examples/

Small, hand-built layouts the Lumo tools and tests use as anchors. Every file is referenced from at least one of:

• the worked examples in skill/SKILL.md
• the smoke commands in README.md
• the CI job that pip-installs lumo-mobile and runs each CLI

Keep them small, keep them honest. If you change one, make sure the findings the README screenshot promises still match.

What each file is for

File	Used by	What it demonstrates
theory_bad_layout.json	lumo-theory check	Several issues at once — undersized icon (close is 32 dp, below 48 dp Material minimum), primary action save in a top corner (reach_primary_in_top_corner), 7 equal-weight nav items (Hick overload). Used as the "find everything wrong" demo.
theory_good_layout.json	lumo-theory check	A well-designed Material screen — bottom nav with 4 items, well-separated content groups, primary CTA in the safe bottom area. Tool returns "OK no findings". Used to show the tool doesn't false-positive on healthy UI.
parity_android.json	lumo-parity diff	The Android side: 5 elements including card_offer at height=16 and a fab_add that has no iOS twin.
parity_ios.json	lumo-parity diff	The iOS side: paired with parity_android.json. card_offer is at height=48 (the classic "iOS uses 3× because Retina" junior bug), nav_back is at 44 pt (legitimate platform default — Lumo whitelists it), fab_add is missing entirely.
lumo.config.json	lumo-parity diff --config	Tiny design system: primary_button_height: 56. Both Android and iOS layouts comply, so this config doesn't add findings — it's here as the "happy path" example. Edit it to 48 and re-run parity to see two new design_system_height_mismatch_* findings appear.
source_bad_compose.kt	lumo-source check	Compose anchor — one of each: Modifier.size(32.dp) (a11y), Color(0xFFAA0000) (token), RoundedCornerShape(13.dp) (consistency). Includes MaterialTheme.* counter-cases that must NOT trip the checker.
source_bad_swiftui.swift	lumo-source check	SwiftUI anchor — one of each: .frame(width: 32, height: 32) (a11y, HIG 44pt), Color(red:green:blue:) (token), .cornerRadius(13) (consistency). Includes Color.red and Color("brandPrimary") counter-cases that must NOT trip the checker.

Layout JSON schema (quick reference)

{
  "screen":  { "width": 411, "height": 891, "unit": "dp" },
  "source":  "measured | code-estimated | description-estimated",
  "elements": [
    {
      "id": "btn_continue",
      "role": "primary_action",
      "x": 24, "y": 800, "w": 363, "h": 56,
      "group": "form_actions",
      "weight": "primary"
    }
  ]
}

The source field is honesty about how the coordinates were obtained:

• measured — from a real device or a snapshot-testing framework (Espresso, XCUITest, Compose onGloballyPositioned, SwiftUI GeometryReader, Paparazzi, xcodebuild test --only-testing).
• code-estimated — parsed statically from Compose / SwiftUI source. Theme tokens, fillMaxWidth, dynamic type are guesses at this layer.
• description-estimated — built from a screenshot description or a natural-language prompt. Lowest confidence.

The Lumo tools propagate this label to every finding so the consumer can weigh trust honestly.

See skill/SKILL.md for the full schema (roles, weights, group semantics) and tools/lumo/theory/core.py / tools/lumo/parity/core.py for the authoritative type definitions.