research-document GV-VE-2026-EBCC

Visual Engineering UI Decision Checklist

Visual Engineering UI Decision Checklist

Before implementation

  • What is the primary user task?
  • What must be recognized immediately?
  • What requires deliberate verification?
  • What is the intended reading and action order?
  • Which relationships must remain visible?
  • What are the consequences of misunderstanding or error?
  • Which existing product and design-system constraints apply?
  • What is authoritative for the state, uncertainty, completion, unresolved work, and legal actions this view presents?
  • Which state distinctions change what the user may safely conclude or do?
  • Which semantic scopes must remain distinct rather than being combined into one status?
  • Which visual channels carry consequential meaning, and what happens if hue, iconography, border, position, motion, or another channel is unavailable?
  • What is the Perceptual Failure Boundary for each important or critical semantic state?
  • Is a proposed accessibility preset based on measured need, explicit user choice, or an unsupported assumption about a diagnosis?
  • Does this task contain strong linguistic context, or is it low-context content such as an identifier, code, unfamiliar name, or critical value?

During implementation

  • Does visual order agree with semantic, DOM, reading, and focus order?
  • Is emphasis proportional to importance?
  • Are related elements closer or more strongly grouped than unrelated elements?
  • Can users distinguish status without color?
  • Are labels meaningful without implementation knowledge?
  • Does density support the actual scanning or comparison task?
  • Are components based on durable semantics or bounded behavior?
  • Is native HTML being replaced without a demonstrated benefit?
  • Does responsive behavior preserve meaning and task priority?
  • Are loading, empty, error, success, and recovery states designed?
  • Has an unknown, partial, stale, or unverifiable state been strengthened into a definite success, failure, empty, or complete state?
  • Where authoritative capabilities or legal actions exist, do controls derive from them rather than recreate legality locally?
  • Do summaries preserve the scope and uncertainty of the facts they aggregate?
  • Is decision-relevant unresolved work visible enough that users do not have to infer it from missing controls or generic errors?
  • Does consequential state remain identifiable without hue alone?
  • If CVD simulation is used, is it treated as screening rather than proof?
  • If presentation adapts to a user profile, are semantic meaning, legal actions, reading order, and programmatic state unchanged?
  • Are user preference and measured performance being kept distinct when either is used to justify a design decision?

Required verification

  • Keyboard-only navigation
  • Visible and unobscured focus
  • 200% text scaling and browser zoom
  • Narrow viewport and content reflow
  • Reduced motion
  • Forced colors or high contrast
  • Color-independent state recognition
  • Grayscale plus protan, deutan, and tritan simulation as screening conditions where color carries categorization
  • Text-spacing overrides where reading content is material
  • Semantic channel dropout for consequential states: remove each declared visible channel independently and verify important and critical meaning survives
  • Run the Visual Engineering robustness manifest when the repository defines important or critical semantic states
  • Review the minimal failure sets and improve low-boundary states when practical
  • Long, missing, and extreme content
  • Screen-reader semantics for critical workflows
  • First-glance hierarchy inspection
  • Deliberate verification of consequential information
  • Exercise at least one consequential non-happy semantic state and compare the rendered actions/status with the authoritative product state
  • Where stale state can change meaning or legal action, verify that old presentation evidence is not treated as proof of the current state

Motion model selection

Before adding or reviewing motion:

  • What does the motion communicate?
  • What state or input is authoritative?
  • Is the phenomenon inertial/spring, gravity-derived, constant-velocity/cadence, direct-manipulation/value tracking, or perceptual interpolation?
  • Is overshoot physically and semantically safe?
  • Can the user interrupt or rapidly repeat the interaction without stale queued motion?
  • Does direct manipulation track the pointer, scroll position, or authoritative value without decorative lag?
  • Can determinate progress ever visually exceed the authoritative value?
  • Does repeated activity stop when the activity stops?
  • Is a new duration/easing inherited from the shared model, or is there a documented reason for a new one?
  • What is the model-specific reduced-motion substitution?
  • Is the interface still correct when progressive animation support is unavailable?

During verification:

  • Exercise rapid repeated activation and interruption.
  • Verify final visual state equals actual semantic state.
  • Verify focus movement/restoration does not wait on decoration.
  • Verify pointer drag/resize/scrub does not lag.
  • Verify determinate progress remains bounded.
  • Verify indeterminate cadence does not imply false progress.
  • Verify reduced motion removes spatial/repeated effects without losing state or feedback.
  • Verify simultaneous effects do not clobber one another.
  • Verify animation is not the only carrier of consequential meaning.

Application polish gate

  • Have applicable polish dimensions been identified?
  • Have reachable non-happy states been exercised rather than inferred?
  • Have high-risk state and environment seams been tested?
  • Have pathological content and boundary fixtures been used?
  • Does persistence presentation tell the truth about local, pending, persisted, failed and conflicting state?
  • Are untested environments and states recorded as unknown rather than passed?
  • Is there evidence behind a zero-finding review?
  • Have recurring defect classes been pushed upstream into shared components, rules or fixtures where appropriate?

Application polish gate

  • Have applicable polish dimensions been identified?
  • Have reachable non-happy states been exercised rather than inferred?
  • Have high-risk state and environment seams been tested?
  • Have pathological content and boundary fixtures been used?
  • Does persistence presentation tell the truth about local, pending, persisted, failed and conflicting state?
  • Are untested environments and states recorded as unknown rather than passed?
  • Is there evidence behind a zero-finding review?
  • Have recurring defect classes been pushed upstream into shared components, rules or fixtures where appropriate?

Agent handoff

Report:

  • Visual Engineering context version and source commit
  • Principles applied
  • Verification performed
  • Material deviations and rationale
  • Unresolved evidence or product questions
  • Material semantic-projection assumptions, including the authoritative state source when it is not obvious
  • Perceptual robustness manifest result, failure boundaries, and any accepted low-boundary states