summaryrefslogtreecommitdiff
path: root/android-app/app/src/main/res
diff options
context:
space:
mode:
authorClaude <noreply@anthropic.com>2026-05-26 06:39:05 +0000
committerClaude <noreply@anthropic.com>2026-05-26 06:39:05 +0000
commit8dbf3f883274fc0ddd16e5507f732629c85e3c91 (patch)
tree87a18390a9e3c17d6cf6bd1fc17cdcc6a7632cf5 /android-app/app/src/main/res
parent36bee4a5aa2b3ecbf233bc02d70eb01d7ac737a8 (diff)
Redesign tack detection: time-based windows + circularMAD stability
Replace the point-count window (HALF_WIN=2) with 30s time-based settle windows that give consistent reliability regardless of GPS update rate (1 Hz FULL vs 0.2 Hz ECONOMY). Key changes: - Before/after windows are time-based (T_SETTLE=30s) around a guard zone (T_MANEUVER=30s), not point counts. - Stability uses circularMAD (<20°) across all fixes in each window, replacing the single-pair spread check. This correctly handles 0°/360° wrap and suppresses random anchor-watch COG noise (MAD≈90°). - No SOG gate: averaging 30 fixes over 30s reduces noise by √30 regardless of boat speed. A boat at anchor has MAD≈90° and is rejected by the stability check. - MAX_DELTA raised to 160° (covers deep jibes the old 140° missed). - START_SKIP reduced to 60s (stability check handles cold-start noise). - De-duplication uses adjacent-candidate comparison so a long stream of raw candidates from one maneuver stays in a single group even if it spans >MIN_GAP_MS in total. - Position refined to max instantaneous heading rate within the maneuver zone. - Mode: CRUISING (conservative). Race mode noted as future backlog. 16/16 unit tests pass at 1 Hz (FULL mode GPS rate). https://claude.ai/code/session_01YUbuZNDAoLea4cf9UGQ9qn
Diffstat (limited to 'android-app/app/src/main/res')
0 files changed, 0 insertions, 0 deletions