Skip to content

PPC residual wrong-FIX analysis

Reference truth is used only after the run to label wrong FIX; the source and gate columns use data already present in the real-time POS/selector outputs.

Wrong-FIX threshold: 0.500 m 3D error.

Key findings

  • filter residual wrong FIX is all baseline-selected and did not match a selector rule.
  • Best low-collateral deployable gate for filter in this POS-only replay: nis_per_obs > 20 catches 37.352% of wrong FIX while touching 5.103% of good FIX.
  • smoother_v22 residual wrong FIX is all baseline-selected and did not match a selector rule.
  • Best low-collateral deployable gate for smoother_v22 in this POS-only replay: nis_per_obs > 20 catches 57.156% of wrong FIX while touching 6.040% of good FIX.

Profile summary

Profile FIX epochs Wrong FIX Wrong/FIX Wrong source Rule matched
filter 44212 6238 14.109% baseline=6238 false=6238
smoother_v22 42386 2236 5.275% baseline=2236 false=2236

filter

Run FIX epochs Wrong FIX Wrong/FIX Wrong p95 Longest wrong span
tokyo_run1 9339 1936 20.730% 93.317 m 189314.800-189389.000 (74.400 s)
tokyo_run2 7718 350 4.535% 45.321 m 177731.400-177748.600 (17.400 s)
tokyo_run3 12381 993 8.020% 18.113 m 181699.600-181746.000 (46.600 s)
nagoya_run1 6026 875 14.520% 8.789 m 551153.000-551266.200 (113.400 s)
nagoya_run2 5938 790 13.304% 130.037 m 556155.400-556181.400 (26.200 s)
nagoya_run3 2810 1294 46.050% 78.754 m 554728.600-554769.600 (41.200 s)

Real-time discriminator percentiles

Metric Good FIX p50/p90/p95 Wrong FIX p50/p90/p95
ratio 28.000/112.700/180.700 4.600/35.100/89.500
baseline_m 2869.429/9427.663/9534.781 4730.074/9826.517/10091.115
nsat 17.000/23.000/25.000 14.000/19.000/20.000
prefit_rms_m 0.544/3.450/16.366 2.471/17.822/38.037
prefit_max_m 5.107/31.139/68.444 20.047/70.043/104.680
post_rms_m 0.536/1.456/3.694 1.421/7.139/7.848
post_max_m 5.107/23.451/27.944 18.720/28.920/29.454
nis_per_obs 0.619/5.135/20.342 6.664/58.370/73.126
outliers 0.000/1.000/5.000 0.000/8.000/12.000

Deployable gate simulation

Gate Wrong caught Wrong caught Good FIX harmed Good FIX harmed
ratio < 6 3719 59.618% 4449 11.716%
ratio < 8 4194 67.233% 6637 17.478%
post_rms > 4 m 2102 33.697% 1558 4.103%
post_rms > 6 m 1388 22.251% 482 1.269%
nis_per_obs > 20 2330 37.352% 1938 5.103%
nis_per_obs > 50 963 15.438% 278 0.732%
prefit_rms > 5 m 2518 40.366% 3409 8.977%
baseline > 7000 m and post_rms > 4 m 1138 18.243% 613 1.614%

smoother_v22

Run FIX epochs Wrong FIX Wrong/FIX Wrong p95 Longest wrong span
tokyo_run1 8760 656 7.489% 94.363 m 189461.600-189500.000 (38.600 s)
tokyo_run2 7513 99 1.318% 46.619 m 178229.200-178241.000 (12.000 s)
tokyo_run3 12153 701 5.768% 27.792 m 181691.400-181744.400 (53.200 s)
nagoya_run1 6023 42 0.697% 27.939 m 550968.200-550971.400 (3.400 s)
nagoya_run2 5531 433 7.829% 85.505 m 556150.000-556180.400 (30.600 s)
nagoya_run3 2406 305 12.677% 83.461 m 553814.800-553831.600 (17.000 s)

Real-time discriminator percentiles

Metric Good FIX p50/p90/p95 Wrong FIX p50/p90/p95
ratio 15.700/67.800/117.200 3.900/25.900/61.000
baseline_m 2945.070/9441.536/9632.869 3041.909/8961.940/9379.419
nsat 17.000/23.000/25.000 15.000/21.000/23.000
prefit_rms_m 0.555/3.817/16.288 5.408/10.861/35.272
prefit_max_m 5.510/32.038/68.619 23.194/51.986/95.049
post_rms_m 0.546/1.583/3.672 4.551/6.873/7.529
post_max_m 5.506/23.500/27.822 21.706/28.914/29.373
nis_per_obs 0.728/8.347/22.899 27.266/91.103/94.904
outliers 0.000/1.000/5.000 0.000/4.000/9.000

Deployable gate simulation

Gate Wrong caught Wrong caught Good FIX harmed Good FIX harmed
ratio < 6 1436 64.222% 8099 20.172%
ratio < 8 1619 72.406% 11097 27.639%
post_rms > 4 m 1189 53.175% 1607 4.002%
post_rms > 6 m 702 31.395% 507 1.263%
nis_per_obs > 20 1278 57.156% 2425 6.040%
nis_per_obs > 50 817 36.538% 498 1.240%
prefit_rms > 5 m 1180 52.773% 3653 9.098%
baseline > 7000 m and post_rms > 4 m 392 17.531% 650 1.619%