making the adaptive UI
check inconclusive. A uniquely identified QA copy now launches; its synthetic
fixture import was still pending at last observation. See openplan3d-ios/docs/native-isolated-ui-qa.md.
Inspect that same import before retrying. Gesture/layout qualification and all
remaining NEXT requirements stay open.
Preserving original sandbox entitlements fixed the isolated QA setup; its
synthetic import completed. Fresh UI checks confirmed adaptive context-toolbar
fitting, a completed label drag and undo restoration. See openplan3d-ios/docs/native-sandbox-qa-validation.md.
Use OpenPlan3D-Sandbox-QA.app for new checks. Actual cancellation, pinch/device
interaction and remaining NEXT requirements stay open.
The full iPhone 17 Pro / iOS 26.5 target passed after measured toolbars and
pending-drag cancellation: 240 tests, two optional skips, zero failures. See
openplan3d-ios/docs/native-ios-adaptive-editor-validation.md. This does not simulate interrupted gestures; cancellation,
pinch/device checks and remaining NEXT requirements stay open.
A checked-in helper creates a distinct QA app while preserving sandbox
entitlements and refusing existing destinations. Syntax, signing, entitlement
equality and overwrite-refusal checks passed. See openplan3d-ios/docs/sandbox-qa-helper.md.
Runtime/device qualification, signed releases and remaining NEXT work stay open.
A new accessible editor action recomputes current-floor bounds and resets pan/
zoom. Build and sandboxed Catalyst UI checks passed: a cropped, panned sample
returned fully into view without an undo entry. See openplan3d-ios/docs/native-fit-current-floor-validation.md.
Multi-floor/changed-geometry/zoom/device checks and remaining NEXT work stay open.
A sandboxed UI check moved furniture beyond the original walls; Fit included
the new extent. One Undo restored the edit, and refitting restored the original
view without extra history. See openplan3d-ios/docs/native-fit-current-floor-validation.md. Multi-floor, zoom/iOS/device
checks and remaining NEXT requirements stay open.
The full iPhone 17 Pro / iOS 26.5 test target passed with Fit Current Floor:
240 tests, two optional skips, zero failures. See openplan3d-ios/docs/native-fit-current-floor-validation.md.
This covers compilation and regressions; multi-floor, zoom recovery, iOS UI
and physical-device checks remain open alongside the remaining NEXT work.
Switching floors now resets bounds/pan/zoom to the active floor; initial loading
uses that floor too. Catalyst build and an isolated empty-to-populated floor UI
check passed, with no extra undo entry. See openplan3d-ios/docs/native-floor-switch-fit-validation.md.
Distant populated floors, zoom, iOS/device checks and remaining NEXT work stay open.
Undoing creation of the active floor no longer leaves a nonexistent floor
selected. The editor selects an existing floor and refits without adding history.
Catalyst build and before/after UI checks passed; see openplan3d-ios/docs/native-floor-undo-recovery.md.
Broader history, multi-floor and iOS/device qualification remain open.
The full iPhone 17 Pro / iOS 26.5 target passed after automatic floor fitting
and removed-floor recovery: 240 tests, two optional skips, zero failures.
See openplan3d-ios/docs/native-floor-undo-recovery.md. iOS UI, broader history/navigation sequences
and physical-device qualification remain open with the wider NEXT backlog.
Reproduced an old-floor elevation remaining visible after selecting an empty
upper floor. Floor changes now clear mismatched elevation targets and return
to Plan; the canvas also requires matching floor ownership. Catalyst build
passed. See openplan3d-ios/docs/native-elevation-floor-switch.md; post-fix UI/iOS checks remain pending.
Fresh Catalyst UI confirmed floor switching clears the old elevation and restores
Plan tools. The full iOS 26.5 target passed: 240 tests, two optional skips, zero
failures. See openplan3d-ios/docs/native-elevation-floor-switch.md. Same-floor wall history, iOS interaction
and broader device/NEXT qualification remain open.
Undo/Redo now leave elevation when its wall disappears; direct deletion chooses
a replacement only on the active floor. Catalyst build and a drawn-wall Undo UI
check passed. See openplan3d-ios/docs/native-elevation-history.md. Redo/deletion, iOS/device and wider
NEXT qualification remain open.
Live Catalyst Redo removal restored Plan controls from the removed wall’s
elevation. Full iOS integration passed: 240 tests, two optional skips, zero
failures. See openplan3d-ios/docs/native-elevation-history.md. Cross-floor direct deletion and broader
iOS/device/NEXT qualification remain open.
Deleting the upper floor’s sole wall from elevation returned to Plan without
selecting a ground-floor wall. Temporary floor/geometry changes were undone.
See openplan3d-ios/docs/native-elevation-history.md. Elevation counter numbering still spans all floors
while cycling is floor-specific; align that counter in a follow-up. Broader
iOS/device qualification and remaining NEXT requirements stay open.
The elevation title and previous/next controls now share the active floor’s wall
collection, so their index/total agree. Catalyst build passed. See openplan3d-ios/docs/native-elevation-counter.md.
Fresh multi-floor UI verification and broader iOS/device work remain open.
Live Catalyst checks showed upper-floor counts of 1/1 and 2/2, with forward/back
wrap between 1/2 and 2/2 and matching lengths, excluding ground-floor walls.
Temporary edits were undone. See openplan3d-ios/docs/native-elevation-counter.md. Broader iOS/device
and remaining NEXT qualification stay open.
New tracing imports use unique filenames, preserving prior image bytes for
Undo. The displayed cache now follows the restored filename. Catalyst build
and nine package regressions passed. See openplan3d-ios/docs/native-trace-history.md. Live two-image
history and iOS integration remain pending with the broader NEXT backlog.
The full iOS 26.5 target passed after trace-history preservation and elevation
counter changes: 240 tests, two optional skips, zero failures. See openplan3d-ios/docs/native-trace-history.md.
Direct two-image Undo/Redo UI and broader device/NEXT qualification remain open.
Mac Trace Image now opens a local image file picker. File and Photos imports
share the same history-preserving save path; Catalyst build passed. See openplan3d-ios/docs/native-trace-file-import.md.
Live picker/two-image history checks and iOS integration remain pending.
The desktop file picker imported synthetic red/blue traces. Undo restored red
after removing it and importing blue; Redo restored blue. Temporary changes
were undone. See openplan3d-ios/docs/native-trace-file-import.md. Cancel/error, reopen/export and iOS
checks remain open with broader NEXT requirements.
Native Cancel returned without error or history changes. The full iOS 26.5
target passed: 240 tests, two optional skips, zero failures. See openplan3d-ios/docs/native-trace-file-import.md.
Invalid-file/write-failure, reopen/export and broader device checks remain open.
The desktop picker rejected a synthetic corrupt PNG with an import error.
Dismissing it preserved disabled Undo and available Redo. See openplan3d-ios/docs/native-trace-file-import.md.
Write-failure, reopen/export and broader device/NEXT checks remain open.
A saved synthetic trace reappeared after quitting/relaunching the isolated QA
app and reopening its editor. The persisted plan references its unique image
file. See openplan3d-ios/docs/native-trace-file-import.md. Initial Draw a Plan’s Done saved without
navigating away; investigate that behavior. Export/write-failure and broader
NEXT qualification remain open.
Editor Close/Done now explicitly clear the presenting view’s cover state.
Catalyst build and fresh Draw a Plan → Done → library UI check passed.
See openplan3d-ios/docs/native-editor-dismissal.md. Review/error/iOS paths and broader NEXT work remain open.
Both Close and Done returned a reopened editor to its saved-plan preview in
the isolated Catalyst build, with unchanged element counts. See openplan3d-ios/docs/native-editor-dismissal.md.
Error paths, iOS integration and broader NEXT qualification remain open.
The full iOS 26.5 target passed after explicit editor dismissal: 240 tests,
two optional skips, zero failures. See openplan3d-ios/docs/native-editor-dismissal.md. Load/save-error UI
and broader device/NEXT qualification remain open.
A controlled invalid destination in the isolated QA session kept the editor
open with an error. Restoring the original file preserved its hash; retry
succeeded and returned to preview. See openplan3d-ios/docs/native-editor-dismissal.md. Load-error UI,
full-disk/unsaved-edit cases and broader NEXT qualification remain open.
A truncated QA plan showed recovery controls instead of an editable fallback.
Error-screen Close returned to preview; restoring the original preserved its
hash and allowed a normal reopen. See openplan3d-ios/docs/native-editor-dismissal.md. Broader recovery
and device/NEXT qualification remain open.
A new storage regression verifies unique trace reference/placement/scale and
exact JPEG bytes through export, independent import, reload and re-export.
All ten package tests passed on Catalyst. See openplan3d-ios/docs/native-trace-file-import.md. Export-dialog,
write-failure and broader device/NEXT checks remain open.
All ten package tests passed on iOS 26.5, including unique trace reference and
JPEG preservation through export/import/re-export. See openplan3d-ios/docs/native-trace-file-import.md.
This focused run does not replace full-suite or device/UI qualification.
A newly drawn wall survived a controlled save failure and was written on retry.
The editor remained open after this changed-document save, unlike the prior
unchanged-document check. See openplan3d-ios/docs/native-editor-dismissal.md. Investigate preview-refresh/
dismissal ordering; changed-document dismissal remains unresolved.
Moving preview refresh after dismissal did not fix the changed-document editor
remaining open. The experiment was fully reverted. A fresh QA run reproduced
the issue without save-failure injection; five walls were saved. See openplan3d-ios/docs/native-editor-dismissal.md.
Changed-document dismissal remains unresolved.
Both previously open QA editors eventually returned to preview without more
clicks. Treat the issue as delayed dismissal; latency is unmeasured. A two-second
process sample during a new occurrence showed an idle main event loop, and
clearing selection did not immediately resolve it. See openplan3d-ios/docs/native-editor-dismissal.md.
Presentation timing remains under investigation.
Disabling close-transition animations did not establish prompt dismissal after
a wall edit. The experiment was reverted. See openplan3d-ios/docs/native-editor-dismissal.md. Instrument
callback/state timing before choosing another fix; dismissal delay remains open.
Temporary instrumentation measured save/write/close callbacks completing in
9.3 ms, including Home setting its cover item to nil. No cover onDismiss event
appeared within the next 52.4 seconds. See openplan3d-ios/docs/native-editor-dismissal.md. Instrumentation
was removed; investigate state application/cover lifecycle next.
Measured dismissal at 95.0 s after wall dragging versus 0.54 s with no edit
and 0.53 s after tool selection only. A pan-only drag also delayed completion
(>18.8 s), so geometry changes are not required. See openplan3d-ios/docs/native-editor-dismissal.md.
Investigate canvas gesture lifecycle and distinguish app from input effects.
Pan onEnded cleared its operation, gesture activity reset, and Save observed
no active drag/frozen viewport with gesture zoom 1.0. See openplan3d-ios/docs/native-editor-dismissal.md.
Temporary logging was removed; investigate lower-level presentation/input
behavior rather than resetting already-cleared app flags.
A canvas click without movement also delays dismissal despite cleared gesture
state and completed Save/Home close callbacks. Direct pointer Done activation
during the pending dismissal did not immediately resolve it. Movement is not
required; isolate canvas recognizers next. Details are in native
docs/native-editor-dismissal.md. No production source changes.
The dismissal delay reproduces in a diagnostic build with both canvas drag and
pinch recognizers removed (over 32 seconds after Home close). Removing pinch
alone also did not resolve it. Production source was restored; investigate
presentation and pointer-input behavior, not speculative drag-state resets.
See native docs/native-editor-dismissal.md for measured controls.
Transition logging plus a delayed diagnostic inspection produced prompt
dismissal in three controls, including a saved fifth wall. A fresh uninstrumented
build still showed the editor after saving; dispatching close asynchronously
also did not demonstrate a fix and was reverted. This remains timing-sensitive
and unresolved; see native docs/native-editor-dismissal.md. No production
gesture or scheduling workaround has been retained.
A real permission-denied image write through the desktop picker reports an
error without changing session files or adding history. Undo still restores
the original red trace. QA directory permissions were restored and all four
file hashes verified unchanged. See native docs/native-trace-file-import.md.
Desktop export-dialog and live-device checks remain open.
Exported the saved trace project through the native Save dialog. ZIP integrity,
complete plan value and exact original JPEG bytes passed independent checks.
The options sheet required Escape before the queued export proceeded, so
unassisted dismissal remains open. See native docs/native-trace-file-import.md.
Fixed the cached trace preview’s 2,000-pixel limit: UIKit points and default
Retina renderer scale previously allowed oversized bitmaps. Import/reload now
check source pixel dimensions and render at scale 1, preserving original saved
image bytes. Four new preview regressions reproduced the defect; all 14 preview
and package tests pass on Catalyst and iOS Simulator. See native
docs/native-trace-preview-pixels.md. Peak import memory/device budgets remain open.
Reopen/history reload now requests a transformed ImageIO thumbnail directly
from the saved file at the 2,000-pixel limit, avoiding the previous full-image
UIImage drawing path. Source bytes stay intact; missing/corrupt files return
no preview. All 17 preview/package tests pass on Catalyst and iOS Simulator,
including file dimensions and EXIF rotation. See native
docs/native-trace-preview-pixels.md; peak import/device measurements remain open.
Replaced the obsolete FEATURES mockup checklist with a web capability matrix
linked to regression tests, source and validation reports. Marked the old
comparison as historical, refreshed README package import/export coverage and
removed its fixed catalog count. Inspected the linked test scopes and verified
all 44 matrix file targets; git diff --check passed. No runtime tests were rerun
for documentation-only edits. Contributor/release templates and device gates remain open.
Added contributor setup/testing guidance, a minimal-fixture bug template, a PR template and a release evidence checklist. Commands and browser isolation match package scripts/Playwright/CI; the release checklist links the existing native, Storage migration and billing gates rather than asserting they are complete. README links both guides. Local links/frontmatter and diff checks passed; no application tests were repeated for documentation-only changes.
Verified the existing native root README against shared schemes, bundle IDs and platform minimums. Updated the linked native handoff with current resume guidance, marked stale pause instructions historical, and qualified old test counts as dated checkpoints. Setup documentation is present; native licensing/publication, final release branding and device gates remain open.
Added a generated manifest for 189 catalog entries and 93 mapped GLBs, including
default centimetre dimensions, web representation, asset hashes and embedded
metadata. CI checks drift through npm run catalog:check. Provenance is explicitly
unverified; source/license curation, product measurements, native support and
unmapped/texture assets remain open. See inventory scope.
All 93 mapped GLBs exactly match official Kenney archive members (29 Furniture Kit, 64 Nature Kit). Added durable archive/member/model hashes and local CC0 notices; inventory generation validates source evidence and license-notice hashes. See provenance scope. Unmapped assets, textures, product dimensions and native representation remain separate curation work.
Expanded source-byte verification and CI inventory checks from 93 mapped models to all 204 bundled GLBs (140 Furniture Kit, 64 Nature Kit). All match official archive members. The 111 unmapped assets are identified by empty catalogIds; an altered unmapped-asset provenance hash is rejected. Textures, product measurements and native representation remain separate work.
The complete FloorPlanTests target passed at native b6ee323 on Catalyst and
iPhone 17 Pro / iOS 26.5 Simulator: 248 executed, two optional external-worker
checks skipped, zero failures on each platform. Suite durations were 65.894 s
and 55.705 s respectively. See native docs/native-trace-integration-baseline.md.
UI dismissal, physical-device and release gates remain open.
Catalog inventory now covers all 20 bundled material textures with hashes, material IDs and source asset IDs from existing credits. Each has one current material mapping. CI rejects missing/duplicate credit records and missing mapped files. Attribution remains documented locally, not independently source-byte verified; conversion/source comparison and physical scale remain open.
Runtime texture loading and the inventory now share a typed filename module, replacing formatting-dependent source scraping. Existing 20 material/file associations and the generated manifest are unchanged. Catalog check, asset regression, Svelte check (zero errors/warnings) and production build passed.
Fixed permanently pending texture IDs after image-load errors. Wall and floor loaders now permit a later draw to retry after a 30-second cooldown, avoiding per-frame requests; success wakes rendering and reuses the cached image. Both regression cases failed before the fix; all three focused recovery/asset tests, zero-warning Svelte check, catalog check and production build passed. See validation.
Six production-browser cases pass across Chromium, Firefox and WebKit. Wall and floor texture requests recover after an injected abort; cooldown redraws do not add requests, and fulfilled retries update canvas pixels without further input. The test advances Date.now past the cooldown; physical-device/outage qualification remains separate. See report.
All 12 wall/floor × 2D/3D recovery cases pass across Chromium, Firefox and WebKit. 3D checks settle WebGL drawing before releasing the retry and view floors from above so their changed surface is visible. Shared texture callbacks rebuild the 3D scene without further input; no additional production change was needed. See report.
Failed GLB downloads no longer stay cached until page reload. Later requests can retry after a 30-second cooldown while concurrent callers still share one load and cloned resources remain independent. Existing fallback instances stay intact until rebuilt. The new regression failed before the change; all 14 focused model/texture tests, zero-warning Svelte checks, catalog checks and production build pass. See report.
npm test passed all 931 tests in 86 files against faaa9cd, zero failures,
21.14 seconds. The source already passed zero-warning Svelte checks and production
build. Focused texture browser checks remain separately reported; no new full
browser-suite or physical-device result is claimed.
Native 8e2605d now prepares ImageIO thumbnails directly from encoded Files/Photos
input and preserves PNG/JPEG attachment bytes instead of full-resolution UIKit
JPEG re-encoding. Other formats retain conversion. All 22 targeted preview/import
and package tests passed on Catalyst and iPhone 17 Pro / iOS 26.5 Simulator,
including PNG/JPEG package round trips. Original metadata/transparency and source
resolution are preserved; larger package sizes, other-format conversion and
physical-device peak memory remain open. See native
docs/native-trace-preview-pixels.md. No new picker UI qualification is claimed.
Added typed reactive translations and a persisted Settings language selector,
adapting the relevant community PR #15 strings. Language changes preserve dialog
identity/focus and initialize after hydration. Four unit tests, zero-warning
Svelte check, production build and all three browser-engine cases passed.
The remaining interface, nested settings panels, Portuguese review and device
coverage are still open; see docs/reviews/2026-09-10-localization.md.
Translated floor elevations/slab thickness, wall snapping, OpenAI provider settings and the shared model picker, plus both Settings entry points. Floor names, model IDs and numeric measurement inputs stay unchanged. Five unit tests, zero-warning Svelte check and production build pass. Language switching and Portuguese floor/provider interactions are verified across all three engines, with native model suggestions committed before the next button action. Broader localization and diagnostic-message translation remain open. See the localization record.
Translated the welcome flow, quick tour, template display labels, import-error controls and contextual tips. Template factories and generated data retain their original identifiers/names. Five localization unit tests, zero-warning Svelte check and production build pass. Six Portuguese browser cases pass across all three engines at 390 × 900, including bad-import recovery, template export and tour dismissal persistence. Library/package dialogs, wider editor translation and fluent-speaker/device review remain open. See the localization record.
Translated the library shell, counts/dates, template picker, action menus and rename/delete dialogs. Menu typeahead uses visible Portuguese labels; project content and action identifiers remain unchanged. Header controls wrap on narrow screens. Five localization unit tests, zero-warning Svelte check and production build pass. Three 390-pixel browser cases pass across all engines, covering menu keyboard behavior, geometry-preserving rename, deletion cancellation and template labels. Restore/package dialogs and broader editor translation remain open.
Translated library-restore and package-import controls, previews, confirmation, completion and dialog-owned recovery messages. Five unit tests, zero-warning Svelte check and production build pass. Six browser cases pass across all engines at 390 × 900, verifying unchanged library data before confirmation, byte-identical original downloads and one imported/restored copy. Service diagnostics/warnings and broader editor translation remain open; see the localization record.
Translated alignment/distribution, selection-toolbar actions, contextual menus and Undo History controls. Saved history descriptions and wider editor UI remain open. Five localization unit tests, zero-warning Svelte check and production build pass. Twelve Portuguese alignment cases pass at desktop/phone widths across all three engines, verifying scaled/rotated geometry, locked items, Undo and Redo. See the localization record for the exact coverage limits.
Translated command/action names, categories, search labels and keyboard hints; search accepts queries without Portuguese accents. Five localization unit tests, zero-warning Svelte check and production build pass. Three browser cases pass across all engines, verifying searches, Settings/tool/grid execution after modal closure and focus restoration. Catalog names/categories and wider editor controls remain open. See the localization record.
All 936 web unit tests across 87 files pass at 31e263d (21.68 seconds, exit 0).
The existing English palette/modal-field flow passes in all three engines at
1440 and 390 pixels (six cases, 46.4 seconds). This supplements the focused
Portuguese checks; it does not close the remaining localization or release gates.
Translated the version panel, entry points, time labels, confirmations and panel-owned errors. Snapshot descriptions/data remain unchanged. Five localization unit tests, zero-warning Svelte check and production build pass. Three phone-width browser cases pass across all engines, verifying cancellation preserves history and confirmed restore recovers the selected project name and wall height. Service diagnostics and wider editor translation remain open.
Translated summary labels, categories and controls, preserving room names and
measurement logic. Five localization unit tests, zero-warning Svelte check and
production build pass. Three phone-width browser cases pass across all engines,
comparing measurements across languages and retaining unknown room categories.
The older 931-test checkpoint above is explicitly historical; the latest full
unit baseline remains 936 tests at 31e263d. Wider localization remains open.
Translated print entry/controls, warnings and fallback errors. Paper IDs, scale values and project content are unchanged. Five localization unit tests, zero-warning Svelte check and production build pass. Three phone-width browser cases pass across all engines, verifying overflow/fit behavior, A4 portrait proportions and PDF download. Printed sheet/schedule text and physical print qualification remain open.
Translated sheet captions/instructions and the PDF schedule heading with explicit locale input. Layout math, calibration and project content remain unchanged. Fifteen print-scale/localization unit tests, zero-warning Svelte check and production build pass. Three browser cases pass across all engines, inspecting canvas labels and the downloaded PDF heading alongside scale/fit checks. Physical print, broader canvas text and general export localization remain open.
PDF/PNG notices now retain typed message keys and translate at display time. Nine unit tests, zero-warning Svelte check and production build pass. Three Portuguese browser cases pass across all engines, checking real 2D PNG export from 3D mode and failure feedback from menu/palette with no extra download. Export-menu labels, deployment notices and wider localization remain open.
Translated update, loading-failure and save-before-leaving notices with reactive keys. Twenty-one targeted unit tests, zero-warning Svelte check and production build pass. Three Portuguese browser cases pass across all engines using an isolated server, verifying blocked navigation on save failure, JSON recovery and successful retry to the intended destination. No live deployment changed; wider localization remains open.
All 938 unit tests across 87 files pass at e12cff4 (11.98 seconds, exit 0).
Three English reload-safety browser cases also pass across all engines (26.6
seconds), covering failed-save protection, JSON recovery and retry navigation.
Remaining localization, device and release gates stay open.
Corrected one-result model-discovery wording in English and Portuguese. Five localization unit tests, zero-warning Svelte check, production build and all three provider browser cases pass. Broader localization and fluent-speaker review remain open.
Translated export/import actions and project-package help. Five localization unit tests, zero-warning Svelte check, production build and six browser cases pass across all three engines, including original-name JSON preservation and PNG export/failure feedback. Remaining editor localization and device/release gates stay open.
Desktop and compact view controls now translate labels, tooltips and state text. Five localization unit tests, zero-warning check, production build and nine browser cases pass across all engines, including 390px/1440px controls and language-switching area-summary preservation. Other editor localization, fluent review and physical-device/release requirements remain open.
Translated floor controls without changing stored names or seed behavior. Fifteen focused unit tests, zero-warning check, production build and six Portuguese browser cases pass, covering both toolbar widths, floor framing/camera restoration and exported data preservation. Broader localization and device/release gates remain open.
Translated save-state/recovery controls and made elapsed-time tooltips react to language changes immediately. Five localization unit tests, zero-warning check, production build and six browser cases pass, including storage-failure backup, retry, persisted edits and deployment recovery. Raw storage diagnostics, remaining editor localization and device/release gates stay open.
Layers item-list labels are translated. Five unit tests, zero-warning check, production build and six browser cases pass, using the L shortcut at both widths to verify visibility/selection and exported-data preservation. Catalog names remain untranslated; the canvas visibility popover was translated in the follow-up below.
Reproduced issue, subsequently fixed in browser checks (see below): at 390px, clicking the canvas “Layers” control is blocked by the floating zoom toolbar (Fit selection intercepts the click in Chromium). The item-list toggle is desktop-only; its keyboard shortcut works on the narrow layout but does not establish touch access. Fix control overlap and make item-list access clear on touch devices; verify real clicks without force or DOM dispatch. See the localization record for failed/successful run evidence.
The canvas status/actions strip scrolls horizontally above the zoom controls on phones; its visibility popover opens above both strips. Added a Layers entry to the compact toolbar menu to open/close the item list without a keyboard. Six browser cases pass across Chromium, Firefox and WebKit at 1440px/390px (60.0s). The narrow cases use normal pointer clicks to open/close the visibility popover, toggle walls and open the item list, then verify selection and exported floor data. Production build and zero-warning Svelte check pass. Physical touch devices, other narrow viewport combinations and general toolbar localization remain open.
Translated visibility checkbox labels, lower-floor text and the canvas trigger. Five unit tests, zero-warning check, production build and six browser cases pass (57.5 seconds), including narrow-layout pointer access, room-label toggles and the disabled no-lower-floor option. Remaining canvas/editor strings and physical-device qualification stay open.
Translated canvas zoom/reset/fit labels and shortcut hints without changing camera math or limits. Five unit tests, zero-warning check, production build and six Portuguese large-fit browser cases pass (36.3 seconds) across all engines and both widths, covering small zoom values, extents and zoom/fit restoration. Remaining canvas/editor localization and physical-device gates stay open.
Translated canvas display toggles and exposed their on/off state with aria-pressed. Five unit tests, zero-warning check, production build and the final nine-case focused browser run pass, including pointer toggles and unchanged exported settings. An initial WebKit phone-width grid click did not toggle; isolated and full reruns passed unchanged. Keep this intermittent observation in view during broader WebKit/touch qualification; its cause is not established. Remaining translation and physical-device gates stay open.
Translated project navigation/rename, undo/redo labels and import/export error headings. Five unit tests, zero-warning check, production build and six browser cases pass (36.4 seconds), verifying preserved names, backup/retry and guarded back-navigation. Raw diagnostics, remaining editor panels and device/release requirements stay open.
Translated empty-plan/elevation-picking guidance and inline editor labels. Five unit tests, zero-warning check, production build and six browser cases pass (30.5 seconds), including drawing a real wall and canceling elevation picking at both widths. Remaining editor strings and physical-device gates remain open.
Expanded the existing annotation keyboard test to English/Portuguese and verified
save/reopen/download equality for literal-brace text. All six desktop browser
cases pass across three engines (24.4 seconds), including field focus and context
menu bounds. Room/dimension inline editing and phone qualification remain separate
work. Runtime source is unchanged from 102537f.
Translated panel tabs, primary tool/help text and import entry captions. Five unit tests, zero-warning check, production build and six browser cases pass (26.1 seconds), selecting the translated wall tool through desktop/phone UI and verifying its exported wall. Catalogs, remaining room/object content, import dialogs and physical-device qualification stay open.
Translated all door/window choice labels and descriptions reactively, retaining semantic types, dimensions and drag payloads. Five unit tests, zero-warning check, production build and three desktop browser cases pass (10.3 seconds), checking all labels plus representative door/window placement and exported types/widths. Remaining room/object catalogs, dialogs and device qualification stay open.
Translated room shapes/templates through display-only maps and localized item counts. Five unit tests, zero-warning check, production build and six browser cases pass (36.5 seconds), covering desktop/phone placement, original geometry and furniture IDs, and grouped undo. Object catalogs, dialogs, remaining editor text and device/release requirements remain open.
Translated search/results/favorites controls and accessible names while preserving search and stored IDs. Five unit tests, zero-warning check, production build and 12 bilingual browser cases pass (1.2 minutes), including keyboard focus, literal search text and independent favorite/placement actions. Item/category names, remaining dialogs/editor panels and physical-device requirements remain open.
Translated category filters/preview labels while preserving identifiers. Search accepts original/translated categories without requiring accents. Five unit tests, zero-warning check, production build and three browser cases pass (36.3 seconds), including category isolation and original recent-item IDs. Individual object names, remaining editor panels/dialogs and device/release gates remain open.
Translated import-option labels/help and confirmation/cancel actions. Five unit tests, zero-warning check, production build and six Portuguese browser cases pass (39.4 seconds), verifying edited-option cancellation leaves stored projects intact and restores dialog focus at both widths. Remaining diagnostics/editor strings and physical-device/release gates stay open.
Six browser cases pass across all engines at both widths (43.6 seconds), confirming
prepared RoomPlan imports in English/Portuguese and comparing exported walls,
openings and linkage while retaining original names and prepared-option defaults.
Runtime source remains 9fd5922. Raw-scan/device qualification and remaining
localization/release requirements remain open.
Translated built-in symbols, category/place hints and upload captions, sharing display labels with Layers. Five unit tests, zero-warning check, production build and three browser cases pass (41.9 seconds), retaining person ID/size and undo. Custom names stay unchanged; individual furniture names, remaining editor panels and physical-device/release requirements remain open.
Added translated inline read/decode/size errors and corrected the supported-format caption. Five unit tests, zero-warning check, production build and three browser cases pass (41.8 seconds), verifying failed uploads preserve prior data and valid PNG retry preserves original bytes/names. Remaining editor localization and physical-device/release gates stay open.
Added asynchronous read-error fault injection to upload recovery coverage. Three
browser cases pass (33.4 seconds), retaining prior data and verifying successful
retry after restoring the native reader. Runtime source remains 950fc36; real
hardware/device failure qualification and remaining NEXT requirements stay open.
Translated door/window Properties controls while retaining numeric handlers and stored type IDs. Production check/build and five localization unit tests pass; three desktop browser cases pass (12.2 seconds), verifying width, sill height, type, hinge and opening direction edits through JSON export with host walls unchanged. Other Properties sections and physical-device/release gates remain open.
Translated furniture editing controls and finish display names, retaining stored IDs and catalog names. Production check/build, five localization unit tests and three desktop browser cases pass (25.8 seconds). Export comparisons verify appearance/transform edits, reset behavior and unchanged neighboring furniture. Other Properties sections, 3D/device qualification and remaining NEXT gates stay open.
Translated room controls and type/color/material choices while retaining original preset names and material IDs. Production check/build and five localization unit tests pass. Three desktop browser cases pass (35.6 seconds), checking literal names, category/material/color edits, reversible floor openings and unchanged geometry. Remaining Properties sections and broader usability/device/release gates stay open.
Translated symbol Properties controls and reused built-in display names. Check, build and five localization unit tests pass; three browser cases pass (50.1 seconds) for numeric/opacity/lock edits and delete/undo data preservation. An initial WebKit placement undo assertion failed before new edits, then passed unchanged in isolated and full reruns; its intermittent cause remains unresolved. See the localization record. Remaining Properties and device/release gates stay open.
Reproduced the symbol Undo failure on WebKit run four. Trace evidence shows autosave changing the status label during the click and shifting toolbar targets. Reserved space for the widest translated status; undo/history logic is unchanged. A new position-stability assertion passes across all engines with save/symbol workflows (six cases), followed by five consecutive WebKit passes. Production check/build pass. This addresses the observed save-label shift; other intermittent canvas observations and remaining NEXT requirements remain open.
Translated stair editing controls while retaining layout IDs and handlers. Check, build and five localization unit tests pass; three desktop browser cases pass (18.5 seconds), verifying dimensions, riser count, layout, rotation, direction and undo through exported data. Remaining Properties and device/release gates stay open.
Translated column labels/color presets and corrected a reproduced accessible-name issue caused by a label wrapping both shape buttons. Shape controls now use a named group and pressed states. Check/build, five localization unit tests and three browser cases pass (29.4 seconds), retaining dimensions/rotation/color across shape changes. Other Properties/accessibility/device/release work remains open.
Reproduced missing accessible names for the first hinge/direction buttons. Replaced wrapping form labels with named groups for door hinge/opening and stair direction; added selected-state semantics. Production check/build and six browser cases pass (51.2 seconds), including edited-data export and stair undo. Remaining accessibility, localization and device/release requirements stay open.
Translated annotation Properties labels. Production check/build and five localization unit tests pass. Six bilingual browser cases pass (53.2 seconds), checking focus, reopen, literal multiline text and font/rotation/position edits through exported data. Remaining Properties/device/release gates stay open.
Translated image adjustment, lock, calibration-entry and removal labels. Check, build and five localization unit tests pass; three browser cases pass (31.1 seconds), verifying image byte/position preservation, adjustment values and removal/undo. Calibration execution and remaining Properties/device/release gates stay open.
Translated wall length/thickness, fixed-endpoint choices, related help and minimum length validation. Check/build, five localization unit tests and three browser cases pass (33.0 seconds), retaining connected geometry/opening values and invalid input undo behavior. Other wall controls/diagnostics and broader device/release requirements remain open.
Translated endpoint-height controls, clipping warning, equalize and reverse actions. Check/build, five localization unit tests and three browser cases pass (40.0 seconds), verifying saved opening dimensions, swapped endpoints/heights and equalized heights. Remaining wall controls/diagnostics and broader device/release scope stay open.
Translated curve states and elevation entry; added a named curve toggle with pressed state. Check/build, five localization unit tests and three browser cases pass (47.5 seconds), verifying saved curve geometry, unchanged openings and straight wall restoration. Wall materials/diagnostics and broader device/release gates stay open.
Translated wall color/texture choices and side controls while retaining material IDs. Check/build, five localization unit tests and three browser cases pass (53.2 seconds), verifying independent interior/exterior finish values and texture removal without losing other wall/opening data. Diagnostics, child detail panels and broader device/release requirements stay open.
Translated notes/cost, construction-material and room-use/ceiling metadata labels, retaining identifiers and values. Check/build, five localization unit tests and six browser cases pass (56.1 seconds), verifying literal notes, cost precision, room-use ID and ceiling override. Photos/diagnostics and broader native/device/release requirements stay open.
Six English/Portuguese Swift-return fixture cases pass (54.6 seconds), checking translated construction material, native room-use/height and cost export. Fixed a test-loading dependency by validating exported photos with the browser decoder; six desktop/compact photo integration cases pass (1.6 minutes). Runtime unchanged; new physical/native runs and remaining photo/localization/release gates stay open.
Translated item-photo add/download/remove controls and guidance. Check/build, five localization unit tests and three browser cases pass (33.8 seconds), preserving small-PNG bytes and retained assets while verifying remove/undo references. Retained file management, diagnostics and broader device/release requirements stay open.
Translated retained-file controls and confirmation text, preserving source filenames. Check/build, five localization unit tests and three browser cases pass (41.9 seconds), verifying reuse, cancellation retaining bytes and confirmed deletion. Diagnostics and broader device/release requirements remain open.
Translated known detail-panel notices at render time, preserving unknown service errors. Check/build, five localization unit tests and six browser cases pass (1.0 minute), verifying invalid-cost recovery, attachment notices and rejection of used-file deletion. Broader diagnostics/device/release requirements remain open.
Reproduced Infinity calibration producing null scale in JSON export. Added finite, positive distance/result checks and translated the prompt. Check/build, five localization unit tests and three browser cases pass (46.8 seconds), checking invalid/cancelled input preservation and valid scale doubling. Physical calibration and remaining device/release requirements stay open.
Reproduced a stale calibration prompt after Escape. The canvas Escape handler now clears calibration mode and points. Check/build and three browser cases pass (51.8 seconds), preserving image data and allowing fresh calibration afterward. Other cancellation/device/release requirements remain open.
Reproduced calibration surviving floor addition. The shared floor-context reset now discards calibration mode and points. All 11 floor unit tests pass, including switch/add/remove and floor-changing undo/redo. Production check/build and three browser cases pass (59.2 seconds), preserving both floors’ background images and allowing fresh calibration. Physical-device and remaining release work stay open.
Translated wall cycling, wall counter and interaction guidance, and named the elevation canvas for accessibility. Check/build, five localization unit tests and three browser cases pass (51.3 seconds), covering navigation, wraparound and Escape with unchanged exported floor data. Opening drag/device qualification and the remaining localization/release work remain open.
Reproduced a moved window remaining changed after Undo when Escape interrupted its elevation drag. View teardown now closes the pending drag’s undo group. Production check/build and three browser cases pass (41.3 seconds), checking position/sill changes, retained dimensions and exact whole-floor Undo/Redo. Other gesture exits, physical-device and remaining release work stay open.
Translated shortcut help and copied reference together, plus the 3D loading message. Production check/build, five localization unit tests and three browser cases pass (22.6 seconds), verifying help entry/dismissal and the translated clipboard payload. OS clipboard permissions, remaining UI/device/release work stay open.
Translated retry/backup, return-to-projects, loading/importing and capture-error shell controls. Check/build, five localization unit tests and three browser cases pass (22.3 seconds). Injected migration failure preserves exact backup bytes; retry opens the original project without duplication or saved-data changes. Service diagnostics, physical-device recovery and other release work remain open.
Translated known capture-import errors with code/status parameters while retaining unexpected diagnostics. Check/build, five localization unit tests and three browser cases pass (35.8 seconds), each covering six routed failure paths and dismissal back to an available editor. Live handoff, physical-device and remaining release requirements stay open.
Three browser cases pass at 1700d1e (29.6 seconds), confirming a routed
multi-floor capture remains exportable after a project-write quota failure. Retry
saves one project with exact backup floor data, and reload avoids re-downloading
the capture. No runtime changes were needed. Live cloud/device and remaining
release qualification stay open.
Translated tools/layers/history controls and exposed expanded state. Check/build and five localization unit tests pass. All 30 affected browser cases pass across three engines in split runs, including desktop/narrow panel use, catalog placement and bilingual RoomPlan preservation. See the localization record for run details. Physical touch and remaining UI/release requirements stay open.
Translated primary viewer controls and camera/walkthrough guidance. Check/build, five localization unit tests and three browser cases pass (47.0 seconds), covering view toggles, camera placement cancellation, PNG download and unchanged project floors/settings. Remaining viewer panels, physical-device and release requirements stay open.
Translated lighting labels and presets, exposing panel/preset states. Check/build, five localization unit tests and three browser cases pass (57.5 seconds), covering all preset values, keyboard ambient adjustment, selection reset and unchanged exported project data. Render-quality, remaining viewer UI and device/release requirements stay open.
Three Portuguese browser cases pass at 9cf4ec2 (35.0 seconds), observing
rendered camera movement, isolated eye-height changes and Top-Down exit under
simulated mouse-lock denial. Three existing English input-reset cases also pass
(39.2 seconds). No runtime changes were needed. Physical input/performance and
remaining release requirements stay open.
Translated preview, movement, camera adjustment and capture controls, naming movement arrows accessibly. Check/build, five localization unit tests and three browser cases pass (58.6 seconds), covering preview rendering, control use, 1920×1080 PNG capture and context cleanup after reposition/close. AI panel, physical-device and remaining release requirements stay open.
Translated AI panel controls/disclosure and style/lighting/mood display choices while preserving model IDs and provider prompt values. Check/build, five localization unit tests and three browser cases pass (37.8 seconds), with literal extra instructions retained and no external requests. Render execution, remaining diagnostics/device and release qualification stay open.
Six existing English local-provider browser cases pass at edd0846 across
three engines and desktop/narrow layouts (2.1 minutes), covering image output,
byte-identical download, failure/cancellation and settings cleanup. No runtime
changes or live AI requests were needed. Portuguese execution, device and
remaining release qualification stay open.
Translated placement controls/guidance and reused category labels, with accessible category selection states. Check/build, five localization unit tests and three browser cases pass (46.1 seconds), verifying catalog identity and exact floor Undo/Redo after placement. Item names, remaining viewer text and device/release requirements stay open.
Reproduced missing feedback on clipboard denial. Copy now catches failure, reports translated status, supports retry and ignores late results after dialog reopen. Check/build, five localization unit tests and six browser cases pass (28.6 seconds), including delayed completion and no page errors. Actual OS permissions and remaining device/release requirements stay open.
Translated export labels/status and stacked-floor elevation. Check/build, five localization unit tests and six Portuguese browser cases pass (52.0 seconds), covering deterministic scene downloads, geometry/height/scope and backup heights at desktop/narrow widths. External rendering, failure branches and remaining device/release qualification stay open.
Translated five known app-owned AI error messages while preserving unknown/raw diagnostics. Check/build, five localization unit tests and three browser cases pass (35.8 seconds), including missing-key recovery without network requests. Other failure branches, physical-device and release requirements stay open.
Translated object/geometry counts with singular/plural forms, selection and zoom text. Check/build, five localization unit tests and twelve browser cases pass (1.3 minutes), covering desktop/narrow layer counts and welcome/template recovery with retained project data. Remaining UI/device/release requirements stay open.
Translated duplicate/delete, swing-flip and midpoint-split controls. Check/build, five localization unit tests and six Portuguese browser cases pass (56.3 seconds), covering mixed/symbol duplication and deletion with IDs, wall references and Undo/Redo preservation. Swing/split, device and remaining release qualification stay open.
Three Portuguese browser cases pass at 2c620e9 (24.2 seconds), checking saved
swing changes, midpoint endpoints and sloped heights, opening dimensions/centers
and exact Undo restoration. No runtime changes were needed. Openings spanning
the split point, curved splitting and device/release requirements stay open.
Reproduced door/window crossing splits mutating their owning wall. Since an opening has one wall reference, splitting through its interior now returns null before taking an undo snapshot. Splits exactly at opening edges remain allowed. All three canvas split entry points use a shared wrapper with an English/Portuguese dismissible explanation. Check/build, 943 unit tests and nine browser cases pass; browser checks cover crossing door/window preservation, notice dismissal and existing safe split/swing/Undo behavior. Curved splitting, physical gestures and the remaining native/release backlog stay open.
Reproduced the context menu opening without focused actions. Menus now focus the first enabled action, wrap with Up/Down, support Home/End and restore prior focus on Escape or activation. Tab closes the menu; menu key events no longer bubble to editor shortcuts. Removed unused action imports and added visible focus styling. Check/build and six English/Portuguese Chromium/Firefox/WebKit cases pass (52.4 seconds), including keyboard activation, Escape, viewport bounds and annotation save/reload/edit preservation. These tests open the menu with a pointer; native keyboard context-menu invocation and physical accessibility checks remain open.
Shift+F10 and the Menu key now open a context menu from the focused canvas. The menu targets the selected supported element (or canvas for multi-selection), with its anchor clamped to the canvas. It does not depend on pointer position or place measurement points. A regression at 211fb0f could not find the wall menu after Shift+F10; the fixed checks pass.
Check/build and 12 Chromium/Firefox/WebKit cases pass (1.4 minutes), covering selected-wall keyboard invocation with both keys, opening-safe split activation, Delete isolation and exact floor preservation, plus English/Portuguese canvas menus, keyboard activation, Escape/Tab dismissal and annotation persistence. Furniture/opening/room keyboard target branches and physical assistive-technology qualification remain to be exercised; broader NEXT scope stays open.
Reproduced Properties leaving keyboard focus on the canvas. The explicit action now waits for the selected panel to render and focuses its first enabled control, with guards against teardown or a changed selection. Twelve Portuguese browser cases pass across Chromium/Firefox/WebKit at 1440px/390px (1.1 minutes), checking selected door/window menu targeting, focus transfer, width edits and exact floor Undo restoration. Check/build pass with zero Svelte diagnostics. Furniture/room menu actions and physical assistive-technology checks remain open.
Reproduced Escape from inline room rename leaving focus outside the canvas. Enter/Escape now prevent their default action and restore canvas focus after commit/cancel. Blur still allows focus to move to the requested control. Check/build and six Portuguese browser cases across Chromium/Firefox/WebKit at 1440px/390px pass (48.5 seconds). They cover room-menu keyboard targeting, rename focus, cancellation with exact floor preservation, ordinary blur focus transfer, name-only updates and exact Undo restoration. Other room-menu actions and physical accessibility remain open.
Change Floor Texture now selects the room and focuses its current material (or the first available choice). Material choices have a named group and expose their selected state. Check/build and six Portuguese browser cases pass across Chromium/Firefox/WebKit at 1440px/390px (20.8 seconds), covering menu focus, keyboard material selection, material-only changes and exact Undo alongside the existing rename cases. Room reset/delete keyboard flows and physical accessibility remain open.