fix: #150 follow-ups (drawer header contrast, Live heat sync, RX days race, IATA interval cap)

[dborup]

Relates to #150

The four follow-ups from the review of the issue batch (#109–#126). Each item has a test commit followed by a fix commit. The fifth point in the issue comment (analytics warm-up vs. frontend retry) is not part of this PR.

1. Nav drawer header unreadable in the light theme

style.css gave .nav-drawer-header the background --surface-2 (white in light), while nav-drawer.css paints the title with --nav-text (white) and the close button with --nav-text-muted. The header now uses --nav-bg2, which the customizer already maps (navBg2 in THEME_CSS_MAP) and every preset pairs with navText. No new variable, so the customizer is unchanged.

• Test: step (k) in test-nav-drawer-1064-e2e.js computes, from computed styles, the contrast of the title (≥ 4.5:1) and the close button (≥ 3:1) against the header background in light and dark.
• master: light 1.00:1 (white on white), dark passes. Branch: both pass (title white on rgb(26, 26, 46)).
• In dark mode the header changes from #232340 to #1a1a2e, the same colour as the top nav.

2. Live: Heat checkbox vs. heat layer

syncHeatToggleToMatrix(false) now also sets ht.checked = heatEnabled. Switching Matrix off after init also rebuilds the heat layer when Heat is on. Without that rebuild, the checkbox would now be checked while the layer stayed hidden.

New steps in test-issue-125-live-toggles-early-e2e.js:

• Matrix and Heat saved ON, Matrix clicked OFF while init is held: Heat is checked and enabled, the layer is shown after init, and Matrix ON/OFF after init brings Heat back as saved. Red on master.
• Same with Heat saved OFF: Heat stays unchecked with no layer.
• fix(live): wire every persisted view toggle before Live init awaits #135 regression: leave Live while init awaits loadNodes(), come back, then one click on each of the nine toggles must write localStorage exactly once. api() shares the in-flight /api/nodes request, so both inits continue when it is answered. This step is green on master because fix(live): wire every persisted view toggle before Live init awaits #135 fixed the bug. Against master with live.js from before fix(live): wire every persisted view toggle before Live init awaits #135 (d778bc40^1) it fails with "8 of 9 toggles" and 2 writes per click, which matches the issue.

3. RX Coverage: stale days response

drawCoverage() and loadBoard() each take a sequence number (coverageSeq, boardSeq) next to the mount generation from #136. A response, or a leaderboard failure, renders only if no newer request of its kind has started. As a side effect, the same now holds for coverage requests from quick pans and rx changes.

• Test 13 in test-issue-124-rx-coverage-viewport.js: switch days 7 → 30, answer the days=30 requests, then release the delayed days=7 responses. The days=30 data must stay. A failed older leaderboard request must not replace a newer one either. Red on master (the days=7 leaderboard replaces the days=30 one).

4. Ingestor: iataWarnIntervalSec overflow

IATAWarnInterval() caps the configured value at 86400 s (24 h). Before, 1e10 overflowed time.Duration to a negative interval, so every drop was logged. The cap is documented in config.go and config.example.json; the field is not in the README.

• TestIATAWarnIntervalClampsHugeValues: 1e10 s (direct and from JSON), MaxInt, 86401 s and a year all give 24 h; 86399 s stays as is. Red on master.
• TestIATADropThrottleOldestTracksStaggeredEntries and TestIATADropThrottleOldestFollowsAnEarlierDrop cover the oldest lower bound with staggered and out-of-order timestamps. They pass on master, which is expected: they pin the existing optimisation and kill four mutants the old tests let survive (see below).

Tests

Suite	master	branch
test-nav-drawer-1064-e2e.js	12/13 (new light step red)	13/13
test-issue-125-live-toggles-early-e2e.js	16/17 with the new steps (Heat sync red); master's own file 14/14	17/17
same, live.js from before #135	2/17 (regression step: 8 of 9 toggles ×2)	–
test-issue-124-rx-coverage-viewport.js	12/13 (test 13 red)	13/13
go test ./... ingestor (-run IATA)	TestIATAWarnIntervalClampsHugeValues red	green

Unchanged on master and branch: test-frontend-helpers.js has 2 favStar failures (known baseline). test-rx-coverage-escape.js fails on both with "could not locate row-builder end"; it is not run in CI. test-pr-1490-live-map-gpu-animations-e2e.js cannot load @playwright/test locally on either side; it is not in CI.

Mutants

All run in copies, never in the worktree:

Mutant	Caught by
header background var(--surface) / back to var(--surface-2, var(--surface))	(k) light theme
drop ht.checked = heatEnabled	#150 Matrix+Heat step
ht.checked = true	Heat-OFF step, plus two existing Heat steps
drop the heat rebuild on Matrix OFF	#150 Matrix+Heat step
live.js from before #135	#135 regression step
board .then / coverage .then / board .catch without the sequence check	test 13
boardSeq not bumped	test 13
clamp removed / cap 25 h	TestIATAWarnIntervalClampsHugeValues
sweep leaves oldest = now	…OldestTracksStaggeredEntries
insert always sets oldest = now	both new oldest tests
insert condition without now.Before	…OldestFollowsAnEarlierDrop
insert condition without len == 0	…OldestTracksStaggeredEntries (sweep count)

The four oldest mutants survived the tests on master.

Perf

None of the changes is in a hot path. Item 3 adds one integer compare per response, item 2 one assignment per Matrix toggle, and item 4 one compare per dropped message.

Notes

• No new npm dependencies and no new test files, so test-all.sh and deploy.yml are unchanged. The Kpa-clawbot guard count is still 9.
• scripts/check-xss-sinks.sh --diff and scripts/check-css-vars.js are clean.

🤖 Generated with Claude Code