houseplan-card/docs/design/505-summary-panel/ACCEPTANCE.md

#505 — author evidence and visual comparison

Issue: #505. Spec: 505-summary-panel-design-parity.md. This is implementation evidence, not the independent S7 verdict.

Reproduce the paired visual check

After npm ci and npm run bundle:sync:

node demo/capture_summary_panel_505.mjs --probe-only
node demo/capture_summary_panel_505.mjs --output artifacts/summary-panel-505

The fixture instructions explain the pinned official HA wheel, isolated loopback server, cache integrity and optional offline wheel. No production HA instance is accessed. The genuine hp-dialog → ha-dialog → wa-dialog → dialog chain is exercised; only the HA app's private loader is exposed, not a replacement dialog implementation.

Author run on Windows/Chromium, product commit 05778a47 (pre-rebase a038149c, recorded in the original capture): source fingerprint 23ed457d01a2622aae7bb6937116f20f0044b8216b723388ed3dcf85dacd4a1e. Rebase onto dev 2e387ac1 preserved identical frontend source and bundle bytes; both unrelated #495 and #505 changelog entries were retained. gate:small was rerun successfully after the rebase. The full capture completed all five paired panel contexts and native/real-HA settings in desktop, 320/390px, light/dark and enlarged-text cases. Every settings surface/editor/footer had no horizontal overflow and a visible footer; no page errors, external requests or WebSockets occurred. The separate authentic probe measured small=320, medium=580, inherited medium=920, mobile=390 fullscreen. The product's first opening was 920px in both native and genuine HA.

Long German source fields were additionally re-captured using --only for the edge-*-text-200-long-de* scenarios after replacing an incorrect assumption about scroll ownership in the capture. Real HA scrolls its shadow .body, native scrolls .summary-editor. The report records both scroll owners and the actual visible band between header and footer; the source name and entity ID were visually inspected in the corrected start/end frames. Product code was not changed for this capture correction.

These are diagnostic frames, not canonical golden/docs baselines. The main matrix and corrected source frames were personally inspected, including desktop light/dark, bottom portrait, kiosk, mobile View, native/HA settings, and enlarged German text. The portable command produces the same comparisons for the independent reviewer. A completed capture alone is not visual approval.

Comparison result and bounded adaptations

Surface	Result
Header/kiosk control	Designer outline gear on the left, sidebar on the right; only the right half is active. Normal control is last in View actions, kiosk control floats above the panel.
Panel	Separate header, muted scroller, white/theme block cards, header dividers and left-label/right-bold-value rows. Compact content height, 12px radius and stage insets; right/bottom anchoring preserved.
Settings	General/Blocks cards, block count, grip/arrows/name/eye header, scope row, label/source/remove value rows, dashed add actions and separate destructive block footer match the composition.
Responsive form	Wide 920px shell instead of the reference's 780px accommodates readable text and 44px controls. Narrow fields stack. HA retains its own fullscreen rule and close-button placement; native keeps its normal modal close position.
Themes and text	HA theme colors replace hard-coded cyan/white. Text is larger than the miniature reference and follows root font size; HA's 14px root yields a 19.25px title, native 16px root yields 22px. At root32px the title is 44px and wraps without clipping.
Data and actions	Real HA-formatted values, stable source picker, permissions and shared/local ownership retained. Save stays visible but disabled when unchanged, unlike the demo's omitted Save. No size controls in this dialog; stored scales preserved.

The 800×500 effective-viewport scenario is a layout proxy for 200% zoom, not an actual browser-zoom claim. Separate root32px scenarios exercise enlarged text. Different fake HA chrome, background plan and empty non-summary glyph stubs are outside comparison scope. No pixel-identical rendering is claimed.

Executed gates

• npm run gate:small — PASS: frontend units, typecheck/build, no-new-any, bundle parity and budget. The initial View graph is 298558B gzip, only 1B above the pre-task build; all added summary shell CSS is lazy. No budget or baseline was increased.
• node demo/smoke_summary_panel.mjs — PASS, retaining live values, lost-ACK, source identity, 10k-entity/200-row/multi-card scenarios.
• node demo/smoke_summary_panel_polish.mjs — PASS, actual frame sampling, lifecycle boundaries, native geometry and persisted-setting combinations.
• node demo/smoke_kiosk.mjs, node demo/smoke_houseplan_panel.mjs, node demo/smoke_dialog_footer_width.mjs, node demo/smoke_nav_persist.mjs, node demo/smoke_lazy_editor_chunk.mjs, node demo/smoke_warm_dialogs.mjs — PASS for neighbouring View/navigation/dialog contracts.
• node scripts/check-docs.mjs --screenshots=warn — PASS with the expected stale canonical screenshot warning. Strict capture/acceptance belongs to the beta candidate; no Windows baseline acceptance was performed.
• git diff --check, node scripts/check-inputs.mjs --coverage, node scripts/mutation-gate.mjs --check — PASS.

The broader smoke-select list includes unrelated symbols reached through the large card renderer. Geometry, drawing, light, PDF, backend and public models are unchanged; their full release suites and performance/golden captures were not rerun for this task. This is not a release-quality assertion for a beta.

AC and failure witnesses

All registered witnesses below were run with node scripts/mutation-gate.mjs --id=<id> on the implementation commit: each clean guard passed, each mutated guard failed, and the runner reported поймано 1 из 1 with exit0. No stale-bundle bypass was used.

AC	Evidence	What detects a regression
AC1–2	Focused smoke geometry/content + paired visual inspection above	Direct actual DOM/bounds/glyph/value comparisons; these are presentation AC, not guards.
AC3	Focused smoke samples opacity, 18px offset, 190ms duration, one retained inert node, reversal and removal	summary-hide-unmounts-before-animation is caught.
AC4	test/summary-panel-presentation.test.mjs plus focused browser lifecycle matrix	summary-animation-stale-completion-unguarded, summary-animation-reset-keeps-presentation, summary-animation-ignores-reduced-motion are caught. Actual browser checks include backgrounding, resize, eligibility, editors and identity changes.
AC5	Native width/actual-control bounds smoke plus authentic HA measured and inspected frames	summary-dialog-loses-wide-shell is caught; it removes wide from the real form rather than changing a test fixture.
AC6	Existing/focused summary smokes and paired settings frames	Actual CRUD/reorder/visibility/source/confirmation results and direct layout comparison.
AC7	Summary runtime units and focused persisted-storage comparisons, admin and local-only	Read-through confirms existing scale load/save authority is untouched; the existing summary-local-scale-authority-disabled witness remains applicable (not rerun here).
AC8	All local/mobile pairs through Save/Cancel/X/Escape/reopen/reload; unchanged and busy form	summary-mobile-ignores-local-off is caught by real disabled/focus/reopen/reload assertions; prior checkbox value remains saved.
AC9	Existing summary runtime units + source picker/live/lost-ACK smoke	Existing #493 guard mechanisms read as unchanged; their registered witnesses remain applicable. This task does not claim a fresh negative run of every unchanged backend/permission guard.
AC10	Four locales, narrow geometry, actual targets, keyboard, camera identity; theme/large-text frames	Actual bounds/focus/value comparisons and frame inspection.
AC11	Gates/docs above; code scope inspection	No new eager implementation except header call order; stored schema and backend untouched.

Code-review r1 follow-up

The independent review accepted the implementation and recorded one Low L1: summary.block_visible had no remaining consumers. The follow-up removes only that unused key from all four dictionaries and rebuilds the bundle. No render, style, event, model or visible string is changed. The paired visual evidence above still describes the accepted implementation; it is not claimed as a new capture of the cleanup commit.

• rg -n 'summary\.block_visible' src demo test scripts — no matches.
• npm run gate:small after rebuilding, synchronizing and staging the complete generated trees — PASS. The preliminary attempts caught unsynchronized bundle manifests and untracked generated assets; both were corrected without changing any test or gate.
• node demo/smoke_summary_panel_polish.mjs on the fresh cleanup bundle — PASS.

Rollback is a revert and normal bundle rebuild. Nothing needs data migration. Independent review and pipeline merge remain required before S8; publication and issue closure belong to a later release request.