Skip to content

Latest commit

 

History

History
54 lines (45 loc) · 2.81 KB

File metadata and controls

54 lines (45 loc) · 2.81 KB

Frontend UX Review Notes

Use these Chronicle-specific lessons together with the installed frontend-design-principles skill and the repository's screenshots skill.

Visual hierarchy

  • Metadata should look like metadata. Use muted text and existing metadata rows; do not give transcript versions, counts, or implementation details a colored panel or prominent chip.
  • Color must communicate action or state. Decorative blue/amber treatments make low-priority information compete with the conversation title and primary task.
  • Do not allocate a dedicated row to one small fact. Fold secondary information into an existing row or omit it when it does not help a decision.
  • Apply the squint test before presenting a UI: the title and primary task should remain dominant, and no incidental control should jump out.

Interaction design

  • Avoid duplicate actions. Destructive and maintenance actions such as reprocessing belong in the three-dots menu when they are not the page's primary task.
  • When making a card clickable, preserve every nested interaction. Use semantic buttons for expanders and explicit event boundaries around editors, dropdowns, playback, and selectable content. Verify both card navigation and nested controls in a browser regression check.
  • A large click target needs a subtle hover/focus affordance; it should not require a redundant "View details" link.
  • Small icon-only controls need an adequate hit area, an accessible name, and a visible keyboard focus state.

Review workflows

  • A review surface must provide enough coverage to support judgment. One or two clips are anecdotes, not a useful acoustic sample.
  • Keep semantically different training data separate. Chronicle uses Noise for non-speech/undeciphered audio and Background Speech for intelligible speech from TV, radio, media, or distant non-participants.
  • Show source context, timestamps, playback, coverage counts, and enough diverse candidates to compare acoustic conditions. Balance quality with responsiveness; the current background sampler targets 20 candidates per bucket across eight conversations.

Visual verification

  • Capture the rendered page after the final change at 1440×1000 using skills/screenshots/SKILL.md; authenticated and dynamic routes must use real data.
  • Inspect the screenshot rather than treating capture as proof. Look for wasted rows, competing accents, inconsistent control styling, tiny targets, raw IDs, redundant links, and controls that appear when they have no content.
  • Exercise interactions separately with Playwright. A static screenshot cannot detect event bubbling, navigation conflicts, empty expanders, or unusable menus.
  • Treat login redirects, blank pages, unresolved loading, and development-server reload loops as failed verification—not successful screenshots.