Fix code-review findings C1–C3, H1, H2: line init, midnight rollover, detection window, input validation, buffer wraparound

[TheAngryRaven]

Addresses all five findings from the professional code review (scored 4.5/10). Each fix lands with regression tests; the C3 and H2 tests were verified to fail against the pre-fix code. All 89 host tests green, and the pinned NMEA replay goldens are unchanged (no timing drift on real-track data).

C1 — Start/finish line never initialized

• All 12 line-coordinate members are now zero-initialized in the header.
• New startFinishLineConfigured flag (mirroring the sector flags) gates checkStartFinish() in loop() — calling loop() before setStartFinishLine() is now a safe no-op instead of UB on indeterminate doubles.
• All three line setters reject degenerate (A == B, e.g. the 0.00 placeholders shipped in basic_oled_example) or non-finite lines, with a debug warning.
• New public getter: isStartFinishLineConfigured().

C2 — Midnight rollover / backwards GPS time

• New shared timeSinceMidnightDelta() helper normalizes every duration across the 86,399,999 → 0 UTC midnight wrap. Applied at all subtraction sites in DovesLapTimer (lap, all three sectors, getCurrentLapTime, getPaceDifference) and WaypointLapTimer.
• The crossing interpolator computes its fix-pair delta in double, wrap-normalizes it, and clamps it to CROSSING_MAX_FIX_GAP_MS (10 s) before any conversion back to unsigned long — the old path underflowed unsigned subtraction and pushed an out-of-range double through double→unsigned conversion ([conv.fpint] UB).
• Crossing validity is now an explicit bool through interpolateCrossingPoint() / _detectLineCrossing() instead of the time != 0 sentinel, so a legitimate crossing at exactly 00:00:00.000 is no longer discarded.
• Documented limitation (CLAUDE.md): a small backwards time step that isn't a midnight wrap is indistinguishable from one; the interpolator clamp catches it inside a crossing zone, lap-level deltas do not.

C3 — Course detection false-matched longer layouts

• _checkWaypointProximity() now re-anchors _waypointOdometer after a completed proximity pass with no match, so distanceSinceWaypoint can't accumulate across laps and falsely match a 2×-length layout on the next pass. Same root-cause class as the v4.1.0 rejection-cooldown fix (CLAUDE.md issue little review with opus #13); regression test mirrors that one.

H1 — GPS input validation

• New GeoMath.h helpers geoIsFinite() / geoCoordinatesValid() (no isnan/isinf dependency — portable across Arduino cores and host).
• DovesLapTimer::loop(), WaypointLapTimer::loop(), and CourseDetector::update() reject NaN/Inf, |lat|>90, |lng|>180, and exact (0,0) fixes; non-finite altitude/speed are sanitized rather than dropping the fix.
• Teleport rejection: a single-fix jump > GPS_MAX_PLAUSIBLE_JUMP_METERS (500 m) is dropped; after GPS_JUMP_REACCEPT_COUNT (3) consecutive far fixes the new position is accepted as a genuine relocation and re-seeded without crediting the gap to the odometer.

H2 — Crossing-buffer wraparound

• interpolateCrossingPoint() unwinds the ring chronologically (oldest → newest) before scanning for the straddling pair. Once the ring wrapped (kart parked in the zone: standing start, red flag, stop-and-go), physical index order ≠ chronological order and the seam pair (newest adjacent to oldest) masqueraded as the crossing — fabricating a crossing from points minutes apart or invalidating the real one. Verified failing pre-fix: the park-cross-park scenario dropped the crossing entirely.

Tests

• test_midnight_rollover.cpp (7): lap/sector/current-lap/crossing-zone timing across the wrap, WaypointLapTimer across the wrap, delta-helper units.
• test_input_validation.cpp (11): adversarial fixes, teleport/relocation, unconfigured + degenerate line safety, detector/WLT guards.
• test_buffer_wraparound.cpp (3): park-cross-park, standing start on the grid, normal lap after a wrapped start.
• test_course_detector.cpp (+2): no-match re-anchor + post-re-anchor genuine match.

CHANGELOG, README, CLAUDE.md, keywords.txt, and test/README.md updated accordingly.

https://claude.ai/code/session_01XNmVH6ySJ6Dz2QKDjB69mE

---

Generated by Claude Code

[github-actions]

GCC Code Coverage Report

📂 Overall coverage

Metric	Coverage
Lines	🔴 833/1242 (67.1%)
Functions	🔴 127/195 (65.1%)
Branches	🔴 410/776 (52.8%)

📄 File coverage

File	Lines	Functions	Branches
src/CourseDetector.cpp	🟢 94/100 (94.0%)	🟡 17/19 (89.5%)	🟡 39/46 (84.8%)
src/CourseManager.cpp	🔴 0/142 (0.0%)	🔴 0/22 (0.0%)	🔴 0/78 (0.0%)
src/CourseManager.h	🔴 0/24 (0.0%)	🔴 0/8 (0.0%)	🔴 0/16 (0.0%)
src/DovesLapTimer.cpp	🟡 426/545 (78.2%)	🟡 43/57 (75.4%)	🔴 268/440 (60.9%)
src/DovesLapTimer.h	🔴 123/187 (65.8%)	🟡 41/48 (85.4%)	🔴 38/88 (43.2%)
src/GeoMath.h	🟢 21/21 (100.0%)	🟢 4/4 (100.0%)	🟡 21/24 (87.5%)
src/WaypointLapTimer.cpp	🔴 148/199 (74.4%)	🔴 15/29 (51.7%)	🔴 37/68 (54.4%)
src/WaypointLapTimer.h	🟡 21/24 (87.5%)	🟡 7/8 (87.5%)	🔴 7/16 (43.8%)