--navi776-tc + adaptive-R extension A/B (Tokyo run1 full)

Objective

PR #438 added two default-off adaptive-R variants (--rtk-adaptive-noise-phase-only, --rtk-adaptive-noise-per-system-alpha). The Phase 3 TC main line ships --navi776-tc (M3 closed loop + M4 velocity + SD Doppler rows + gated shared-alpha adaptive noise). This A/B checks whether combining the new knobs with the TC preset helps or hurts, before any preset-promotion decision.

Method

Tokyo run1 full (11845 scored epochs), clang-ninja gnss_fuse, --preset low-cost --ratio 2.4 --max-subset-ar-drop-steps 18 --rtk-snr-weighting --no-arfilter --max-epochs 0. Scored against reference.csv ECEF.

Result

Config	Fix %	rmsH	maxH	<50cm
OFF (basic only)	76.70%	16.09 m	215.5 m	8644
--navi776-tc	79.77%	14.54 m	215.5 m	9337
--navi776-tc + phase-only	55.06%	17.25 m	215.5 m	6151
--navi776-tc + per-system	77.90%	15.01 m	215.5 m	8772

The --navi776-tc shared-alpha preset is reproduced (79.77% fix, matching the sign-off table). Adding phase-only collapses the fix rate to 55.06% and <50 cm to 6151: the code-variance adaptation is essential to the TC preset's fix-rate gain, so removing it is strongly negative. Adding per-system alpha also regresses (77.90% fix, 8772 <50 cm): the shared phase 0.9 / code 0.5 constants are better on this dataset than the per-system table.

Conclusion

The Phase 3 TC main line should keep the shared-alpha adaptive noise inside --navi776-tc. The two new adaptive-R knobs are not folded into the TC preset; combining them is measurably worse. Both knobs remain default-OFF and independent; no preset-promotion, sealed run2/run3 untouched. This closes the "reflect adaptive-R into the Phase 3 TC main line" backlog item with a negative result.