Skip to content

Latest commit

 

History

History
69 lines (48 loc) · 8.68 KB

design-qa.md

File metadata and controls

69 lines (48 loc) · 8.68 KB

Balanced Atelier — visual acceptance

final result: passed

Evidence and scope

  • Selected visual truth: docs/design/references/balanced-atelier.png (first displayed option; 1487 × 1058 pixels).
  • Real implementation: the existing Electron application, isolated synthetic project “골목의 오후”, description at 01:24 selected. No writer project or credentials used.
  • CSS viewport: 1440 × 1024, deviceScaleFactor 1. Implementation capture: output/playwright/atelier-review-1440.png, 1440 × 1024.
  • Comparison normalization: both images contained into 1440 × 1024 panels, separated by 24px. The reference includes an illustrated title bar; the implementation screenshot excludes the genuine native title bar. These frame offsets are not findings.
  • Full comparison: output/playwright/atelier-round1/atelier-comparison-full.png.
  • Focused table comparison: output/playwright/atelier-round1/atelier-comparison-detail.png.
  • Additional states: 800 and 1100px wide, 28px long Korean text, tools dialog, export menu, high contrast. Generated film still is original synthetic QA media, never a production image.

Findings — comparison round 1

  1. P2 · Notifications obscure persistent actions. Bottom-centered export toast overlaps undo/next controls, especially at 800 and 1100px. Evidence: output/playwright/atelier-round1/atelier-review-800.png, atelier-large-text-1100.png. Relocate it to a non-overlapping status region and expose its status accessibly.
  2. P2 · Excess table density drift. Redundant per-row durations add a second time line and show roughly seven rows where the selected design shows eight. Remove duplicate duration lines; retain the selected interval duration in the media context. Preserve 16px default script text.
  3. P3 · Long precise time values. Code review notes 52px inputs cannot display supported 125:43.250 values fully. Accommodate longer values and wrapping without widening every ordinary time field.

Required fidelity surfaces

  • Typography: Segoe UI with installed Malgun Gothic Korean fallback, not bundled Apple fonts. Actual script 16px default and 11–28px persisted control. Clear heading/body hierarchy; keyboard focus retained. Korean rasterization differs from generated mock; no fabricated font match claim.
  • Spacing/layout: 42/58 main workspace matches selected wide composition. Narrow layout stacks the media/context region above the script. Density finding above must be fixed; native minimum allows 800px content width (816px outer Windows frame in this environment).
  • Colors: white/warm neutral surfaces, deep teal #145f59, restrained separators, distinct approval/draft badges. Status is conveyed in text, not color alone.
  • Image quality: original wide synthetic alley shot plays as an actual MP4. object-fit: contain intentionally differs from the mock's portrait crop, preserving visual evidence important to audio-description writers. First premature screenshot showed buffering; waiting for a decoded non-seeking frame fixed the capture, not production code.
  • Copy/content: Korean labels, honest autosave/review state, separate HWPX all-script and SRT dialogue-only explanations. Extra “작업 도구” and next-unreviewed controls retain existing capabilities. No design-award claims or prompt language in UI.
  • Icons: licensed Microsoft Fluent regular icons, no Apple assets or custom icon approximations.

Comparison round 2 — accepted

  • Fix commit: a199786. Full comparison reopened together at output/playwright/atelier-comparison-full.png; focused table comparison reopened at output/playwright/atelier-comparison-detail.png.
  • Notifications are now normal-flow live status regions in the media context and the active modal, not a footer overlay. Native geometry checks verify no overlap with any footer action at 1440, 1100 and 800px. Evidence: output/playwright/atelier-responsive-800.png and atelier-responsive-1100.png.
  • Eight ordinary rows are fully visible at 1440 × 1024 with 16px text; an actual bounding-box assertion checks the eighth row inside the table viewport. Redundant time-cell durations are removed; selected duration remains in the media context.
  • 125:43.250 is fully visible in the expanding time field and wraps with the end time within the time column; input scrollWidth <= clientWidth is verified at 1100px. Evidence: output/playwright/atelier-precise-time-1100.png. This deliberately invalid fixture range also leaves approval disabled without saving it.
  • The five fidelity surfaces above were rechecked. No actionable P0/P1/P2 remains. Video playback controls are genuine Chromium controls; waiting for readiness and the native loading-control fade makes the paused frame reproducible. Saved selected-row state differs from the mock's pending-draft state, so the neutral “확인 전” badge is correct.
  • Remaining P3: Windows font rasterization differs from the image-generated type; a future writer field trial can guide typography preference. The supported text-size controls and clear keyboard focus are retained.
  • Stable handoff copies: docs/design/atelier/desktop.png, comparison-full.png, comparison-detail.png. Full transient test artifacts remain under ignored output/playwright.

Final integration review — additional blocking finding

P1 · Newly selected text can remain offscreen. Whole-branch review found that next-unreviewed/confirm-next changed selection but did not reveal it in the scrollable table. A new native test reproduces this by navigating from r2 to r9; it fails because the selected editor is outside the visible reading region. Fix automatic reveal after sizing, accounting for the sticky header/footer, including wraparound and large text. Final result is reopened until that regression passes and the revised state is inspected.

Resolution — round 3

Fixed in 96fc8df. Selection changes now reveal the selected editor after sizing, with sticky-header and visible-footer bounds. Container scrolling is preferred; minimum-height layouts move the page only as much as needed to expose the first two text lines. The actual editor offset and line height determine that space; no fixed row-height assumption or focus/playback call is used. Outer page scrolling is suppressed while the tools modal is open.

The initial new test failed on the original implementation. Intermediate tests exposed the difference between the entire table and its visible portion at 800 × 900, then the insufficient reading area at 800 × 600. Those constraints were corrected and all nine Atelier scenarios now pass, including offscreen progression, wraparound, long 28px text, minimum-window reveal and save-failure preservation afterward. Tests wait for layout frames and measure actual sticky TH bounds, not the offscreen THEAD container.

Post-fix evidence opened: output/playwright/atelier-next-offscreen-1440.png, atelier-large-text-1100.png, atelier-next-compact-800.png, atelier-next-minimum-800x600.png. Full and focused source/implementation comparisons were rebuilt and reopened; the selected visual composition is unchanged. P1 is resolved. No P0/P1/P2 remains; final result restored to passed.

Interaction evidence

  • Atelier native suite: 9 checks passed, no page errors (draft/approval failure safety, double click, focus, consent persistence, export, responsive sizes, text scaling, selected-text reveal).
  • Existing workflow: 8 checks passed (including proposal application/source preservation, snapshot restore, actual export artifacts and subsequent editing).
  • YouTube controlled iframe boundary: 4 checks passed. Live public YouTube IFrame API example (M7lc1UVf-VE): another 4 checks passed for dialogue/description selection and active editor seeking while paused, preserving drafts. No video download or AI request.
  • Guide/About: 5 checks passed, no failures or page errors.
  • Final implementation: 240 unit tests across 30 files, typecheck and build passed in an independent parent run. Task code review approved with no important findings; its precise-time readability note is fixed. Whole-branch selected-text reveal finding is addressed in round 3, and independent delta re-review approved with no remaining critical or important findings.

Implementation checklist

  • Fix notification overlap and table density.
  • Address precise-time legibility and test it.
  • Rebuild, recapture matching state, compare full + detail again.
  • Verify no remaining P0/P1/P2 and update final result.

Residual validation limits

This is automated/native visual QA, not a screen-description writer field trial. Windows Narrator, installed Hancom rendering, signed installer/updater and a design award are not certified by these checks. No merge, push, version bump or release has occurred for this redesign.