[PurpleSentinel]

Closes #219.

#219 was already half fixed

#219 and #221 are the same road test, filed twice. #221 was closed by #224, which gave Race its own screen — so the headline complaint is fixed and on the device:

criterion	state
current lap counting up in mm:ss.sss	✅ RaceScreen, nine cells; update_elapsed advances it on every observation
fastest lap	✅
session remaining	✅
--:--.--- before the first crossing	✅ presenter, asserted in test_ui_foundation
Track Day unchanged	✅ TrackdayScreen untouched
a decision recorded on lap_delta	❌
host tests cover Race populated and Track Day not	❌ the Race half only, at presenter level
fields repaint only when their value changes	❌ for the track name

This closes the three.

The routing was never asserted

running_screen() compared the mode to DeviceMode::race inside a private member of screen_router.cpp. That comparison is the only thing keeping a Track Day session off a screen with lap figures on it, and no test could reach it.

ui::running_screen_for(DeviceMode) now states the rule in the component, and the router asks it. The default is deliberately the withholding screen: a mode added without a thought for this rule gets the one that cannot show a lap time.

Four assertions, the last of them the one worth having:

// a mode is drawn on RaceScreen exactly when it is permitted lap times
assert((running_screen_for(mode) == RunningScreen::race) == visibility_for(mode).lap_times);

Grant laps to a mode routed to TrackdayScreen and they silently never appear — which is exactly the bug this issue reported. Route a mode without them to RaceScreen and it shows what #132 forbids. Neither can now land without failing a test.

lap_delta: the flag goes

Race was granted lap_delta when its running screen had no lap fields at all. The screen it has now shows three figures and no delta, and #58 — the reference mapping and validity rules a delta worth trusting needs — is still open.

So it is false for every mode, with a test holding it there. A permission for something nothing draws describes a display that does not exist, which is the same mistake as a plausible reading. The field stays in ModeVisibility because #58 is what flips it back.

One real repaint

set_track_name was called on every display tick, and lv_label_set_text invalidates whether or not the text differs. The armed track cannot change during a session, so that was a repaint of the widest label on the screen 25 times a second — sitting next to three lap rows that were already change-gated for precisely that reason (#213). Same gate, same place.

Verification

• make check → CHECK_RC=0, including the four new assertions in device-mode-test
• make firmware-container-build → FW_RC=0, 32% of the app partition free
• Bench-verifiable: the routing and the delta decision are host tests. The track-name gate is not directly observable — it removes a repaint rather than changing a pixel, so what is checkable on the bench is that the Race screen still shows the track name it did before.
• A lap time end to end still needs a drive, and A session interrupted by power loss is lost entirely, including its laps #222 means the re-test has to be stopped deliberately on the device before power-off.

🤖 Generated with Claude Code