Pareto DeltaC benchmark v2

Benchmark v2 scores on the DET_ML 6–10 window, public split only (one third of the positions is held out and opened only at milestones). Axes: A1 = count-weighted rms of the per-bin mean(z); A2 = rms of sd(z) − 1; A3 = rms error of P(DET_ML ≥ L) for L = 6, 8, 10 (worst shown). The noise-corrected value is sqrt(max(raw² − floor², 0)); intervals are two-level (observation-cluster) bootstrap percentiles of the raw score. Params = numbers fitted to data; in-sample entries were fitted on rows this layer scores and form a separate front. Highlighted rows are on a front.

Layers. C-all: 4XMM DR14 M2 band 4 point sources (v1 catalogue population). C-iso: C-all restricted to isolated sources (no extended source within 2′, nearest neighbour ≥ 60″), the primary layer for theory claims. L0 (ideal fitter) and L1 (emldetect vs ideal fitter) appear when their production runs finish.

C-all

Combined front: V5c_eta
out-of-sample front: T3_joint, T3_E147mix_jointW; selected: not applicable
in-sample front: V5c_eta; selected: V5c_eta
Selection noise: two-level paired bootstrap percentile interval, B=1000, clusters=image_id

Entry ID	Model	In-sample	Params	A1 mean bias: noise-corrected raw [two-level 95% CI of raw]	A2 width bias: noise-corrected raw [two-level 95% CI of raw]	A3 threshold calibration, worst nc per-L CI of raw
C_all_V5c_eta	V5c_eta	yes	6	0.0633917 raw 0.0699251 [0.0671465, 0.092584]	0.0430508 raw 0.0478422 [0.0454043, 0.0695616]	0.0179404 L6: [0.0194693, 0.029019] L8: [0.0194912, 0.0299005] L10: [0.0109848, 0.0190579]
C_all_T3_joint	T3_joint	no	0	0.375442 raw 0.376596 [0.36268, 0.395939]	0.171561 raw 0.172821 [0.160084, 0.189835]	0.0949639 L6: [0.0667045, 0.0765713] L8: [0.0910258, 0.103069] L10: [0.0684243, 0.0799606]
C_all_T3_E147mix_jointW	T3_E147mix_jointW	no	0	0.377525 raw 0.378673 [0.365442, 0.397195]	0.144066 raw 0.145565 [0.133207, 0.163979]	0.0966641 L6: [0.0660592, 0.0761812] L8: [0.0928374, 0.105132] L10: [0.071126, 0.0826479]
C_all_T3_joint_V4c	T3_joint_V4c	no	2	0.381529 raw 0.382664 [0.3685, 0.402317]	0.189783 raw 0.190923 [0.177798, 0.207679]	0.0954547 L6: [0.0685922, 0.0784889] L8: [0.0915178, 0.103491] L10: [0.0680096, 0.0796229]
C_all_T3_joint_L07	T3_joint_L07	no	1	0.46525 raw 0.466182 [0.451667, 0.484664]	0.17106 raw 0.172323 [0.159538, 0.189382]	0.116639 L6: [0.0799326, 0.0898422] L8: [0.112531, 0.124353] L10: [0.0834837, 0.0949823]
C_all_V4c_no_eta	V4c_no_eta	no	2	0.691062 raw 0.691689 [0.676429, 0.709815]	0.189631 raw 0.190771 [0.177959, 0.207342]	0.168957 L6: [0.109768, 0.120097] L8: [0.164091, 0.176191] L10: [0.120553, 0.131879]

C-iso

Entry ID	Model	In-sample	Params	A1 mean bias: noise-corrected raw [two-level 95% CI of raw]	A2 width bias: noise-corrected raw [two-level 95% CI of raw]	A3 threshold calibration, worst nc per-L CI of raw
C_iso_V5c_eta	V5c_eta	yes	6	0.124529 raw 0.129793 [0.119148, 0.152315]	0.0494115 raw 0.0557756 [0.0531539, 0.0750932]	0.0354606 L6: [0.0346884, 0.04715] L8: [0.0321495, 0.0458784] L10: [0.0139554, 0.0252493]
C_iso_T3_joint	T3_joint	no	0	0.29579 raw 0.298042 [0.283813, 0.321422]	0.107294 raw 0.110366 [0.100098, 0.12717]	0.0782548 L6: [0.056247, 0.0699308] L8: [0.0748427, 0.0889652] L10: [0.0523412, 0.0660436]
C_iso_T3_joint_V4c	T3_joint_V4c	no	2	0.300693 raw 0.302909 [0.288402, 0.326669]	0.125729 raw 0.128361 [0.117289, 0.14476]	0.0785286 L6: [0.058077, 0.0717391] L8: [0.0750962, 0.089179] L10: [0.0517791, 0.0657443]
C_iso_T3_E147mix_jointW	T3_E147mix_jointW	no	0	0.31482 raw 0.316938 [0.302625, 0.339816]	0.0766064 raw 0.0808545 [0.0727077, 0.0991898]	0.084452 L6: [0.0559391, 0.0697195] L8: [0.0807863, 0.0948442] L10: [0.0581327, 0.0718153]
C_iso_T3_joint_L07	T3_joint_L07	no	1	0.380769 raw 0.382521 [0.36676, 0.405358]	0.106933 raw 0.110016 [0.0998087, 0.126981]	0.0996256 L6: [0.0709451, 0.0845088] L8: [0.0955361, 0.109889] L10: [0.0654907, 0.079059]
C_iso_V4c_no_eta	V4c_no_eta	no	2	0.596976 raw 0.598095 [0.580778, 0.62086]	0.126056 raw 0.128681 [0.117674, 0.14515]	0.151828 L6: [0.105731, 0.119737] L8: [0.146563, 0.161338] L10: [0.0985027, 0.111892]