Visual consistency: review flagged typography differences across cards

[tashda]

Summary

While doing the typography token sweep (#33 follow-up), several places where similar visual elements use different font sizes were flagged. Most are probably intentional, but they should be confirmed before considering the design "consistent."

The tokens already exist in two flavors per role (e.g. `eyebrowLabel` 11pt vs `eyebrowLabelLarge` 12pt), so resolving each item here is a one-line token swap once you decide.

Flagged differences

A. Eyebrow size — 11pt vs 12pt

Cards using 11pt	Cards using 12pt
Climate, Fan, Light, Cover, Switch, Lock, Group, DeviceCard	SensorCard, RemoteCard, GenericExposeCard

The 12pt cards are content-heavy tiles rather than hero cards. Possibly intentional. Decision needed:

• Treat all card headers identically and unify on 11pt? (cleaner consistency)
• Keep 12pt for content tiles as a deliberate emphasis tier? (semantic distinction)

B. Tile metric value — 24pt vs 30pt

Cards using 30pt (primary tile)	Cards using 24pt (identity tile)
FanControl, RemoteCard, SensorCard, SwitchControl	GroupCard, ClimateControl, DeviceCard

Likely intentional — 24pt reads as "secondary identity stats" (linkquality/battery), 30pt as "primary card metric." Confirm.

C. Metric unit size — 13 / 14 / 15pt

• 15pt unit pairs with 30pt value (RemoteCard, SensorCard, etc.)
• 14pt unit pairs with 24pt value (DeviceCard, GroupCard)
• 13pt unit appears in LightControl snapshot rows and footer bars

Looks consistent — unit size scales with value size. No action needed unless eyeballing reveals otherwise.

D. Notification banner level icon — 14pt vs 15pt

`FastTrackBanner` uses 14pt, `InAppNotificationBanner` uses 15pt for the same level glyph. Probably should match. Trivial fix.

E. Hero "Off" state — 48pt vs 56pt for active metric

Cards drop from 56pt to 48pt when there's no numeric metric (Light "Off", Switch "Off"). Likely deliberate but worth a glance — does the size shift read as a state change or as inconsistency?

Acceptance criteria

For each item above:

• Decision recorded (unify, keep distinct, or out-of-scope).
• If unifying: swap the outlier site's token to the dominant variant; remove the now-unused `...Large` token if no longer used.
• If keeping distinct: add a one-line code comment at the outlier site explaining the intent so future audits don't re-flag it.

Notes

This is design-decision work, not engineering — none of these are bugs. Schedule alongside the next UI polish pass.

[tashda]

Resolved in `c60f97d`. Recap of what was decided:

#	Item	Decision	Reason
A	Eyebrow 11pt vs 12pt	Unified on 11pt	The 12pt cards (Sensor / Remote / GenericExpose) were dual-using their eyebrow as a card title. Card eyebrows everywhere else are 11pt; consistency wins over the dual-duty rationale.
B	Tile metric 24pt vs 30pt	Kept distinct, renamed tokens	The two sizes serve different roles: 24pt for packed 2×2 identity grids (DeviceCard/GroupCard) and the climate setpoint number, 30pt for prominent feature tiles with room to breathe. Tokens renamed (`identityTileValue` / `featureTileValue`) so the semantic boundary is explicit.
C	Unit size scales with value size	No action	This is the right pattern — proportional unit sizing keeps each tile visually balanced.
D	Banner glyph 14pt vs 15pt	Unified on 15pt	The 1pt difference wasn't doing visual work; FastTrackBanner already differs via padding and capsule shape. Dropped the redundant token.
E	Hero state 48pt vs metric 56pt	No action	The 8pt drop is deliberate — signals "no numeric data here" vs "data to scan", same pattern Apple Home uses on tile cards.

Net change: 2 visible (A, D), 1 token rename (B), 2 leaves (C, E). All inline `// NOTE: see #36.X` markers and the now-unused tokens (`eyebrowLabelLarge`, `fastTrackLevelIcon`) are gone.

Folded into PR #35.