Raw data processing behind the two known instrument calibration issues — computed live from the current trip history, not a frozen snapshot.
Each sample is one independent offset estimate from a single real back-to-back tack: (raw AWA on the leg before the tack + raw AWA on the leg after) / 2. This estimator cancels out to zero on a correctly calibrated sensor and is robust to a constant additive miscalibration — it uses only the boat's own raw AWA readings at real tack events, no GPS heading assumption required for this particular number.
The paddlewheel log (STW) reads faster on starboard tack than port when close-hauled, and the gap grows with wind strength. If that's a real speed difference, GPS speed over ground (SOG) — which has no way to know or care which way the boat is heeled — should show the same pattern. Below: the same tack comparison run twice, once on STW and once on SOG, both pooled across all trips and per-trip (since current direction differs on every trip, a bias that's consistently in the same direction across many unrelated days is much harder to explain by current than by something constant like a heel-dependent sensor reading).