AsianPlayer's picture
|
download
raw
170 kB

DMS baselines — operating points at equal alert load (baselines_dms/report_dms.py; numbers copied from eval/operating_points.py CSVs)

item dms_r0 — R0 fixed gaze/eye rule (own HALPE face pack) — 2026-09-14T14:07:57Z

Status: adapted; zero-cost fixed temporal rule on our own face landmarks; no video decoded; single deterministic run (no seeds). Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_r0_20260914T140622Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 54.674 326 / 30 0.500 [0.320, 0.730] 0.113 [0.065, 0.161] --
dms_matched engaged head 48.641 183 / 32 0.469 [0.286, 0.710] 0.300 [0.120, 0.448] --
dms_matched pooled head -- 509 / 62 0.484 [0.353, 0.620] 0.164 [0.093, 0.232] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 31.302 326 / 30 0.133 [0.000, 0.300] 0.148 [0.000, 0.238] --
low10 engaged head 23.026 183 / 32 0.125 [0.000, 0.333] 0.400 [0.000, 0.727] --
low10 pooled head -- 509 / 62 0.129 [0.021, 0.250] 0.216 [0.042, 0.364] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 20 -0.280 [-1.808, +1.364] EQUAL (CI ⊃ 0)
dms_matched engaged 18 -1.006 [-2.587, +1.696] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -0.655 [-1.948, +0.807] EQUAL (CI ⊃ 0)
low10 manual 2 +1.590 [-0.165, +2.888] EQUAL (CI ⊃ 0)
low10 engaged 7 -0.500 [-2.414, +2.261] EQUAL (CI ⊃ 0)
low10 pooled 9 +0.511 [-0.979, +2.192] EQUAL (CI ⊃ 0)

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 8.531 0.214 0.196 [0.134, 0.246]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_videomae2_learned — VideoMAEv2 ViT-S frozen features, learned read-out — 2026-09-16T04:38:53Z

Status: adapted; frozen 384-d window embeddings -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_videomae2_learned_20260916T042722Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.586 329 / 31 0.645 [0.486, 0.790] 0.235 [0.140, 0.358] --
dms_matched engaged head 49.358 183 / 32 0.688 [0.357, 0.857] 0.367 [0.117, 0.592] --
dms_matched pooled head -- 512 / 63 0.667 [0.488, 0.810] 0.290 [0.160, 0.438] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.966 329 / 31 0.194 [0.038, 0.357] 0.667 [0.200, 1.000] --
low10 engaged head 10.039 183 / 32 0.188 [0.000, 0.279] 0.857 [0.800, 1.000] --
low10 pooled head -- 512 / 63 0.190 [0.080, 0.288] 0.750 [0.500, 0.926] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 12.423 329 / 31 0.258 [0.077, 0.452] 0.667 [0.333, 0.909] --
same_recall engaged head 21.313 183 / 32 0.406 [0.143, 0.538] 0.722 [0.333, 0.882] --
same_recall pooled head -- 512 / 63 0.333 [0.182, 0.466] 0.700 [0.455, 0.871] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 1.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.016 [0.000, 0.059] 1.000 [1.000, 1.000] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.053 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 183 / 32 0.031 [0.000, 0.111] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.032 [0.000, 0.078] 1.000 [1.000, 1.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 5.018 329 / 31 0.097 [0.000, 0.200] 0.600 [0.200, 1.000] --
low5 engaged head 4.980 183 / 32 0.094 [0.000, 0.152] 0.750 [0.667, 1.000] --
low5 pooled head -- 512 / 63 0.095 [0.029, 0.146] 0.667 [0.400, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.358 329 / 31 0.645 [0.486, 0.790] 0.222 [0.131, 0.343] --
dms_matched engaged head 53.382 183 / 32 0.719 [0.400, 0.886] 0.359 [0.120, 0.585] --
dms_matched pooled head -- 512 / 63 0.683 [0.523, 0.820] 0.279 [0.152, 0.422] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.966 329 / 31 0.194 [0.038, 0.357] 0.667 [0.200, 1.000] --
low10 engaged head 10.039 183 / 32 0.188 [0.000, 0.279] 0.857 [0.800, 1.000] --
low10 pooled head -- 512 / 63 0.190 [0.080, 0.288] 0.750 [0.500, 0.926] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 24.635 329 / 31 0.323 [0.143, 0.529] 0.476 [0.200, 0.800] --
same_recall engaged head 32.945 183 / 32 0.500 [0.208, 0.641] 0.471 [0.138, 0.736] --
same_recall pooled head -- 512 / 63 0.413 [0.250, 0.563] 0.473 [0.256, 0.698] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 1.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.016 [0.000, 0.059] 1.000 [1.000, 1.000] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.053 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 183 / 32 0.031 [0.000, 0.111] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.032 [0.000, 0.078] 1.000 [1.000, 1.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 5.018 329 / 31 0.097 [0.000, 0.200] 0.600 [0.200, 1.000] --
low5 engaged head 4.980 183 / 32 0.094 [0.000, 0.152] 0.750 [0.667, 1.000] --
low5 pooled head -- 512 / 63 0.095 [0.029, 0.146] 0.667 [0.400, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.796 329 / 31 0.581 [0.400, 0.769] 0.231 [0.138, 0.356] --
dms_matched engaged head 48.721 183 / 32 0.688 [0.357, 0.857] 0.373 [0.118, 0.594] --
dms_matched pooled head -- 512 / 63 0.635 [0.446, 0.803] 0.292 [0.164, 0.455] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.966 329 / 31 0.194 [0.038, 0.357] 0.667 [0.200, 1.000] --
low10 engaged head 10.039 183 / 32 0.188 [0.000, 0.279] 0.857 [0.800, 1.000] --
low10 pooled head -- 512 / 63 0.190 [0.080, 0.288] 0.750 [0.500, 0.926] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 11.019 329 / 31 0.226 [0.077, 0.389] 0.636 [0.333, 0.889] --
same_recall engaged head 16.453 183 / 32 0.344 [0.000, 0.500] 0.786 [0.000, 1.000] --
same_recall pooled head -- 512 / 63 0.286 [0.120, 0.414] 0.720 [0.429, 0.875] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 1.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.016 [0.000, 0.059] 1.000 [1.000, 1.000] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.053 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 183 / 32 0.031 [0.000, 0.111] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.032 [0.000, 0.078] 1.000 [1.000, 1.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 5.018 329 / 31 0.097 [0.000, 0.200] 0.600 [0.200, 1.000] --
low5 engaged head 4.980 183 / 32 0.094 [0.000, 0.152] 0.750 [0.667, 1.000] --
low5 pooled head -- 512 / 63 0.095 [0.029, 0.146] 0.667 [0.400, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 23 -0.510 [-1.832, +0.704] EQUAL (CI ⊃ 0)
dms_matched engaged 22 +2.197 [-0.280, +3.641] EQUAL (CI ⊃ 0)
dms_matched pooled 45 +0.865 [-0.661, +2.296] EQUAL (CI ⊃ 0)
low10 manual 2 +0.819 [-0.831, +2.094] EQUAL (CI ⊃ 0)
low10 engaged 6 -0.181 [-3.954, +2.135] EQUAL (CI ⊃ 0)
low10 pooled 8 +0.311 [-1.466, +1.819] EQUAL (CI ⊃ 0)
same_recall manual 9 -4.010 [-5.553, -2.816] DMS EARLIER
same_recall engaged 13 -1.813 [-5.367, +0.444] EQUAL (CI ⊃ 0)
same_recall pooled 22 -2.894 [-4.883, -0.973] DMS EARLIER
low1 manual 0 -0.665 [-1.889, +0.278] EQUAL (CI ⊃ 0)
low1 engaged 3 -2.309 [-4.472, -1.206] DMS EARLIER
low1 pooled 3 -1.500 [-2.634, -0.731] DMS EARLIER
low3 manual 0 -0.113 [-1.588, +1.027] EQUAL (CI ⊃ 0)
low3 engaged 5 -1.187 [-4.448, +0.615] EQUAL (CI ⊃ 0)
low3 pooled 5 -0.659 [-2.237, +0.513] EQUAL (CI ⊃ 0)
low5 manual 0 +0.284 [-1.407, +1.686] EQUAL (CI ⊃ 0)
low5 engaged 5 -0.725 [-4.416, +1.418] EQUAL (CI ⊃ 0)
low5 pooled 5 -0.229 [-2.046, +1.209] EQUAL (CI ⊃ 0)
dms_matched manual 22 -0.506 [-2.226, +1.218] EQUAL (CI ⊃ 0)
dms_matched engaged 24 +1.712 [-0.337, +2.870] EQUAL (CI ⊃ 0)
dms_matched pooled 46 +0.621 [-0.826, +1.835] EQUAL (CI ⊃ 0)
low10 manual 3 +0.426 [-1.179, +1.567] EQUAL (CI ⊃ 0)
low10 engaged 5 +0.175 [-3.508, +2.499] EQUAL (CI ⊃ 0)
low10 pooled 8 +0.298 [-1.637, +1.919] EQUAL (CI ⊃ 0)
same_recall manual 11 -3.310 [-5.269, -1.483] DMS EARLIER
same_recall engaged 19 -0.566 [-3.381, +1.304] EQUAL (CI ⊃ 0)
same_recall pooled 30 -1.916 [-3.868, -0.241] DMS EARLIER
low1 manual 1 -1.058 [-2.364, -0.012] DMS EARLIER
low1 engaged 2 -1.953 [-4.053, -0.822] DMS EARLIER
low1 pooled 3 -1.513 [-2.760, -0.655] DMS EARLIER
low3 manual 2 -0.506 [-1.849, +0.448] EQUAL (CI ⊃ 0)
low3 engaged 4 -0.831 [-3.991, +0.967] EQUAL (CI ⊃ 0)
low3 pooled 6 -0.671 [-2.317, +0.555] EQUAL (CI ⊃ 0)
low5 manual 2 -0.110 [-1.603, +1.024] EQUAL (CI ⊃ 0)
low5 engaged 4 -0.369 [-3.927, +1.805] EQUAL (CI ⊃ 0)
low5 pooled 6 -0.241 [-2.114, +1.232] EQUAL (CI ⊃ 0)
dms_matched manual 21 -0.242 [-1.709, +1.265] EQUAL (CI ⊃ 0)
dms_matched engaged 22 +2.222 [-0.204, +3.653] EQUAL (CI ⊃ 0)
dms_matched pooled 43 +1.010 [-0.610, +2.416] EQUAL (CI ⊃ 0)
low10 manual 4 +0.535 [-1.154, +1.858] EQUAL (CI ⊃ 0)
low10 engaged 7 -0.319 [-3.608, +1.754] EQUAL (CI ⊃ 0)
low10 pooled 11 +0.102 [-1.667, +1.463] EQUAL (CI ⊃ 0)
same_recall manual 8 -3.735 [-5.359, -2.393] DMS EARLIER
same_recall engaged 12 -2.166 [-5.823, +0.260] EQUAL (CI ⊃ 0)
same_recall pooled 20 -2.938 [-4.949, -1.084] DMS EARLIER
low1 manual 1 -0.948 [-2.405, +0.141] EQUAL (CI ⊃ 0)
low1 engaged 3 -2.447 [-4.219, -1.300] DMS EARLIER
low1 pooled 4 -1.710 [-2.870, -0.849] DMS EARLIER
low3 manual 2 -0.397 [-1.916, +0.803] EQUAL (CI ⊃ 0)
low3 engaged 6 -1.325 [-4.138, +0.288] EQUAL (CI ⊃ 0)
low3 pooled 8 -0.868 [-2.346, +0.202] EQUAL (CI ⊃ 0)
low5 manual 2 -0.000 [-1.733, +1.411] EQUAL (CI ⊃ 0)
low5 engaged 6 -0.862 [-4.074, +1.088] EQUAL (CI ⊃ 0)
low5 pooled 8 -0.438 [-2.113, +0.842] EQUAL (CI ⊃ 0)

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.819 0.054 0.043 [0.035, 0.084]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_videomae2_ctx — VideoMAEv2 ViT-S frozen features + CockpitWM current-state context — 2026-09-16T05:01:07Z

Status: matched-context arm (C ii): adapted; frozen 384-d window embeddings -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_videomae2_ctx_20260916T044942Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.581 [0.385, 0.741] 0.254 [0.143, 0.370] --
dms_matched engaged head 49.318 183 / 32 0.562 [0.273, 0.706] 0.360 [0.111, 0.579] --
dms_matched pooled head -- 512 / 63 0.571 [0.400, 0.702] 0.298 [0.169, 0.446] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.896 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.119] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.048 [0.000, 0.094] 0.750 [0.500, 1.000] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 26.565 329 / 31 0.258 [0.106, 0.423] 0.500 [0.250, 0.778] --
same_recall engaged head 25.974 183 / 32 0.406 [0.154, 0.533] 0.591 [0.200, 0.815] --
same_recall pooled head -- 512 / 63 0.333 [0.192, 0.446] 0.553 [0.320, 0.763] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.916 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 2.948 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.323 329 / 31 0.581 [0.385, 0.741] 0.240 [0.135, 0.349] --
dms_matched engaged head 53.382 183 / 32 0.625 [0.400, 0.741] 0.345 [0.128, 0.548] --
dms_matched pooled head -- 512 / 63 0.603 [0.463, 0.703] 0.286 [0.165, 0.426] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.896 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.119] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.048 [0.000, 0.094] 0.750 [0.500, 1.000] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 33.232 329 / 31 0.323 [0.132, 0.500] 0.417 [0.179, 0.684] --
same_recall engaged head 38.881 183 / 32 0.500 [0.231, 0.636] 0.457 [0.138, 0.711] --
same_recall pooled head -- 512 / 63 0.413 [0.245, 0.537] 0.441 [0.220, 0.683] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.916 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 2.948 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.796 329 / 31 0.548 [0.367, 0.724] 0.258 [0.152, 0.370] --
dms_matched engaged head 48.681 183 / 32 0.562 [0.273, 0.706] 0.360 [0.111, 0.579] --
dms_matched pooled head -- 512 / 63 0.556 [0.395, 0.682] 0.302 [0.175, 0.451] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.896 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.119] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.048 [0.000, 0.094] 0.750 [0.500, 1.000] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 24.424 329 / 31 0.226 [0.069, 0.405] 0.583 [0.250, 0.900] --
same_recall engaged head 17.807 183 / 32 0.344 [0.120, 0.467] 0.786 [0.399, 1.000] --
same_recall pooled head -- 512 / 63 0.286 [0.143, 0.398] 0.692 [0.400, 0.889] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.916 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 2.948 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 23 -0.306 [-1.627, +1.082] EQUAL (CI ⊃ 0)
dms_matched engaged 21 +1.537 [-0.986, +2.995] EQUAL (CI ⊃ 0)
dms_matched pooled 44 +0.630 [-0.736, +1.849] EQUAL (CI ⊃ 0)
low10 manual 0 -0.132 [-1.728, +1.104] EQUAL (CI ⊃ 0)
low10 engaged 7 -0.069 [-3.119, +1.783] EQUAL (CI ⊃ 0)
low10 pooled 7 -0.100 [-1.836, +1.190] EQUAL (CI ⊃ 0)
same_recall manual 11 -3.929 [-5.547, -2.650] DMS EARLIER
same_recall engaged 16 -0.913 [-4.304, +1.215] EQUAL (CI ⊃ 0)
same_recall pooled 27 -2.397 [-4.561, -0.618] DMS EARLIER
low1 manual 0 -0.645 [-1.889, +0.318] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.616 [-4.448, -1.483] DMS EARLIER
low1 pooled 1 -1.646 [-2.681, -0.741] DMS EARLIER
low3 manual 0 -0.645 [-1.889, +0.318] EQUAL (CI ⊃ 0)
low3 engaged 3 -2.197 [-4.401, -1.083] DMS EARLIER
low3 pooled 3 -1.433 [-2.633, -0.602] DMS EARLIER
low5 manual 0 -0.568 [-1.878, +0.523] EQUAL (CI ⊃ 0)
low5 engaged 4 -1.634 [-3.974, -0.485] DMS EARLIER
low5 pooled 4 -1.110 [-2.388, -0.320] DMS EARLIER
dms_matched manual 22 -0.316 [-1.961, +1.610] EQUAL (CI ⊃ 0)
dms_matched engaged 24 +1.325 [-0.505, +2.418] EQUAL (CI ⊃ 0)
dms_matched pooled 46 +0.517 [-0.721, +1.609] EQUAL (CI ⊃ 0)
low10 manual 1 -0.526 [-2.007, +0.607] EQUAL (CI ⊃ 0)
low10 engaged 6 +0.287 [-2.591, +2.126] EQUAL (CI ⊃ 0)
low10 pooled 7 -0.113 [-1.826, +1.291] EQUAL (CI ⊃ 0)
same_recall manual 13 -3.542 [-5.457, -1.800] DMS EARLIER
same_recall engaged 20 +0.059 [-2.810, +1.627] EQUAL (CI ⊃ 0)
same_recall pooled 33 -1.713 [-3.739, -0.141] DMS EARLIER
low1 manual 0 -1.039 [-2.364, +0.023] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.259 [-4.047, -1.130] DMS EARLIER
low1 pooled 1 -1.659 [-2.823, -0.706] DMS EARLIER
low3 manual 0 -1.039 [-2.364, +0.023] EQUAL (CI ⊃ 0)
low3 engaged 2 -1.841 [-3.891, -0.681] DMS EARLIER
low3 pooled 2 -1.446 [-2.760, -0.512] DMS EARLIER
low5 manual 0 -0.961 [-2.354, +0.175] EQUAL (CI ⊃ 0)
low5 engaged 3 -1.278 [-3.508, -0.156] DMS EARLIER
low5 pooled 3 -1.122 [-2.567, -0.185] DMS EARLIER
dms_matched manual 21 -0.252 [-1.653, +1.287] EQUAL (CI ⊃ 0)
dms_matched engaged 20 +1.253 [-1.396, +2.817] EQUAL (CI ⊃ 0)
dms_matched pooled 41 +0.513 [-1.017, +1.769] EQUAL (CI ⊃ 0)
low10 manual 1 -0.416 [-2.121, +0.912] EQUAL (CI ⊃ 0)
low10 engaged 8 -0.206 [-2.738, +1.404] EQUAL (CI ⊃ 0)
low10 pooled 9 -0.310 [-1.917, +0.867] EQUAL (CI ⊃ 0)
same_recall manual 9 -3.655 [-5.336, -2.234] DMS EARLIER
same_recall engaged 10 -2.497 [-5.879, -0.450] DMS EARLIER
same_recall pooled 19 -3.067 [-4.898, -1.432] DMS EARLIER
low1 manual 0 -0.929 [-2.405, +0.179] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.753 [-4.276, -1.529] DMS EARLIER
low1 pooled 1 -1.856 [-3.039, -0.827] DMS EARLIER
low3 manual 0 -0.929 [-2.405, +0.179] EQUAL (CI ⊃ 0)
low3 engaged 3 -2.334 [-4.155, -1.184] DMS EARLIER
low3 pooled 3 -1.643 [-2.853, -0.728] DMS EARLIER
low5 manual 0 -0.852 [-2.386, +0.332] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.772 [-3.612, -0.784] DMS EARLIER
low5 pooled 5 -1.319 [-2.626, -0.506] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.864 0.050 0.057 [0.048, 0.100]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_l2cs_gaze_learned — L2CS-Net (Gaze360), learned read-out — 2026-09-16T05:14:34Z

Status: adapted; gaze angles -> 8 s causal summaries -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_l2cs_gaze_learned_20260916T050514Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.656 329 / 31 0.419 [0.176, 0.609] 0.110 [0.036, 0.209] --
dms_matched engaged head 49.199 183 / 32 0.531 [0.355, 0.840] 0.195 [0.095, 0.299] --
dms_matched pooled head -- 512 / 63 0.476 [0.317, 0.642] 0.146 [0.070, 0.234] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 10.001 329 / 31 0.032 [0.000, 0.107] 0.167 [0.000, 0.667] --
low10 engaged head 10.039 183 / 32 0.125 [0.000, 0.308] 0.222 [0.000, 0.471] --
low10 pooled head -- 512 / 63 0.079 [0.014, 0.151] 0.208 [0.040, 0.444] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 43.690 329 / 31 0.258 [0.091, 0.478] 0.095 [0.028, 0.200] --
same_recall engaged head 34.499 183 / 32 0.406 [0.227, 0.762] 0.220 [0.098, 0.382] --
same_recall pooled head -- 512 / 63 0.333 [0.205, 0.489] 0.147 [0.070, 0.253] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 5.018 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.019 183 / 32 0.094 [0.000, 0.273] 0.300 [0.000, 0.800] --
low5 pooled head -- 512 / 63 0.048 [0.000, 0.112] 0.250 [0.000, 0.625] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.253 329 / 31 0.452 [0.176, 0.677] 0.115 [0.035, 0.223] --
dms_matched engaged head 53.461 183 / 32 0.594 [0.448, 0.840] 0.194 [0.093, 0.293] --
dms_matched pooled head -- 512 / 63 0.524 [0.356, 0.686] 0.150 [0.069, 0.239] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 10.001 329 / 31 0.032 [0.000, 0.107] 0.167 [0.000, 0.667] --
low10 engaged head 10.039 183 / 32 0.125 [0.000, 0.308] 0.222 [0.000, 0.471] --
low10 pooled head -- 512 / 63 0.079 [0.014, 0.151] 0.208 [0.040, 0.444] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 48.533 329 / 31 0.323 [0.150, 0.525] 0.101 [0.035, 0.200] --
same_recall engaged head 45.613 183 / 32 0.500 [0.333, 0.810] 0.208 [0.099, 0.329] --
same_recall pooled head -- 512 / 63 0.413 [0.281, 0.551] 0.148 [0.072, 0.237] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 5.018 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.019 183 / 32 0.094 [0.000, 0.273] 0.300 [0.000, 0.800] --
low5 pooled head -- 512 / 63 0.048 [0.000, 0.112] 0.250 [0.000, 0.625] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.867 329 / 31 0.355 [0.156, 0.546] 0.099 [0.035, 0.190] --
dms_matched engaged head 48.761 183 / 32 0.531 [0.355, 0.840] 0.200 [0.098, 0.306] --
dms_matched pooled head -- 512 / 63 0.444 [0.302, 0.590] 0.143 [0.070, 0.230] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 10.001 329 / 31 0.032 [0.000, 0.107] 0.167 [0.000, 0.667] --
low10 engaged head 10.039 183 / 32 0.125 [0.000, 0.308] 0.222 [0.000, 0.471] --
low10 pooled head -- 512 / 63 0.079 [0.014, 0.151] 0.208 [0.040, 0.444] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 38.566 329 / 31 0.226 [0.087, 0.400] 0.108 [0.035, 0.216] --
same_recall engaged head 30.993 183 / 32 0.344 [0.200, 0.625] 0.208 [0.085, 0.355] --
same_recall pooled head -- 512 / 63 0.286 [0.173, 0.417] 0.153 [0.069, 0.264] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 5.018 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.019 183 / 32 0.094 [0.000, 0.273] 0.300 [0.000, 0.800] --
low5 pooled head -- 512 / 63 0.048 [0.000, 0.112] 0.250 [0.000, 0.625] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 16 -0.774 [-2.878, +0.761] EQUAL (CI ⊃ 0)
dms_matched engaged 19 +0.747 [-1.530, +2.285] EQUAL (CI ⊃ 0)
dms_matched pooled 35 -0.002 [-1.640, +1.065] EQUAL (CI ⊃ 0)
low10 manual 0 -0.852 [-1.847, +0.000] EQUAL (CI ⊃ 0)
low10 engaged 4 -1.706 [-3.000, -0.329] DMS EARLIER
low10 pooled 4 -1.286 [-2.157, -0.465] DMS EARLIER
same_recall manual 12 -2.577 [-4.407, -0.904] DMS EARLIER
same_recall engaged 15 -1.250 [-2.750, +0.604] EQUAL (CI ⊃ 0)
same_recall pooled 27 -1.903 [-3.160, -0.796] DMS EARLIER
low1 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 0 -2.866 [-4.733, -1.766] DMS EARLIER
low1 pooled 0 -1.889 [-2.779, -1.183] DMS EARLIER
low3 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low3 engaged 3 -2.253 [-3.745, -1.103] DMS EARLIER
low3 pooled 3 -1.578 [-2.455, -0.764] DMS EARLIER
low5 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low5 engaged 3 -2.216 [-3.736, -1.007] DMS EARLIER
low5 pooled 3 -1.559 [-2.453, -0.725] DMS EARLIER
dms_matched manual 15 -0.765 [-3.324, +1.267] EQUAL (CI ⊃ 0)
dms_matched engaged 23 +0.837 [-1.329, +2.355] EQUAL (CI ⊃ 0)
dms_matched pooled 38 +0.049 [-1.842, +1.305] EQUAL (CI ⊃ 0)
low10 manual 0 -1.245 [-2.348, -0.385] DMS EARLIER
low10 engaged 4 -1.350 [-2.562, -0.015] DMS EARLIER
low10 pooled 4 -1.298 [-2.205, -0.541] DMS EARLIER
same_recall manual 12 -1.923 [-4.567, +0.228] EQUAL (CI ⊃ 0)
same_recall engaged 19 -0.622 [-2.514, +0.996] EQUAL (CI ⊃ 0)
same_recall pooled 31 -1.262 [-3.081, +0.120] EQUAL (CI ⊃ 0)
low1 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low1 engaged 0 -2.509 [-4.168, -1.400] DMS EARLIER
low1 pooled 0 -1.902 [-2.865, -1.162] DMS EARLIER
low3 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low3 engaged 3 -1.897 [-3.319, -0.707] DMS EARLIER
low3 pooled 3 -1.590 [-2.473, -0.827] DMS EARLIER
low5 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low5 engaged 3 -1.859 [-3.306, -0.657] DMS EARLIER
low5 pooled 3 -1.571 [-2.470, -0.787] DMS EARLIER
dms_matched manual 15 -0.761 [-2.684, +0.794] EQUAL (CI ⊃ 0)
dms_matched engaged 19 +0.747 [-1.505, +2.264] EQUAL (CI ⊃ 0)
dms_matched pooled 34 +0.005 [-1.550, +1.106] EQUAL (CI ⊃ 0)
low10 manual 0 -1.135 [-2.404, -0.238] DMS EARLIER
low10 engaged 4 -1.844 [-2.975, -0.268] DMS EARLIER
low10 pooled 4 -1.495 [-2.416, -0.638] DMS EARLIER
same_recall manual 11 -2.381 [-4.005, -0.643] DMS EARLIER
same_recall engaged 10 -2.097 [-3.705, +0.176] EQUAL (CI ⊃ 0)
same_recall pooled 21 -2.237 [-3.395, -1.036] DMS EARLIER
low1 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low1 engaged 0 -3.003 [-4.405, -1.729] DMS EARLIER
low1 pooled 0 -2.098 [-3.061, -1.268] DMS EARLIER
low3 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low3 engaged 3 -2.391 [-3.581, -1.017] DMS EARLIER
low3 pooled 3 -1.787 [-2.711, -0.932] DMS EARLIER
low5 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low5 engaged 3 -2.353 [-3.552, -0.926] DMS EARLIER
low5 pooled 3 -1.768 [-2.705, -0.895] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.798 0.059 0.036 [0.025, 0.089]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_l2cs_gaze_ctx — L2CS-Net (Gaze360) + CockpitWM current-state context — 2026-09-16T05:29:55Z

Status: matched-context arm (C ii): adapted; gaze angles -> 8 s causal summaries -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_l2cs_gaze_ctx_20260916T051914Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.516 [0.323, 0.750] 0.193 [0.108, 0.283] --
dms_matched engaged head 49.279 183 / 32 0.250 [0.092, 0.520] 0.211 [0.083, 0.343] --
dms_matched pooled head -- 512 / 63 0.381 [0.218, 0.583] 0.198 [0.119, 0.291] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.966 329 / 31 0.097 [0.000, 0.235] 0.750 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.105] 0.500 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.125] 0.667 [0.250, 1.000] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 34.355 329 / 31 0.258 [0.118, 0.445] 0.320 [0.158, 0.500] --
same_recall engaged head 54.417 183 / 32 0.406 [0.150, 0.786] 0.224 [0.089, 0.351] --
same_recall pooled head -- 512 / 63 0.333 [0.176, 0.523] 0.253 [0.135, 0.377] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.358 329 / 31 0.613 [0.407, 0.852] 0.209 [0.122, 0.303] --
dms_matched engaged head 53.422 183 / 32 0.312 [0.121, 0.600] 0.192 [0.076, 0.323] --
dms_matched pooled head -- 512 / 63 0.460 [0.290, 0.684] 0.203 [0.124, 0.299] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.966 329 / 31 0.097 [0.000, 0.235] 0.750 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.105] 0.500 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.125] 0.667 [0.250, 1.000] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 36.636 329 / 31 0.323 [0.167, 0.522] 0.333 [0.172, 0.500] --
same_recall engaged head 61.708 183 / 32 0.500 [0.290, 0.824] 0.213 [0.102, 0.325] --
same_recall pooled head -- 512 / 63 0.413 [0.277, 0.586] 0.248 [0.148, 0.355] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.796 329 / 31 0.516 [0.323, 0.750] 0.211 [0.117, 0.312] --
dms_matched engaged head 48.761 183 / 32 0.250 [0.092, 0.520] 0.216 [0.087, 0.350] --
dms_matched pooled head -- 512 / 63 0.381 [0.218, 0.583] 0.212 [0.127, 0.311] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.966 329 / 31 0.097 [0.000, 0.235] 0.750 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.105] 0.500 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.125] 0.667 [0.250, 1.000] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 24.565 329 / 31 0.226 [0.105, 0.400] 0.500 [0.273, 0.727] --
same_recall engaged head 53.541 183 / 32 0.344 [0.136, 0.667] 0.208 [0.080, 0.341] --
same_recall pooled head -- 512 / 63 0.286 [0.155, 0.438] 0.269 [0.145, 0.407] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 1.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 23 +0.171 [-1.286, +1.849] EQUAL (CI ⊃ 0)
dms_matched engaged 15 -0.609 [-2.044, +1.071] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -0.225 [-1.185, +1.051] EQUAL (CI ⊃ 0)
low10 manual 2 +0.623 [-0.413, +2.032] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.316 [-2.436, +0.092] EQUAL (CI ⊃ 0)
low10 pooled 8 -0.362 [-1.279, +0.737] EQUAL (CI ⊃ 0)
same_recall manual 13 -2.032 [-3.625, -0.041] DMS EARLIER
same_recall engaged 17 -0.075 [-1.628, +1.352] EQUAL (CI ⊃ 0)
same_recall pooled 30 -1.038 [-2.178, +0.067] EQUAL (CI ⊃ 0)
low1 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 2 -2.528 [-3.914, -1.578] DMS EARLIER
low1 pooled 2 -1.717 [-2.477, -1.106] DMS EARLIER
low3 manual 2 -0.239 [-0.756, +0.100] EQUAL (CI ⊃ 0)
low3 engaged 3 -2.306 [-3.611, -1.392] DMS EARLIER
low3 pooled 5 -1.289 [-1.876, -0.663] DMS EARLIER
low5 manual 2 +0.035 [-0.637, +0.739] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.709 [-2.772, -0.615] DMS EARLIER
low5 pooled 7 -0.851 [-1.595, -0.052] DMS EARLIER
dms_matched manual 22 +0.174 [-1.649, +2.342] EQUAL (CI ⊃ 0)
dms_matched engaged 18 -0.969 [-2.408, +0.882] EQUAL (CI ⊃ 0)
dms_matched pooled 40 -0.406 [-1.603, +1.100] EQUAL (CI ⊃ 0)
low10 manual 2 +0.229 [-0.887, +1.438] EQUAL (CI ⊃ 0)
low10 engaged 6 -0.959 [-1.922, +0.501] EQUAL (CI ⊃ 0)
low10 pooled 8 -0.375 [-1.191, +0.561] EQUAL (CI ⊃ 0)
same_recall manual 12 -1.919 [-3.934, +0.642] EQUAL (CI ⊃ 0)
same_recall engaged 23 +0.716 [-0.635, +2.062] EQUAL (CI ⊃ 0)
same_recall pooled 35 -0.581 [-1.938, +0.684] EQUAL (CI ⊃ 0)
low1 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low1 engaged 2 -2.172 [-3.404, -1.304] DMS EARLIER
low1 pooled 2 -1.730 [-2.590, -1.112] DMS EARLIER
low3 manual 2 -0.632 [-1.544, +0.067] EQUAL (CI ⊃ 0)
low3 engaged 3 -1.950 [-3.115, -1.149] DMS EARLIER
low3 pooled 5 -1.302 [-2.089, -0.615] DMS EARLIER
low5 manual 2 -0.358 [-1.252, +0.286] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.353 [-2.281, -0.252] DMS EARLIER
low5 pooled 7 -0.863 [-1.545, -0.136] DMS EARLIER
dms_matched manual 21 +0.348 [-1.361, +2.350] EQUAL (CI ⊃ 0)
dms_matched engaged 15 -1.091 [-2.500, +0.632] EQUAL (CI ⊃ 0)
dms_matched pooled 36 -0.383 [-1.528, +1.226] EQUAL (CI ⊃ 0)
low10 manual 2 +0.339 [-1.008, +1.842] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.453 [-2.431, +0.180] EQUAL (CI ⊃ 0)
low10 pooled 8 -0.571 [-1.552, +0.607] EQUAL (CI ⊃ 0)
same_recall manual 12 -2.148 [-4.162, +0.235] EQUAL (CI ⊃ 0)
same_recall engaged 16 -0.409 [-1.777, +1.152] EQUAL (CI ⊃ 0)
same_recall pooled 28 -1.265 [-2.425, +0.085] EQUAL (CI ⊃ 0)
low1 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low1 engaged 2 -2.666 [-3.704, -1.567] DMS EARLIER
low1 pooled 2 -1.927 [-2.818, -1.200] DMS EARLIER
low3 manual 2 -0.523 [-1.367, +0.037] EQUAL (CI ⊃ 0)
low3 engaged 3 -2.444 [-3.464, -1.336] DMS EARLIER
low3 pooled 5 -1.498 [-2.275, -0.706] DMS EARLIER
low5 manual 2 -0.248 [-1.248, +0.585] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.847 [-2.711, -0.579] DMS EARLIER
low5 pooled 7 -1.060 [-1.837, -0.202] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.841 0.038 0.040 [0.026, 0.100]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_openface3_learned — OpenFace 3.0, learned read-out — 2026-09-16T05:50:21Z

Status: adapted (optional baseline); full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_openface3_learned_20260916T053944Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.387 [0.259, 0.536] 0.169 [0.099, 0.268] --
dms_matched engaged head 49.398 183 / 32 0.375 [0.167, 0.667] 0.300 [0.122, 0.488] --
dms_matched pooled head -- 512 / 63 0.381 [0.234, 0.523] 0.216 [0.123, 0.344] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 10.039 183 / 32 0.031 [0.000, 0.111] 0.333 [0.000, 0.500] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.049] 0.200 [0.000, 0.500] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 49.234 329 / 31 0.258 [0.138, 0.421] 0.170 [0.089, 0.312] --
same_recall engaged head 51.270 183 / 32 0.406 [0.214, 0.667] 0.277 [0.122, 0.432] --
same_recall pooled head -- 512 / 63 0.333 [0.219, 0.452] 0.223 [0.124, 0.356] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.288 329 / 31 0.387 [0.259, 0.536] 0.164 [0.096, 0.262] --
dms_matched engaged head 53.342 183 / 32 0.531 [0.382, 0.682] 0.298 [0.143, 0.433] --
dms_matched pooled head -- 512 / 63 0.460 [0.358, 0.548] 0.223 [0.126, 0.333] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 10.039 183 / 32 0.031 [0.000, 0.111] 0.333 [0.000, 0.500] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.049] 0.200 [0.000, 0.500] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 52.744 329 / 31 0.323 [0.207, 0.476] 0.172 [0.091, 0.300] --
same_recall engaged head 52.346 183 / 32 0.500 [0.346, 0.682] 0.308 [0.154, 0.444] --
same_recall pooled head -- 512 / 63 0.413 [0.320, 0.512] 0.236 [0.133, 0.357] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.796 329 / 31 0.355 [0.222, 0.531] 0.172 [0.105, 0.269] --
dms_matched engaged head 48.800 183 / 32 0.375 [0.167, 0.667] 0.316 [0.135, 0.500] --
dms_matched pooled head -- 512 / 63 0.365 [0.224, 0.507] 0.225 [0.132, 0.353] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 10.039 183 / 32 0.031 [0.000, 0.111] 0.333 [0.000, 0.500] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.049] 0.200 [0.000, 0.500] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 38.601 329 / 31 0.226 [0.118, 0.364] 0.212 [0.115, 0.375] --
same_recall engaged head 47.526 183 / 32 0.344 [0.109, 0.667] 0.314 [0.103, 0.500] --
same_recall pooled head -- 512 / 63 0.286 [0.150, 0.422] 0.265 [0.145, 0.408] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 19 -1.839 [-3.280, -0.826] DMS EARLIER
dms_matched engaged 19 -0.419 [-2.312, +0.465] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -1.117 [-2.373, -0.390] DMS EARLIER
low10 manual 1 -0.687 [-1.880, +0.223] EQUAL (CI ⊃ 0)
low10 engaged 4 -1.628 [-3.004, +0.430] EQUAL (CI ⊃ 0)
low10 pooled 5 -1.165 [-2.049, -0.185] DMS EARLIER
same_recall manual 16 -2.284 [-3.921, -1.107] DMS EARLIER
same_recall engaged 20 +0.047 [-2.032, +1.110] EQUAL (CI ⊃ 0)
same_recall pooled 36 -1.100 [-2.642, -0.127] DMS EARLIER
low1 manual 0 -0.700 [-1.889, +0.208] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.553 [-4.123, -1.578] DMS EARLIER
low1 pooled 1 -1.641 [-2.544, -0.983] DMS EARLIER
low3 manual 0 -0.694 [-1.889, +0.223] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.528 [-4.104, -1.561] DMS EARLIER
low3 pooled 2 -1.625 [-2.530, -0.976] DMS EARLIER
low5 manual 0 -0.694 [-1.889, +0.223] EQUAL (CI ⊃ 0)
low5 engaged 2 -2.331 [-4.025, -1.016] DMS EARLIER
low5 pooled 2 -1.525 [-2.482, -0.775] DMS EARLIER
dms_matched manual 18 -1.629 [-3.082, -0.562] DMS EARLIER
dms_matched engaged 21 -0.369 [-2.200, +0.467] EQUAL (CI ⊃ 0)
dms_matched pooled 39 -0.989 [-2.198, -0.262] DMS EARLIER
low10 manual 1 -1.081 [-2.383, -0.061] DMS EARLIER
low10 engaged 4 -1.272 [-2.493, +0.667] EQUAL (CI ⊃ 0)
low10 pooled 5 -1.178 [-2.063, -0.276] DMS EARLIER
same_recall manual 16 -2.100 [-3.764, -0.825] DMS EARLIER
same_recall engaged 21 -0.450 [-2.210, +0.360] EQUAL (CI ⊃ 0)
same_recall pooled 37 -1.262 [-2.569, -0.458] DMS EARLIER
low1 manual 0 -1.094 [-2.383, -0.079] DMS EARLIER
low1 engaged 1 -2.197 [-3.590, -1.320] DMS EARLIER
low1 pooled 1 -1.654 [-2.616, -0.976] DMS EARLIER
low3 manual 0 -1.087 [-2.383, -0.067] DMS EARLIER
low3 engaged 2 -2.172 [-3.561, -1.293] DMS EARLIER
low3 pooled 2 -1.638 [-2.611, -0.962] DMS EARLIER
low5 manual 0 -1.087 [-2.383, -0.067] DMS EARLIER
low5 engaged 2 -1.975 [-3.495, -0.794] DMS EARLIER
low5 pooled 2 -1.538 [-2.603, -0.766] DMS EARLIER
dms_matched manual 18 -1.587 [-2.852, -0.625] DMS EARLIER
dms_matched engaged 19 -0.500 [-2.311, +0.316] EQUAL (CI ⊃ 0)
dms_matched pooled 37 -1.035 [-2.267, -0.312] DMS EARLIER
low10 manual 1 -0.971 [-2.392, +0.086] EQUAL (CI ⊃ 0)
low10 engaged 4 -1.766 [-3.038, +0.550] EQUAL (CI ⊃ 0)
low10 pooled 5 -1.375 [-2.366, -0.305] DMS EARLIER
same_recall manual 13 -3.348 [-4.528, -2.215] DMS EARLIER
same_recall engaged 17 -0.744 [-2.321, +0.075] EQUAL (CI ⊃ 0)
same_recall pooled 30 -2.025 [-3.305, -1.130] DMS EARLIER
low1 manual 0 -0.984 [-2.405, +0.074] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.691 [-4.011, -1.591] DMS EARLIER
low1 pooled 1 -1.851 [-2.846, -1.092] DMS EARLIER
low3 manual 0 -0.977 [-2.405, +0.086] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.666 [-3.985, -1.563] DMS EARLIER
low3 pooled 2 -1.835 [-2.832, -1.069] DMS EARLIER
low5 manual 0 -0.977 [-2.405, +0.086] EQUAL (CI ⊃ 0)
low5 engaged 2 -2.469 [-3.946, -1.028] DMS EARLIER
low5 pooled 2 -1.735 [-2.814, -0.851] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.864 0.046 0.030 [0.027, 0.080]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_openface3_ctx — OpenFace 3.0 + CockpitWM current-state context — 2026-09-16T06:03:41Z

Status: matched-context arm (C ii): adapted (optional baseline); full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_openface3_ctx_20260916T055450Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.387 [0.263, 0.546] 0.200 [0.121, 0.314] --
dms_matched engaged head 49.318 183 / 32 0.375 [0.196, 0.708] 0.353 [0.176, 0.516] --
dms_matched pooled head -- 512 / 63 0.381 [0.244, 0.550] 0.255 [0.161, 0.368] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.966 329 / 31 0.065 [0.000, 0.152] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.192] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.014, 0.143] 0.571 [0.221, 1.000] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 41.654 329 / 31 0.258 [0.121, 0.424] 0.308 [0.160, 0.464] --
same_recall engaged head 54.218 183 / 32 0.406 [0.212, 0.750] 0.289 [0.149, 0.424] --
same_recall pooled head -- 512 / 63 0.333 [0.205, 0.509] 0.296 [0.192, 0.415] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 5.019 183 / 32 0.031 [0.000, 0.143] 0.500 [0.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.065] 0.500 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.323 329 / 31 0.419 [0.308, 0.552] 0.203 [0.118, 0.308] --
dms_matched engaged head 53.382 183 / 32 0.375 [0.196, 0.708] 0.279 [0.139, 0.414] --
dms_matched pooled head -- 512 / 63 0.397 [0.256, 0.561] 0.234 [0.143, 0.337] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.966 329 / 31 0.065 [0.000, 0.152] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.192] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.014, 0.143] 0.571 [0.221, 1.000] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 46.638 329 / 31 0.323 [0.187, 0.500] 0.238 [0.136, 0.375] --
same_recall engaged head 60.473 183 / 32 0.500 [0.312, 0.808] 0.242 [0.132, 0.345] --
same_recall pooled head -- 512 / 63 0.413 [0.281, 0.557] 0.241 [0.154, 0.342] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 5.019 183 / 32 0.031 [0.000, 0.143] 0.500 [0.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.065] 0.500 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.796 329 / 31 0.387 [0.263, 0.546] 0.211 [0.125, 0.327] --
dms_matched engaged head 48.761 183 / 32 0.375 [0.196, 0.708] 0.364 [0.184, 0.538] --
dms_matched pooled head -- 512 / 63 0.381 [0.244, 0.550] 0.267 [0.167, 0.386] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.966 329 / 31 0.065 [0.000, 0.152] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.062 [0.000, 0.192] 0.667 [0.500, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.014, 0.143] 0.571 [0.221, 1.000] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 32.355 329 / 31 0.226 [0.111, 0.368] 0.350 [0.210, 0.533] --
same_recall engaged head 45.733 183 / 32 0.344 [0.140, 0.708] 0.379 [0.174, 0.591] --
same_recall pooled head -- 512 / 63 0.286 [0.151, 0.449] 0.367 [0.227, 0.520] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low5 engaged head 5.019 183 / 32 0.031 [0.000, 0.143] 0.500 [0.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.065] 0.500 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 21 -0.703 [-1.958, +0.594] EQUAL (CI ⊃ 0)
dms_matched engaged 17 +0.244 [-1.280, +1.456] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -0.222 [-1.317, +0.682] EQUAL (CI ⊃ 0)
low10 manual 3 -0.187 [-0.738, +0.250] EQUAL (CI ⊃ 0)
low10 engaged 7 -0.878 [-2.011, +0.547] EQUAL (CI ⊃ 0)
low10 pooled 10 -0.538 [-1.311, +0.180] EQUAL (CI ⊃ 0)
same_recall manual 19 -1.606 [-3.022, +0.210] EQUAL (CI ⊃ 0)
same_recall engaged 18 +0.619 [-1.031, +2.014] EQUAL (CI ⊃ 0)
same_recall pooled 37 -0.476 [-1.764, +0.677] EQUAL (CI ⊃ 0)
low1 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 2 -2.519 [-3.921, -1.553] DMS EARLIER
low1 pooled 2 -1.713 [-2.464, -1.100] DMS EARLIER
low3 manual 1 -0.761 [-1.640, +0.000] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.438 [-3.841, -1.479] DMS EARLIER
low3 pooled 3 -1.613 [-2.318, -1.004] DMS EARLIER
low5 manual 1 -0.755 [-1.634, +0.000] EQUAL (CI ⊃ 0)
low5 engaged 4 -2.147 [-3.119, -1.288] DMS EARLIER
low5 pooled 5 -1.462 [-2.054, -0.892] DMS EARLIER
dms_matched manual 20 -0.694 [-2.644, +1.209] EQUAL (CI ⊃ 0)
dms_matched engaged 20 -0.194 [-1.479, +1.013] EQUAL (CI ⊃ 0)
dms_matched pooled 40 -0.440 [-1.685, +0.664] EQUAL (CI ⊃ 0)
low10 manual 4 -0.581 [-1.280, -0.119] DMS EARLIER
low10 engaged 7 -0.522 [-1.468, +0.809] EQUAL (CI ⊃ 0)
low10 pooled 11 -0.551 [-1.251, +0.066] EQUAL (CI ⊃ 0)
same_recall manual 19 -1.387 [-3.080, +0.878] EQUAL (CI ⊃ 0)
same_recall engaged 23 +1.450 [+0.317, +2.538] HEAD EARLIER
same_recall pooled 42 +0.054 [-1.304, +1.235] EQUAL (CI ⊃ 0)
low1 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low1 engaged 2 -2.163 [-3.391, -1.283] DMS EARLIER
low1 pooled 2 -1.725 [-2.605, -1.112] DMS EARLIER
low3 manual 1 -1.155 [-2.125, -0.366] DMS EARLIER
low3 engaged 2 -2.081 [-3.239, -1.254] DMS EARLIER
low3 pooled 3 -1.625 [-2.405, -1.059] DMS EARLIER
low5 manual 1 -1.148 [-2.116, -0.363] DMS EARLIER
low5 engaged 4 -1.791 [-2.513, -1.013] DMS EARLIER
low5 pooled 5 -1.475 [-1.990, -1.019] DMS EARLIER
dms_matched manual 20 -0.494 [-1.746, +1.208] EQUAL (CI ⊃ 0)
dms_matched engaged 17 +0.200 [-1.335, +1.434] EQUAL (CI ⊃ 0)
dms_matched pooled 37 -0.141 [-1.230, +0.898] EQUAL (CI ⊃ 0)
low10 manual 4 -0.471 [-1.162, +0.000] EQUAL (CI ⊃ 0)
low10 engaged 7 -1.016 [-2.141, +0.635] EQUAL (CI ⊃ 0)
low10 pooled 11 -0.748 [-1.609, +0.056] EQUAL (CI ⊃ 0)
same_recall manual 15 -2.187 [-3.587, -0.381] DMS EARLIER
same_recall engaged 17 +0.025 [-1.446, +1.176] EQUAL (CI ⊃ 0)
same_recall pooled 32 -1.063 [-2.352, +0.056] EQUAL (CI ⊃ 0)
low1 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low1 engaged 2 -2.656 [-3.721, -1.567] DMS EARLIER
low1 pooled 2 -1.922 [-2.809, -1.202] DMS EARLIER
low3 manual 1 -1.045 [-2.194, -0.208] DMS EARLIER
low3 engaged 2 -2.575 [-3.618, -1.503] DMS EARLIER
low3 pooled 3 -1.822 [-2.641, -1.096] DMS EARLIER
low5 manual 1 -1.039 [-2.173, -0.206] DMS EARLIER
low5 engaged 4 -2.284 [-3.043, -1.168] DMS EARLIER
low5 pooled 5 -1.671 [-2.268, -1.036] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.887 0.033 0.044 [0.032, 0.090]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_ov_gaze_learned — Intel OMZ head-pose / gaze / eye pipeline, learned read-out — 2026-09-16T06:18:03Z

Status: adapted; assembled from Intel OMZ components; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_ov_gaze_learned_20260916T060903Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.161 [0.043, 0.308] 0.076 [0.020, 0.143] --
dms_matched engaged head 49.279 183 / 32 0.312 [0.200, 0.476] 0.323 [0.120, 0.531] --
dms_matched pooled head -- 512 / 63 0.238 [0.158, 0.317] 0.155 [0.082, 0.260] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 9.959 183 / 32 0.031 [0.000, 0.118] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.167 [0.000, 0.500] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 58.253 329 / 31 0.258 [0.114, 0.462] 0.108 [0.052, 0.179] --
same_recall engaged head 50.593 183 / 32 0.406 [0.267, 0.583] 0.361 [0.133, 0.571] --
same_recall pooled head -- 512 / 63 0.333 [0.229, 0.455] 0.191 [0.104, 0.310] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.051] 1.000 [1.000, 1.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.059 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.500 [0.500, 0.500] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.253 329 / 31 0.194 [0.067, 0.368] 0.090 [0.033, 0.164] --
dms_matched engaged head 53.382 183 / 32 0.438 [0.304, 0.632] 0.350 [0.146, 0.538] --
dms_matched pooled head -- 512 / 63 0.317 [0.233, 0.429] 0.187 [0.103, 0.308] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 9.959 183 / 32 0.031 [0.000, 0.118] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.167 [0.000, 0.500] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 63.763 329 / 31 0.323 [0.156, 0.583] 0.103 [0.057, 0.174] --
same_recall engaged head 58.800 183 / 32 0.500 [0.346, 0.677] 0.286 [0.113, 0.446] --
same_recall pooled head -- 512 / 63 0.413 [0.310, 0.542] 0.170 [0.096, 0.274] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.051] 1.000 [1.000, 1.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.059 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.500 [0.500, 0.500] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.831 329 / 31 0.161 [0.043, 0.308] 0.086 [0.023, 0.159] --
dms_matched engaged head 48.721 183 / 32 0.312 [0.200, 0.476] 0.333 [0.125, 0.537] --
dms_matched pooled head -- 512 / 63 0.238 [0.158, 0.317] 0.170 [0.091, 0.279] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 10.036 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged head 9.959 183 / 32 0.031 [0.000, 0.118] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.167 [0.000, 0.500] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 58.183 329 / 31 0.226 [0.088, 0.417] 0.096 [0.042, 0.173] --
same_recall engaged head 49.876 183 / 32 0.344 [0.214, 0.480] 0.344 [0.120, 0.556] --
same_recall pooled head -- 512 / 63 0.286 [0.179, 0.408] 0.171 [0.086, 0.294] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.051] 1.000 [1.000, 1.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged head 5.059 183 / 32 0.031 [0.000, 0.118] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.051] 0.500 [0.500, 0.500] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 15 -1.968 [-3.666, +0.017] EQUAL (CI ⊃ 0)
dms_matched engaged 15 -0.300 [-2.132, +0.822] EQUAL (CI ⊃ 0)
dms_matched pooled 30 -1.121 [-2.211, +0.016] EQUAL (CI ⊃ 0)
low10 manual 1 -0.019 [-1.160, +1.208] EQUAL (CI ⊃ 0)
low10 engaged 4 -1.884 [-3.506, -0.648] DMS EARLIER
low10 pooled 5 -0.967 [-1.838, -0.077] DMS EARLIER
same_recall manual 17 -1.684 [-3.337, +0.138] EQUAL (CI ⊃ 0)
same_recall engaged 17 +0.181 [-1.892, +1.355] EQUAL (CI ⊃ 0)
same_recall pooled 34 -0.737 [-1.941, +0.431] EQUAL (CI ⊃ 0)
low1 manual 1 -0.684 [-1.548, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 2 -2.509 [-4.127, -1.402] DMS EARLIER
low1 pooled 3 -1.611 [-2.377, -0.869] DMS EARLIER
low3 manual 1 -0.677 [-1.548, +0.000] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.469 [-4.092, -1.339] DMS EARLIER
low3 pooled 3 -1.587 [-2.346, -0.847] DMS EARLIER
low5 manual 1 -0.613 [-1.472, +0.057] EQUAL (CI ⊃ 0)
low5 engaged 4 -2.034 [-3.604, -0.953] DMS EARLIER
low5 pooled 5 -1.335 [-2.099, -0.619] DMS EARLIER
dms_matched manual 15 -1.835 [-3.610, +0.353] EQUAL (CI ⊃ 0)
dms_matched engaged 20 +0.188 [-2.071, +1.438] EQUAL (CI ⊃ 0)
dms_matched pooled 35 -0.808 [-2.135, +0.481] EQUAL (CI ⊃ 0)
low10 manual 1 -0.413 [-1.483, +0.670] EQUAL (CI ⊃ 0)
low10 engaged 4 -1.528 [-2.809, -0.526] DMS EARLIER
low10 pooled 5 -0.979 [-1.857, -0.156] DMS EARLIER
same_recall manual 17 -1.229 [-3.050, +0.837] EQUAL (CI ⊃ 0)
same_recall engaged 23 +1.275 [-0.826, +2.453] EQUAL (CI ⊃ 0)
same_recall pooled 40 +0.043 [-1.379, +1.419] EQUAL (CI ⊃ 0)
low1 manual 1 -1.077 [-2.045, -0.300] DMS EARLIER
low1 engaged 2 -2.153 [-3.531, -1.261] DMS EARLIER
low1 pooled 3 -1.624 [-2.440, -0.967] DMS EARLIER
low3 manual 1 -1.071 [-2.032, -0.300] DMS EARLIER
low3 engaged 2 -2.112 [-3.474, -1.206] DMS EARLIER
low3 pooled 3 -1.600 [-2.408, -0.935] DMS EARLIER
low5 manual 1 -1.006 [-1.980, -0.191] DMS EARLIER
low5 engaged 4 -1.678 [-2.917, -0.881] DMS EARLIER
low5 pooled 5 -1.348 [-2.211, -0.698] DMS EARLIER
dms_matched manual 13 -1.806 [-3.535, +0.227] EQUAL (CI ⊃ 0)
dms_matched engaged 15 -0.275 [-2.095, +0.836] EQUAL (CI ⊃ 0)
dms_matched pooled 28 -1.029 [-2.059, -0.004] DMS EARLIER
low10 manual 1 -0.303 [-1.629, +0.956] EQUAL (CI ⊃ 0)
low10 engaged 4 -2.022 [-3.147, -0.731] DMS EARLIER
low10 pooled 5 -1.176 [-2.078, -0.173] DMS EARLIER
same_recall manual 16 -1.223 [-3.014, +0.748] EQUAL (CI ⊃ 0)
same_recall engaged 16 +0.025 [-2.088, +1.187] EQUAL (CI ⊃ 0)
same_recall pooled 32 -0.589 [-1.787, +0.575] EQUAL (CI ⊃ 0)
low1 manual 1 -0.968 [-2.104, -0.166] DMS EARLIER
low1 engaged 2 -2.647 [-3.765, -1.513] DMS EARLIER
low1 pooled 3 -1.821 [-2.632, -1.058] DMS EARLIER
low3 manual 1 -0.961 [-2.088, -0.153] DMS EARLIER
low3 engaged 2 -2.606 [-3.693, -1.450] DMS EARLIER
low3 pooled 3 -1.797 [-2.597, -1.022] DMS EARLIER
low5 manual 1 -0.897 [-2.027, -0.020] DMS EARLIER
low5 engaged 4 -2.172 [-3.212, -1.041] DMS EARLIER
low5 pooled 5 -1.544 [-2.404, -0.782] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 0.961 0.071 0.062 [0.042, 0.107]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_ov_gaze_ctx — Intel OMZ head-pose / gaze / eye pipeline + CockpitWM current-state context — 2026-09-16T06:31:13Z

Status: matched-context arm (C ii): adapted; assembled from Intel OMZ components; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_ov_gaze_ctx_20260916T062317Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.551 329 / 31 0.290 [0.135, 0.478] 0.158 [0.071, 0.256] --
dms_matched engaged head 49.358 183 / 32 0.375 [0.217, 0.546] 0.333 [0.125, 0.520] --
dms_matched pooled head -- 512 / 63 0.333 [0.229, 0.447] 0.226 [0.131, 0.345] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.931 329 / 31 0.097 [0.000, 0.219] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.143] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.147] 0.444 [0.142, 0.778] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 37.303 329 / 31 0.258 [0.108, 0.444] 0.296 [0.154, 0.462] --
same_recall engaged head 52.904 183 / 32 0.406 [0.261, 0.609] 0.302 [0.129, 0.457] --
same_recall pooled head -- 512 / 63 0.333 [0.229, 0.452] 0.300 [0.177, 0.427] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.107] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.031 [0.000, 0.143] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.091] 0.667 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.358 329 / 31 0.290 [0.135, 0.478] 0.155 [0.070, 0.253] --
dms_matched engaged head 53.382 183 / 32 0.438 [0.304, 0.632] 0.286 [0.123, 0.438] --
dms_matched pooled head -- 512 / 63 0.365 [0.283, 0.466] 0.215 [0.130, 0.321] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.931 329 / 31 0.097 [0.000, 0.219] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.143] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.147] 0.444 [0.142, 0.778] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 57.621 329 / 31 0.323 [0.167, 0.500] 0.152 [0.072, 0.244] --
same_recall engaged head 58.640 183 / 32 0.500 [0.364, 0.700] 0.254 [0.113, 0.387] --
same_recall pooled head -- 512 / 63 0.413 [0.333, 0.512] 0.202 [0.120, 0.298] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.107] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.031 [0.000, 0.143] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.091] 0.667 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.867 329 / 31 0.290 [0.135, 0.478] 0.167 [0.077, 0.264] --
dms_matched engaged head 48.761 183 / 32 0.375 [0.217, 0.546] 0.333 [0.125, 0.520] --
dms_matched pooled head -- 512 / 63 0.333 [0.229, 0.447] 0.233 [0.135, 0.357] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.931 329 / 31 0.097 [0.000, 0.219] 0.500 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.031 [0.000, 0.143] 0.333 [0.000, 1.000] --
low10 pooled head -- 512 / 63 0.063 [0.015, 0.147] 0.444 [0.142, 0.778] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 37.303 329 / 31 0.258 [0.108, 0.444] 0.296 [0.154, 0.462] --
same_recall engaged head 47.805 183 / 32 0.344 [0.167, 0.500] 0.344 [0.111, 0.526] --
same_recall pooled head -- 512 / 63 0.302 [0.192, 0.405] 0.322 [0.176, 0.467] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low3 engaged head 2.988 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 5.018 329 / 31 0.032 [0.000, 0.107] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.031 [0.000, 0.143] 1.000 [1.000, 1.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.091] 0.667 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 22 -0.435 [-2.352, +1.873] EQUAL (CI ⊃ 0)
dms_matched engaged 18 -0.197 [-1.953, +1.716] EQUAL (CI ⊃ 0)
dms_matched pooled 40 -0.314 [-1.444, +1.122] EQUAL (CI ⊃ 0)
low10 manual 3 +0.639 [+0.036, +1.431] HEAD EARLIER
low10 engaged 6 -1.431 [-2.645, -0.039] DMS EARLIER
low10 pooled 9 -0.413 [-1.227, +0.414] EQUAL (CI ⊃ 0)
same_recall manual 14 -1.919 [-3.535, +0.113] EQUAL (CI ⊃ 0)
same_recall engaged 19 +0.762 [-0.594, +2.065] EQUAL (CI ⊃ 0)
same_recall pooled 33 -0.557 [-1.811, +0.610] EQUAL (CI ⊃ 0)
low1 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 0 -2.866 [-4.733, -1.766] DMS EARLIER
low1 pooled 0 -1.889 [-2.779, -1.183] DMS EARLIER
low3 manual 1 -0.781 [-1.661, +0.000] EQUAL (CI ⊃ 0)
low3 engaged 4 -2.119 [-3.648, -1.190] DMS EARLIER
low3 pooled 5 -1.460 [-2.310, -0.782] DMS EARLIER
low5 manual 2 -0.216 [-1.041, +0.492] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.872 [-3.023, -0.944] DMS EARLIER
low5 pooled 7 -1.057 [-1.739, -0.426] DMS EARLIER
dms_matched manual 21 -0.426 [-2.510, +2.184] EQUAL (CI ⊃ 0)
dms_matched engaged 21 +0.531 [-0.829, +1.872] EQUAL (CI ⊃ 0)
dms_matched pooled 42 +0.060 [-1.214, +1.409] EQUAL (CI ⊃ 0)
low10 manual 4 +0.245 [-0.379, +0.835] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.075 [-1.991, -0.024] DMS EARLIER
low10 pooled 10 -0.425 [-1.208, +0.301] EQUAL (CI ⊃ 0)
same_recall manual 22 -0.219 [-2.164, +2.206] EQUAL (CI ⊃ 0)
same_recall engaged 22 +1.100 [-0.367, +2.293] EQUAL (CI ⊃ 0)
same_recall pooled 44 +0.451 [-0.783, +1.727] EQUAL (CI ⊃ 0)
low1 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low1 engaged 0 -2.509 [-4.168, -1.400] DMS EARLIER
low1 pooled 0 -1.902 [-2.865, -1.162] DMS EARLIER
low3 manual 1 -1.174 [-2.176, -0.373] DMS EARLIER
low3 engaged 4 -1.763 [-3.085, -0.927] DMS EARLIER
low3 pooled 5 -1.473 [-2.436, -0.805] DMS EARLIER
low5 manual 3 -0.610 [-1.366, +0.090] EQUAL (CI ⊃ 0)
low5 engaged 5 -1.516 [-2.413, -0.769] DMS EARLIER
low5 pooled 8 -1.070 [-1.734, -0.504] DMS EARLIER
dms_matched manual 20 -0.119 [-2.165, +2.442] EQUAL (CI ⊃ 0)
dms_matched engaged 18 -0.191 [-1.964, +1.774] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -0.156 [-1.375, +1.360] EQUAL (CI ⊃ 0)
low10 manual 4 +0.355 [-0.412, +1.168] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.569 [-2.492, -0.330] DMS EARLIER
low10 pooled 10 -0.622 [-1.562, +0.307] EQUAL (CI ⊃ 0)
same_recall manual 13 -1.458 [-3.148, +0.796] EQUAL (CI ⊃ 0)
same_recall engaged 18 -0.241 [-2.030, +1.682] EQUAL (CI ⊃ 0)
same_recall pooled 31 -0.840 [-2.079, +0.477] EQUAL (CI ⊃ 0)
low1 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low1 engaged 0 -3.003 [-4.405, -1.729] DMS EARLIER
low1 pooled 0 -2.098 [-3.061, -1.268] DMS EARLIER
low3 manual 1 -1.065 [-2.215, -0.218] DMS EARLIER
low3 engaged 4 -2.256 [-3.388, -1.115] DMS EARLIER
low3 pooled 5 -1.670 [-2.651, -0.900] DMS EARLIER
low5 manual 3 -0.500 [-1.478, +0.207] EQUAL (CI ⊃ 0)
low5 engaged 5 -2.009 [-2.851, -0.936] DMS EARLIER
low5 pooled 8 -1.267 [-2.034, -0.491] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 1.189 0.056 0.049 [0.032, 0.103]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_mp_face_ctx — MediaPipe Face Landmarker + CockpitWM current-state context — 2026-09-16T06:51:18Z

Status: matched-context arm (C ii): adapted; blendshapes + head pose -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_mp_face_ctx_20260916T064105Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.586 324 / 31 0.581 [0.432, 0.778] 0.182 [0.110, 0.262] --
dms_matched engaged head 49.358 179 / 32 0.438 [0.276, 0.667] 0.389 [0.200, 0.517] --
dms_matched pooled head -- 503 / 63 0.508 [0.386, 0.667] 0.237 [0.154, 0.319] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.966 324 / 31 0.065 [0.000, 0.208] 0.500 [0.000, 1.000] --
low10 engaged head 10.039 179 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 503 / 63 0.032 [0.000, 0.087] 0.400 [0.000, 1.000] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 30.355 324 / 31 0.258 [0.125, 0.421] 0.348 [0.156, 0.600] --
same_recall engaged head 47.247 179 / 32 0.406 [0.250, 0.600] 0.382 [0.207, 0.517] --
same_recall pooled head -- 503 / 63 0.333 [0.214, 0.453] 0.368 [0.210, 0.517] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 1.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 1.036 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 324 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 503 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.358 324 / 31 0.581 [0.432, 0.778] 0.176 [0.107, 0.254] --
dms_matched engaged head 53.422 179 / 32 0.500 [0.333, 0.759] 0.340 [0.175, 0.475] --
dms_matched pooled head -- 503 / 63 0.540 [0.422, 0.690] 0.228 [0.151, 0.310] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.966 324 / 31 0.065 [0.000, 0.208] 0.500 [0.000, 1.000] --
low10 engaged head 10.039 179 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 503 / 63 0.032 [0.000, 0.087] 0.400 [0.000, 1.000] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 35.689 324 / 31 0.323 [0.178, 0.500] 0.333 [0.160, 0.536] --
same_recall engaged head 50.752 179 / 32 0.500 [0.333, 0.759] 0.400 [0.212, 0.545] --
same_recall pooled head -- 503 / 63 0.413 [0.290, 0.543] 0.371 [0.226, 0.529] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 1.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 1.036 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 324 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 503 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.831 324 / 31 0.581 [0.432, 0.778] 0.202 [0.122, 0.286] --
dms_matched engaged head 48.761 179 / 32 0.438 [0.276, 0.667] 0.389 [0.200, 0.517] --
dms_matched pooled head -- 503 / 63 0.508 [0.386, 0.667] 0.256 [0.175, 0.341] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.966 324 / 31 0.065 [0.000, 0.208] 0.500 [0.000, 1.000] --
low10 engaged head 10.039 179 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 503 / 63 0.032 [0.000, 0.087] 0.400 [0.000, 1.000] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 29.127 324 / 31 0.226 [0.087, 0.417] 0.333 [0.125, 0.583] --
same_recall engaged head 43.223 179 / 32 0.344 [0.205, 0.600] 0.393 [0.250, 0.571] --
same_recall pooled head -- 503 / 63 0.286 [0.174, 0.418] 0.367 [0.222, 0.528] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 1.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low1 engaged head 1.036 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 324 / 31 0.032 [0.000, 0.133] 1.000 [1.000, 1.000] --
low3 engaged head 3.028 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 503 / 63 0.016 [0.000, 0.061] 1.000 [1.000, 1.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 324 / 31 0.032 [0.000, 0.133] 0.500 [0.000, 1.000] --
low5 engaged head 5.019 179 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 503 / 63 0.016 [0.000, 0.061] 0.500 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 25 +1.361 [+0.212, +2.333] HEAD EARLIER
dms_matched engaged 21 +1.012 [-0.987, +2.292] EQUAL (CI ⊃ 0)
dms_matched pooled 46 +1.184 [+0.154, +2.170] HEAD EARLIER
low10 manual 3 +0.506 [+0.045, +1.319] HEAD EARLIER
low10 engaged 3 -2.159 [-3.450, -1.085] DMS EARLIER
low10 pooled 6 -0.848 [-1.510, -0.046] DMS EARLIER
same_recall manual 16 -2.806 [-3.865, -1.645] DMS EARLIER
same_recall engaged 20 +0.634 [-1.652, +2.045] EQUAL (CI ⊃ 0)
same_recall pooled 36 -1.059 [-2.659, +0.546] EQUAL (CI ⊃ 0)
low1 manual 2 -0.600 [-1.348, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 1 -2.731 [-4.248, -1.678] DMS EARLIER
low1 pooled 3 -1.683 [-2.343, -0.974] DMS EARLIER
low3 manual 2 -0.265 [-0.780, +0.100] EQUAL (CI ⊃ 0)
low3 engaged 1 -2.725 [-4.225, -1.677] DMS EARLIER
low3 pooled 3 -1.514 [-2.146, -0.793] DMS EARLIER
low5 manual 3 +0.187 [-0.029, +0.573] EQUAL (CI ⊃ 0)
low5 engaged 2 -2.416 [-3.765, -1.331] DMS EARLIER
low5 pooled 5 -1.135 [-1.718, -0.344] DMS EARLIER
dms_matched manual 24 +1.352 [-0.056, +2.757] EQUAL (CI ⊃ 0)
dms_matched engaged 24 +0.953 [-1.050, +2.237] EQUAL (CI ⊃ 0)
dms_matched pooled 48 +1.149 [-0.021, +2.222] EQUAL (CI ⊃ 0)
low10 manual 3 +0.113 [-0.835, +1.078] EQUAL (CI ⊃ 0)
low10 engaged 3 -1.803 [-2.960, -0.737] DMS EARLIER
low10 pooled 6 -0.860 [-1.563, -0.009] DMS EARLIER
same_recall manual 16 -2.142 [-3.875, +0.140] EQUAL (CI ⊃ 0)
same_recall engaged 24 +0.550 [-1.100, +1.564] EQUAL (CI ⊃ 0)
same_recall pooled 40 -0.775 [-2.351, +0.633] EQUAL (CI ⊃ 0)
low1 manual 2 -0.994 [-2.015, -0.293] DMS EARLIER
low1 engaged 1 -2.375 [-3.868, -1.295] DMS EARLIER
low1 pooled 3 -1.695 [-2.487, -1.011] DMS EARLIER
low3 manual 2 -0.658 [-1.555, +0.047] EQUAL (CI ⊃ 0)
low3 engaged 1 -2.369 [-3.858, -1.277] DMS EARLIER
low3 pooled 3 -1.527 [-2.274, -0.810] DMS EARLIER
low5 manual 3 -0.206 [-1.037, +0.471] EQUAL (CI ⊃ 0)
low5 engaged 2 -2.059 [-3.209, -1.082] DMS EARLIER
low5 pooled 5 -1.148 [-1.805, -0.433] DMS EARLIER
dms_matched manual 23 +1.503 [+0.187, +2.880] HEAD EARLIER
dms_matched engaged 21 +1.012 [-0.961, +2.293] EQUAL (CI ⊃ 0)
dms_matched pooled 44 +1.254 [+0.103, +2.277] HEAD EARLIER
low10 manual 3 +0.223 [-0.642, +1.113] EQUAL (CI ⊃ 0)
low10 engaged 3 -2.297 [-3.335, -1.031] DMS EARLIER
low10 pooled 6 -1.057 [-1.861, -0.146] DMS EARLIER
same_recall manual 15 -2.365 [-3.655, -0.775] DMS EARLIER
same_recall engaged 19 -0.319 [-2.865, +1.072] EQUAL (CI ⊃ 0)
same_recall pooled 34 -1.325 [-2.906, +0.055] EQUAL (CI ⊃ 0)
low1 manual 2 -0.884 [-1.872, -0.207] DMS EARLIER
low1 engaged 1 -2.869 [-4.131, -1.626] DMS EARLIER
low1 pooled 3 -1.892 [-2.671, -1.095] DMS EARLIER
low3 manual 2 -0.548 [-1.393, +0.036] EQUAL (CI ⊃ 0)
low3 engaged 1 -2.862 [-4.115, -1.617] DMS EARLIER
low3 pooled 3 -1.724 [-2.384, -0.895] DMS EARLIER
low5 manual 3 -0.097 [-0.904, +0.455] EQUAL (CI ⊃ 0)
low5 engaged 2 -2.553 [-3.573, -1.305] DMS EARLIER
low5 pooled 5 -1.344 [-2.040, -0.458] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 1.189 0.048 0.053 [0.039, 0.108]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_ov_action_learned — Intel OMZ driver-action-recognition logits, learned read-out — 2026-09-16T07:26:58Z

Status: adapted; 9 action logits (window cache) -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_ov_action_learned_20260916T071554Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.621 329 / 31 0.290 [0.147, 0.467] 0.122 [0.060, 0.184] --
dms_matched engaged head 49.318 183 / 32 0.219 [0.000, 0.357] 0.152 [0.000, 0.319] --
dms_matched pooled head -- 512 / 63 0.254 [0.139, 0.364] 0.133 [0.063, 0.219] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 9.931 329 / 31 0.032 [0.000, 0.118] 0.250 [0.000, 0.750] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.200 [0.000, 0.600] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 52.849 329 / 31 0.258 [0.111, 0.421] 0.119 [0.052, 0.182] --
same_recall engaged head 60.114 183 / 32 0.406 [0.167, 0.571] 0.191 [0.053, 0.333] --
same_recall pooled head -- 512 / 63 0.333 [0.187, 0.462] 0.156 [0.073, 0.241] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.057] 1.000 [1.000, 1.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 5.053 329 / 31 0.032 [0.000, 0.118] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.500 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.393 329 / 31 0.290 [0.147, 0.467] 0.118 [0.057, 0.180] --
dms_matched engaged head 53.382 183 / 32 0.281 [0.045, 0.455] 0.161 [0.019, 0.316] --
dms_matched pooled head -- 512 / 63 0.286 [0.160, 0.409] 0.136 [0.064, 0.218] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 9.931 329 / 31 0.032 [0.000, 0.118] 0.250 [0.000, 0.750] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.200 [0.000, 0.600] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 61.341 329 / 31 0.323 [0.154, 0.500] 0.108 [0.044, 0.172] --
same_recall engaged head 62.265 183 / 32 0.500 [0.214, 0.694] 0.198 [0.059, 0.333] --
same_recall pooled head -- 512 / 63 0.413 [0.250, 0.559] 0.149 [0.072, 0.232] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.057] 1.000 [1.000, 1.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 5.053 329 / 31 0.032 [0.000, 0.118] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.500 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.761 329 / 31 0.290 [0.147, 0.467] 0.127 [0.062, 0.190] --
dms_matched engaged head 48.681 183 / 32 0.219 [0.000, 0.357] 0.152 [0.000, 0.319] --
dms_matched pooled head -- 512 / 63 0.254 [0.139, 0.364] 0.137 [0.065, 0.222] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 9.931 329 / 31 0.032 [0.000, 0.118] 0.250 [0.000, 0.750] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.200 [0.000, 0.600] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 51.480 329 / 31 0.226 [0.091, 0.375] 0.111 [0.042, 0.174] --
same_recall engaged head 58.282 183 / 32 0.344 [0.107, 0.514] 0.167 [0.036, 0.304] --
same_recall pooled head -- 512 / 63 0.286 [0.159, 0.404] 0.140 [0.066, 0.224] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] -- [--, --] --
low1 engaged head 0.996 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] -- [--, --] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 3.018 329 / 31 0.032 [0.000, 0.118] 1.000 [1.000, 1.000] --
low3 engaged head 3.067 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.016 [0.000, 0.057] 1.000 [1.000, 1.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 5.053 329 / 31 0.032 [0.000, 0.118] 0.500 [0.000, 1.000] --
low5 engaged head 4.980 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low5 pooled head -- 512 / 63 0.016 [0.000, 0.057] 0.500 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 19 -0.642 [-2.519, +0.851] EQUAL (CI ⊃ 0)
dms_matched engaged 17 -0.022 [-3.587, +2.020] EQUAL (CI ⊃ 0)
dms_matched pooled 36 -0.327 [-2.470, +1.237] EQUAL (CI ⊃ 0)
low10 manual 1 -0.345 [-1.447, +0.778] EQUAL (CI ⊃ 0)
low10 engaged 0 -2.556 [-4.330, -1.023] DMS EARLIER
low10 pooled 1 -1.468 [-2.420, -0.504] DMS EARLIER
same_recall manual 19 -0.771 [-2.729, +0.732] EQUAL (CI ⊃ 0)
same_recall engaged 20 +1.678 [-1.605, +3.480] EQUAL (CI ⊃ 0)
same_recall pooled 39 +0.473 [-1.678, +2.113] EQUAL (CI ⊃ 0)
low1 manual 0 -0.729 [-1.786, +0.168] EQUAL (CI ⊃ 0)
low1 engaged 0 -2.866 [-4.733, -1.766] DMS EARLIER
low1 pooled 0 -1.814 [-2.710, -1.013] DMS EARLIER
low3 manual 0 -0.697 [-1.786, +0.228] EQUAL (CI ⊃ 0)
low3 engaged 0 -2.866 [-4.733, -1.766] DMS EARLIER
low3 pooled 0 -1.798 [-2.700, -0.991] DMS EARLIER
low5 manual 0 -0.568 [-1.653, +0.428] EQUAL (CI ⊃ 0)
low5 engaged 0 -2.612 [-4.348, -1.196] DMS EARLIER
low5 pooled 0 -1.606 [-2.510, -0.732] DMS EARLIER
dms_matched manual 18 -0.613 [-2.847, +1.204] EQUAL (CI ⊃ 0)
dms_matched engaged 20 +0.131 [-3.426, +2.107] EQUAL (CI ⊃ 0)
dms_matched pooled 38 -0.235 [-2.456, +1.430] EQUAL (CI ⊃ 0)
low10 manual 2 -0.739 [-1.794, +0.325] EQUAL (CI ⊃ 0)
low10 engaged 0 -2.200 [-3.850, -0.722] DMS EARLIER
low10 pooled 2 -1.481 [-2.370, -0.573] DMS EARLIER
same_recall manual 19 -0.310 [-2.312, +1.403] EQUAL (CI ⊃ 0)
same_recall engaged 23 +1.516 [-1.153, +2.995] EQUAL (CI ⊃ 0)
same_recall pooled 42 +0.617 [-1.291, +1.982] EQUAL (CI ⊃ 0)
low1 manual 0 -1.123 [-2.347, -0.190] DMS EARLIER
low1 engaged 0 -2.509 [-4.168, -1.400] DMS EARLIER
low1 pooled 0 -1.827 [-2.774, -1.042] DMS EARLIER
low3 manual 0 -1.090 [-2.346, -0.119] DMS EARLIER
low3 engaged 0 -2.509 [-4.168, -1.400] DMS EARLIER
low3 pooled 0 -1.811 [-2.759, -1.022] DMS EARLIER
low5 manual 1 -0.961 [-2.070, -0.049] DMS EARLIER
low5 engaged 0 -2.256 [-3.879, -0.896] DMS EARLIER
low5 pooled 1 -1.619 [-2.532, -0.810] DMS EARLIER
dms_matched manual 18 -0.310 [-2.073, +1.082] EQUAL (CI ⊃ 0)
dms_matched engaged 17 +0.003 [-3.559, +2.030] EQUAL (CI ⊃ 0)
dms_matched pooled 35 -0.151 [-2.048, +1.331] EQUAL (CI ⊃ 0)
low10 manual 2 -0.629 [-1.879, +0.371] EQUAL (CI ⊃ 0)
low10 engaged 0 -2.694 [-4.142, -1.009] DMS EARLIER
low10 pooled 2 -1.678 [-2.534, -0.680] DMS EARLIER
same_recall manual 16 -0.484 [-2.447, +1.057] EQUAL (CI ⊃ 0)
same_recall engaged 20 +1.622 [-1.579, +3.400] EQUAL (CI ⊃ 0)
same_recall pooled 36 +0.586 [-1.530, +2.140] EQUAL (CI ⊃ 0)
low1 manual 0 -1.013 [-2.386, -0.010] DMS EARLIER
low1 engaged 0 -3.003 [-4.405, -1.729] DMS EARLIER
low1 pooled 0 -2.024 [-2.983, -1.143] DMS EARLIER
low3 manual 0 -0.981 [-2.365, +0.007] EQUAL (CI ⊃ 0)
low3 engaged 0 -3.003 [-4.405, -1.729] DMS EARLIER
low3 pooled 0 -2.008 [-2.966, -1.097] DMS EARLIER
low5 manual 1 -0.852 [-2.100, +0.073] EQUAL (CI ⊃ 0)
low5 engaged 0 -2.750 [-4.176, -1.163] DMS EARLIER
low5 pooled 1 -1.816 [-2.780, -0.897] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 1.255 0.080 0.093 [0.064, 0.137]
dms 20.000 0.622 0.408 [0.346, 0.459]

item dms_ov_action_ctx — Intel OMZ driver-action-recognition logits + CockpitWM current-state context — 2026-09-16T07:52:17Z

Status: matched-context arm (C ii): adapted; 9 action logits (window cache) -> common read-out; full val+test exposure (tiers A-C), tier-A read-out applied only (no refit), 3 seeds; loads dms_matched / low10 / same_recall + 1, 3, 5 alerts/h. Evaluator: eval/operating_points.py (unchanged): thresholds bisected on the full test split to the load, 60 s per-route cooldown; recall / precision on the human gold at that fixed threshold (shares of the enriched gold set, not natural rates). Source: results/operating_points/dms_ov_action_ctx_20260916T074131Z/.

load regime score alerts/h achieved n gold / pos recall [CI] precision [CI] lead median s
dms_matched manual head 55.621 329 / 31 0.452 [0.289, 0.667] 0.167 [0.095, 0.230] --
dms_matched engaged head 49.318 183 / 32 0.281 [0.105, 0.440] 0.209 [0.062, 0.357] --
dms_matched pooled head -- 512 / 63 0.365 [0.239, 0.508] 0.181 [0.098, 0.254] --
dms_matched manual dms 55.588 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
dms_matched engaged dms 49.378 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
dms_matched pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low10 manual head 10.001 329 / 31 0.065 [0.000, 0.179] 0.333 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.286 [0.000, 0.750] --
low10 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
same_recall manual head 43.620 329 / 31 0.258 [0.133, 0.406] 0.170 [0.070, 0.283] --
same_recall engaged head 53.342 183 / 32 0.406 [0.214, 0.572] 0.241 [0.089, 0.367] --
same_recall pooled head -- 512 / 63 0.333 [0.214, 0.439] 0.208 [0.103, 0.301] --
same_recall manual dms 55.553 328 / 31 0.258 [0.125, 0.421] 0.103 [0.049, 0.169] --
same_recall engaged dms 49.337 183 / 32 0.406 [0.259, 0.619] 0.619 [0.353, 0.824] --
same_recall pooled dms -- 511 / 63 0.333 [0.210, 0.470] 0.212 [0.110, 0.337] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
low5 manual head 4.983 329 / 31 0.065 [0.000, 0.179] 0.667 [0.000, 1.000] --
low5 engaged head 5.059 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.500 [0.000, 1.000] --
low5 manual dms 15.096 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.663 183 / 32 0.062 [0.000, 0.207] 0.667 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.032 [0.000, 0.085] 0.222 [0.000, 0.667] --
dms_matched manual head 56.358 329 / 31 0.452 [0.289, 0.667] 0.165 [0.095, 0.227] --
dms_matched engaged head 53.422 183 / 32 0.406 [0.214, 0.572] 0.241 [0.089, 0.367] --
dms_matched pooled head -- 512 / 63 0.429 [0.295, 0.569] 0.194 [0.108, 0.266] --
dms_matched manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
dms_matched engaged dms 53.425 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
dms_matched pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low10 manual head 10.001 329 / 31 0.065 [0.000, 0.179] 0.333 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.286 [0.000, 0.750] --
low10 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low10 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
same_recall manual head 47.620 329 / 31 0.323 [0.176, 0.483] 0.167 [0.071, 0.266] --
same_recall engaged head 59.636 183 / 32 0.500 [0.276, 0.647] 0.232 [0.082, 0.359] --
same_recall pooled head -- 512 / 63 0.413 [0.277, 0.528] 0.202 [0.101, 0.292] --
same_recall manual dms 56.329 328 / 31 0.323 [0.160, 0.539] 0.105 [0.050, 0.182] --
same_recall engaged dms 53.385 183 / 32 0.500 [0.317, 0.739] 0.485 [0.292, 0.636] --
same_recall pooled dms -- 511 / 63 0.413 [0.298, 0.556] 0.203 [0.108, 0.308] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low1 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low3 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
low5 manual head 4.983 329 / 31 0.065 [0.000, 0.179] 0.667 [0.000, 1.000] --
low5 engaged head 5.059 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.500 [0.000, 1.000] --
low5 manual dms 15.484 328 / 31 0.032 [0.000, 0.107] 0.143 [0.000, 0.250] --
low5 engaged dms 12.749 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.060] 0.125 [0.000, 0.222] --
dms_matched manual head 53.831 329 / 31 0.452 [0.289, 0.667] 0.175 [0.101, 0.242] --
dms_matched engaged head 48.721 183 / 32 0.281 [0.105, 0.440] 0.214 [0.064, 0.364] --
dms_matched pooled head -- 512 / 63 0.365 [0.239, 0.508] 0.189 [0.104, 0.263] --
dms_matched manual dms 53.860 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
dms_matched engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
dms_matched pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low10 manual head 10.001 329 / 31 0.065 [0.000, 0.179] 0.333 [0.000, 1.000] --
low10 engaged head 9.999 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.286 [0.000, 0.750] --
low10 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low10 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low10 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
same_recall manual head 41.268 329 / 31 0.226 [0.125, 0.346] 0.175 [0.075, 0.289] --
same_recall engaged head 52.426 183 / 32 0.344 [0.184, 0.530] 0.216 [0.080, 0.328] --
same_recall pooled head -- 512 / 63 0.286 [0.190, 0.384] 0.198 [0.099, 0.280] --
same_recall manual dms 53.825 328 / 31 0.226 [0.097, 0.375] 0.103 [0.043, 0.170] --
same_recall engaged dms 48.730 183 / 32 0.344 [0.200, 0.577] 0.579 [0.312, 0.790] --
same_recall pooled dms -- 511 / 63 0.286 [0.190, 0.400] 0.207 [0.116, 0.314] --
low1 manual head 0.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged head 1.036 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low1 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low1 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low1 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low3 manual head 2.983 329 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged head 3.028 183 / 32 0.000 [0.000, 0.000] -- [--, --] --
low3 pooled head -- 512 / 63 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low3 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low3 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --
low5 manual head 4.983 329 / 31 0.065 [0.000, 0.179] 0.667 [0.000, 1.000] --
low5 engaged head 5.059 183 / 32 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 pooled head -- 512 / 63 0.032 [0.000, 0.094] 0.500 [0.000, 1.000] --
low5 manual dms 18.694 328 / 31 0.000 [0.000, 0.000] 0.000 [0.000, 0.000] --
low5 engaged dms 15.582 183 / 32 0.031 [0.000, 0.138] 0.500 [0.000, 1.000] --
low5 pooled dms -- 511 / 63 0.016 [0.000, 0.057] 0.100 [0.000, 0.400] --

Paired lead time (baseline − DMS, undetected → 0) and verdict by the pre-stated rule:

load regime n both detected Δ lead mean s [CI] verdict
dms_matched manual 23 +0.110 [-1.340, +1.738] EQUAL (CI ⊃ 0)
dms_matched engaged 18 +0.544 [-2.435, +2.334] EQUAL (CI ⊃ 0)
dms_matched pooled 41 +0.330 [-1.269, +1.731] EQUAL (CI ⊃ 0)
low10 manual 1 +0.213 [-1.104, +1.550] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.559 [-2.465, -0.224] DMS EARLIER
low10 pooled 7 -0.687 [-1.588, +0.570] EQUAL (CI ⊃ 0)
same_recall manual 19 -1.452 [-2.747, +0.196] EQUAL (CI ⊃ 0)
same_recall engaged 18 +1.366 [-2.285, +3.429] EQUAL (CI ⊃ 0)
same_recall pooled 37 -0.021 [-2.043, +1.817] EQUAL (CI ⊃ 0)
low1 manual 0 -0.881 [-1.889, +0.000] EQUAL (CI ⊃ 0)
low1 engaged 0 -2.866 [-4.733, -1.766] DMS EARLIER
low1 pooled 0 -1.889 [-2.779, -1.183] DMS EARLIER
low3 manual 1 -0.710 [-1.785, +0.190] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.478 [-4.185, -1.183] DMS EARLIER
low3 pooled 3 -1.608 [-2.508, -0.767] DMS EARLIER
low5 manual 1 -0.594 [-1.708, +0.352] EQUAL (CI ⊃ 0)
low5 engaged 3 -2.162 [-3.620, -0.723] DMS EARLIER
low5 pooled 4 -1.390 [-2.178, -0.559] DMS EARLIER
dms_matched manual 22 +0.177 [-1.322, +2.032] EQUAL (CI ⊃ 0)
dms_matched engaged 20 +0.831 [-2.528, +2.725] EQUAL (CI ⊃ 0)
dms_matched pooled 42 +0.510 [-1.208, +2.067] EQUAL (CI ⊃ 0)
low10 manual 3 -0.181 [-1.409, +0.957] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.203 [-1.579, -0.412] DMS EARLIER
low10 pooled 9 -0.700 [-1.359, +0.113] EQUAL (CI ⊃ 0)
same_recall manual 20 -0.965 [-2.524, +1.014] EQUAL (CI ⊃ 0)
same_recall engaged 21 +1.362 [-2.097, +3.168] EQUAL (CI ⊃ 0)
same_recall pooled 41 +0.217 [-1.664, +1.822] EQUAL (CI ⊃ 0)
low1 manual 0 -1.274 [-2.384, -0.390] DMS EARLIER
low1 engaged 0 -2.509 [-4.168, -1.400] DMS EARLIER
low1 pooled 0 -1.902 [-2.865, -1.162] DMS EARLIER
low3 manual 2 -1.103 [-2.246, -0.291] DMS EARLIER
low3 engaged 2 -2.122 [-3.516, -1.110] DMS EARLIER
low3 pooled 4 -1.621 [-2.550, -0.888] DMS EARLIER
low5 manual 3 -0.987 [-2.018, -0.239] DMS EARLIER
low5 engaged 3 -1.806 [-2.949, -0.790] DMS EARLIER
low5 pooled 6 -1.403 [-2.111, -0.739] DMS EARLIER
dms_matched manual 22 +0.358 [-1.249, +2.329] EQUAL (CI ⊃ 0)
dms_matched engaged 18 +0.556 [-2.366, +2.342] EQUAL (CI ⊃ 0)
dms_matched pooled 40 +0.459 [-1.208, +1.866] EQUAL (CI ⊃ 0)
low10 manual 3 -0.071 [-1.423, +1.198] EQUAL (CI ⊃ 0)
low10 engaged 6 -1.697 [-2.261, -0.548] DMS EARLIER
low10 pooled 9 -0.897 [-1.764, +0.134] EQUAL (CI ⊃ 0)
same_recall manual 18 -1.281 [-2.671, +0.482] EQUAL (CI ⊃ 0)
same_recall engaged 18 +1.284 [-2.271, +3.290] EQUAL (CI ⊃ 0)
same_recall pooled 36 +0.022 [-2.013, +1.826] EQUAL (CI ⊃ 0)
low1 manual 0 -1.165 [-2.454, -0.248] DMS EARLIER
low1 engaged 0 -3.003 [-4.405, -1.729] DMS EARLIER
low1 pooled 0 -2.098 [-3.061, -1.268] DMS EARLIER
low3 manual 2 -0.994 [-2.327, +0.030] EQUAL (CI ⊃ 0)
low3 engaged 2 -2.616 [-3.747, -1.390] DMS EARLIER
low3 pooled 4 -1.817 [-2.800, -0.936] DMS EARLIER
low5 manual 3 -0.877 [-2.114, +0.074] EQUAL (CI ⊃ 0)
low5 engaged 3 -2.300 [-3.210, -1.029] DMS EARLIER
low5 pooled 6 -1.600 [-2.385, -0.752] DMS EARLIER

Calibration (temperature fitted on clean-val; 10 equal-count bins):

score T ECE raw ECE calibrated [CI]
head 1.128 0.055 0.062 [0.044, 0.119]
dms 20.000 0.622 0.408 [0.346, 0.459]

Xet Storage Details

Size:
170 kB
·
Xet hash:
a8635ffb4ea799469c0828d43fa18f0fb825f097893168a5ac048f0b721ad326

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.