Archive 4 checkpoints (1-4)

paper-releases/pbisc-diffusion/updates/20261003T105229Z-deck-cross-cohort-loco-k1000/diffusion/dataset/server_1/model8/plans/paired_crosscohort_20261002/scores/AGGREGATE.md

@@ -0,0 +1,117 @@
+# Cross-cohort (LOCO) benchmark: scores
+
+Generated by `tools/aggregate_loco.py`; do not edit. Zero-pair arms first, then within-cohort references that used the held-out cohort's own pairs.
+
+## ImMicro (107 samples; reference c5e2547be3ec)
+
+| arm | comp_l1 | p_l1_vs_zs_K1000 | precision | coverage | coverage_dm | pb_r_median |
+|---|---|---|---|---|---|---|
+| locoL_fullsat_s20000_K1000 | 0.208 | 4.29e-19 | 0.74 | 0.386 | 0.719 | 0.886 |
+| locoL_fullsat_s40000_K1000 | 0.211 | 3.93e-18 | 0.729 | 0.368 | 0.738 | 0.883 |
+| locoL_fullsat_s60000_K1000 | 0.223 | 1.14e-16 | 0.709 | 0.349 | 0.72 | 0.882 |
+| locoL_fullsat_s65000_K1000 | 0.213 | 2.67e-17 | 0.704 | 0.346 | 0.713 | 0.882 |
+| locoL_lora16_K1000 | 0.163 | 2.81e-19 | 0.833 | 0.451 | 0.705 | 0.886 |
+| locoL_lora16sat_s20000_K1000 | 0.163 | 2.81e-19 | 0.833 | 0.451 | 0.705 | 0.886 |
+| locoL_lora16sat_s28500_K1000 | 0.175 | 3.24e-19 | 0.841 | 0.452 | 0.685 | 0.887 |
+| locoL_lora64sat_s20000_K1000 | 0.196 | 6.78e-18 | 0.855 | 0.437 | 0.713 | 0.878 |
+| locoL_lora64sat_s39000_K1000 | 0.199 | 1.48e-17 | 0.835 | 0.382 | 0.664 | 0.878 |
+| xcalL_K1000 | 0.172 | 5.67e-19 | 0.902 | 0.572 | 0.738 | 0.889 |
+| zs_K1000 | 0.446 |  | 0.671 | 0.186 | 0.211 | 0.892 |
+| zs_K2000 | 0.463 | 1.69e-10 | 0.552 | 0.132 | 0.147 | 0.889 |
+| [within-cohort calibration] zs__cal | 0.199 | 7.12e-17 | 0.905 | 0.527 | 0.769 | 0.88 |
+| [within-cohort LoRA r16] real_lora__raw | 0.162 | 1.05e-18 | 0.924 | 0.784 | 0.962 | 0.899 |
+| [within-cohort full FT] real_full__raw | 0.157 | 6e-19 | 0.89 | 0.695 | 0.95 | 0.914 |
+
+## Sepsis (82 samples; reference bac12da8a241)
+
+| arm | comp_l1 | p_l1_vs_zs_K1000 | precision | coverage | coverage_dm | pb_r_median | logfc_r | broadcast_r | partial_r | deg50 | deg50_broadcast | deg50_max |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| locoL_fullsat_s100000_K1000 | 0.577 | 0.059 | 0.819 | 0.353 | 0.553 | 0.877 | 0.0778 | 0.316 | 0.00597 | 7 | 21 | 250 |
+| locoL_fullsat_s108500_K1000 | 0.585 | 0.0381 | 0.821 | 0.346 | 0.551 | 0.877 | 0.073 | 0.316 | -7.02e-05 | 5 | 21 | 250 |
+| locoL_fullsat_s20000_K1000 | 0.498 | 0.972 | 0.838 | 0.329 | 0.587 | 0.876 | 0.121 | 0.316 | 0.0503 | 10 | 21 | 250 |
+| locoL_fullsat_s40000_K1000 | 0.53 | 0.399 | 0.835 | 0.335 | 0.59 | 0.876 | 0.112 | 0.316 | 0.045 | 7 | 21 | 250 |
+| locoL_fullsat_s60000_K1000 | 0.546 | 0.22 | 0.83 | 0.339 | 0.57 | 0.877 | 0.0831 | 0.316 | 0.0124 | 7 | 21 | 250 |
+| locoL_fullsat_s80000_K1000 | 0.576 | 0.0648 | 0.828 | 0.342 | 0.577 | 0.878 | 0.0873 | 0.316 | 0.0163 | 6 | 21 | 250 |
+| locoL_lora16_K1000 | 0.5 | 0.873 | 0.866 | 0.347 | 0.572 | 0.865 | 0.125 | 0.316 | 0.0724 | 9 | 21 | 250 |
+| locoL_lora16sat_s20000_K1000 | 0.5 | 0.873 | 0.866 | 0.347 | 0.572 | 0.865 | 0.125 | 0.316 | 0.0724 | 9 | 21 | 250 |
+| locoL_lora16sat_s26750_K1000 | 0.491 | 0.801 | 0.861 | 0.341 | 0.568 | 0.866 | 0.102 | 0.316 | 0.0504 | 8 | 21 | 250 |
+| locoL_lora64sat_s20000_K1000 | 0.55 | 0.0911 | 0.836 | 0.295 | 0.494 | 0.867 | 0.177 | 0.316 | 0.124 | 12 | 21 | 250 |
+| locoL_lora64sat_s40000_K1000 | 0.564 | 0.0682 | 0.827 | 0.291 | 0.489 | 0.865 | 0.215 | 0.316 | 0.162 | 4 | 21 | 250 |
+| locoL_lora64sat_s59000_K1000 | 0.57 | 0.0893 | 0.828 | 0.289 | 0.486 | 0.862 | 0.228 | 0.316 | 0.176 | 4 | 21 | 250 |
+| loco_lora16_K1000 | 0.538 | 0.184 | 0.851 | 0.288 | 0.513 | 0.866 | 0.0916 | 0.316 | 0.0336 | 7 | 21 | 250 |
+| xcalL_K1000 | 0.444 | 0.00276 | 0.896 | 0.464 | 0.652 | 0.866 | 0.0967 | 0.317 | 0.0165 | 13 | 27 | 250 |
+| zs_K1000 | 0.509 |  | 0.617 | 0.074 | 0.0712 | 0.879 | -0.0218 | 0.316 | -0.103 | 9 | 21 | 250 |
+| zs_K2000 | 0.496 | 0.00799 | 0.616 | 0.0572 | 0.0554 | 0.878 | -0.036 | 0.316 | -0.116 | 10 | 21 | 250 |
+| [within-cohort calibration] zs__cal | 0.401 | 0.000189 | 0.903 | 0.431 | 0.683 | 0.861 | 0.123 | 0.278 | 0.0394 | 11 | 22 | 250 |
+| [within-cohort LoRA r16] real_lora__raw | 0.362 | 0.000685 | 0.926 | 0.829 | 0.949 | 0.891 | 0.527 | 0.316 | 0.469 | 23 | 21 | 250 |
+| [within-cohort full FT] real_full__raw | 0.372 | 0.000402 | 0.866 | 0.795 | 0.927 | 0.901 | 0.534 | 0.316 | 0.468 | 24 | 21 | 250 |
+
+## Imperial (157 samples; reference 34b4caa3f8a6)
+
+| arm | comp_l1 | p_l1_vs_zs_K1000 | precision | coverage | coverage_dm | pb_r_median | logfc_r | broadcast_r | partial_r | deg50 | deg50_broadcast | deg50_max |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| locoL_fullsat_s20000_K1000 | 0.632 | 2.64e-13 | 0.763 | 0.6 | 0.692 | 0.877 | 0.532 | 0.661 | 0.325 | 98 | 84 | 250 |
+| locoL_fullsat_s40000_K1000 | 0.609 | 4.89e-13 | 0.787 | 0.603 | 0.704 | 0.878 | 0.503 | 0.661 | 0.293 | 99 | 84 | 250 |
+| locoL_fullsat_s60000_K1000 | 0.565 | 1.38e-15 | 0.798 | 0.611 | 0.703 | 0.881 | 0.498 | 0.661 | 0.291 | 99 | 84 | 250 |
+| locoL_fullsat_s80000_K1000 | 0.576 | 2.57e-15 | 0.809 | 0.605 | 0.701 | 0.88 | 0.466 | 0.661 | 0.251 | 93 | 84 | 250 |
+| locoL_fullsat_s93000_K1000 | 0.58 | 3.45e-15 | 0.802 | 0.608 | 0.701 | 0.879 | 0.459 | 0.661 | 0.244 | 97 | 84 | 250 |
+| locoL_lora16_K1000 | 0.657 | 8.24e-15 | 0.804 | 0.683 | 0.75 | 0.872 | 0.527 | 0.661 | 0.316 | 96 | 84 | 250 |
+| locoL_lora16sat_s20000_K1000 | 0.657 | 8.24e-15 | 0.804 | 0.683 | 0.75 | 0.872 | 0.527 | 0.661 | 0.316 | 96 | 84 | 250 |
+| locoL_lora16sat_s40000_K1000 | 0.635 | 1.01e-14 | 0.82 | 0.684 | 0.755 | 0.871 | 0.491 | 0.661 | 0.268 | 94 | 84 | 250 |
+| locoL_lora64sat_s20000_K1000 | 0.622 | 4.58e-16 | 0.808 | 0.683 | 0.761 | 0.87 | 0.511 | 0.661 | 0.297 | 101 | 84 | 250 |
+| locoL_lora64sat_s40000_K1000 | 0.603 | 6.15e-17 | 0.819 | 0.711 | 0.782 | 0.874 | 0.535 | 0.661 | 0.314 | 101 | 84 | 250 |
+| locoL_lora64sat_s60000_K1000 | 0.577 | 1.98e-18 | 0.818 | 0.723 | 0.783 | 0.876 | 0.568 | 0.661 | 0.361 | 109 | 84 | 250 |
+| locoL_lora64sat_s69250_K1000 | 0.594 | 6.3e-18 | 0.822 | 0.723 | 0.779 | 0.876 | 0.58 | 0.661 | 0.373 | 102 | 84 | 250 |
+| xcalL_K1000 | 0.651 | 1.06e-16 | 0.845 | 0.637 | 0.664 | 0.865 | 0.689 | 0.677 | 0.445 | 108 | 84 | 250 |
+| zs_K1000 | 0.803 |  | 0.494 | 0.113 | 0.157 | 0.884 | 0.642 | 0.661 | 0.428 | 108 | 84 | 250 |
+| [within-cohort calibration] zs__cal | 0.494 | 8.14e-16 | 0.842 | 0.639 | 0.688 | 0.867 | 0.679 | 0.69 | 0.412 | 111 | 97 | 250 |
+| [within-cohort LoRA r16] real_lora__raw | 0.376 | 9.15e-21 | 0.847 | 0.85 | 0.937 | 0.88 | 0.819 | 0.661 | 0.69 | 143 | 84 | 250 |
+
+## Giroux (24 samples; reference dd0152f7bde4)
+
+| arm | comp_l1 | p_l1_vs_zs_K1000 | precision | coverage | coverage_dm | pb_r_median | logfc_r | broadcast_r | partial_r | deg50 | deg50_broadcast | deg50_max |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| locoL_fullsat_s100000_K1000 | 0.49 | 0.0164 | 0.983 | 0.764 | 0.741 | 0.86 | 0.225 | 0.575 | 0.131 | 1 | 5 | 101 |
+| locoL_fullsat_s120000_K1000 | 0.465 | 0.0526 | 0.984 | 0.776 | 0.731 | 0.869 | 0.196 | 0.575 | 0.107 | 2 | 5 | 101 |
+| locoL_fullsat_s20000_K1000 | 0.543 | 0.000742 | 0.985 | 0.753 | 0.785 | 0.856 | 0.208 | 0.575 | 0.131 | 2 | 5 | 101 |
+| locoL_fullsat_s40000_K1000 | 0.532 | 0.000239 | 0.988 | 0.753 | 0.742 | 0.852 | 0.203 | 0.575 | 0.131 | 2 | 5 | 101 |
+| locoL_fullsat_s60000_K1000 | 0.499 | 0.00871 | 0.986 | 0.756 | 0.726 | 0.855 | 0.217 | 0.575 | 0.141 | 2 | 5 | 101 |
+| locoL_fullsat_s80000_K1000 | 0.466 | 0.0691 | 0.984 | 0.764 | 0.729 | 0.858 | 0.228 | 0.575 | 0.14 | 2 | 5 | 101 |
+| locoL_lora16_K1000 | 0.416 | 0.152 | 0.991 | 0.793 | 0.819 | 0.845 | 0.296 | 0.575 | 0.204 | 0 | 5 | 101 |
+| locoL_lora16sat_s20000_K1000 | 0.416 | 0.152 | 0.991 | 0.793 | 0.819 | 0.845 | 0.296 | 0.575 | 0.204 | 0 | 5 | 101 |
+| locoL_lora16sat_s35750_K1000 | 0.431 | 0.16 | 0.995 | 0.819 | 0.835 | 0.844 | 0.299 | 0.575 | 0.203 | 0 | 5 | 101 |
+| locoL_lora64_K1000 | 0.491 | 0.0211 | 0.994 | 0.802 | 0.819 | 0.835 | 0.264 | 0.575 | 0.171 | 2 | 5 | 101 |
+| locoL_lora64sat_s20000_K1000 | 0.491 | 0.0211 | 0.994 | 0.802 | 0.819 | 0.835 | 0.264 | 0.575 | 0.171 | 2 | 5 | 101 |
+| locoL_lora64sat_s40000_K1000 | 0.469 | 0.034 | 0.994 | 0.818 | 0.835 | 0.843 | 0.239 | 0.575 | 0.154 | 0 | 5 | 101 |
+| locoL_lora64sat_s60000_K1000 | 0.43 | 0.114 | 0.994 | 0.815 | 0.845 | 0.847 | 0.272 | 0.575 | 0.187 | 1 | 5 | 101 |
+| locoL_lora64sat_s69250_K1000 | 0.41 | 0.229 | 0.994 | 0.809 | 0.841 | 0.85 | 0.278 | 0.575 | 0.187 | 0 | 5 | 101 |
+| loco_lora16_K1000 | 0.347 | 0.833 | 0.991 | 0.808 | 0.826 | 0.847 | 0.296 | 0.575 | 0.188 | 0 | 5 | 101 |
+| xcalL_K1000 | 0.202 | 0.00015 | 0.988 | 0.891 | 0.889 | 0.885 | 0.238 | 0.569 | 0.115 | 0 | 2 | 101 |
+| zs_K1000 | 0.337 |  | 0.904 | 0.293 | 0.369 | 0.889 | 0.34 | 0.575 | 0.197 | 0 | 5 | 101 |
+| zs_K2000 | 0.322 | 0.0425 | 0.917 | 0.277 | 0.344 | 0.889 | 0.339 | 0.575 | 0.189 | 0 | 5 | 101 |
+| [within-cohort calibration] zs__cal | 0.266 | 0.00958 | 0.995 | 0.906 | 0.902 | 0.865 | 0.269 | 0.564 | 0.133 | 0 | 2 | 101 |
+| [within-cohort LoRA r16] real_lora__raw | 0.258 | 0.0315 | 0.998 | 0.928 | 0.974 | 0.865 | 0.461 | 0.575 | 0.323 | 5 | 5 | 101 |
+
+## Ji (whole blood) (33 samples; reference 19c30f56db69)
+
+| arm | comp_l1 | p_l1_vs_zs_K1000 | precision | coverage | coverage_dm | pb_r_median | logfc_r | broadcast_r | partial_r | deg50 | deg50_broadcast | deg50_max |
+|---|---|---|---|---|---|---|---|---|---|---|---|---|
+| locoL_fullsat_s20000_K1000 | 0.838 | 1e-06 | 0.714 | 0.168 | 0.392 | 0.826 | 0.187 | 0.24 | 0.14 | 4 | 4 | 250 |
+| locoL_fullsat_s40000_K1000 | 1.01 | 4.66e-10 | 0.611 | 0.112 | 0.282 | 0.818 | 0.149 | 0.24 | 0.105 | 2 | 4 | 250 |
+| locoL_fullsat_s60000_K1000 | 1.03 | 4.66e-10 | 0.541 | 0.0836 | 0.23 | 0.814 | 0.116 | 0.24 | 0.0787 | 0 | 4 | 250 |
+| locoL_fullsat_s80000_K1000 | 1.05 | 2.33e-10 | 0.512 | 0.0721 | 0.2 | 0.815 | 0.132 | 0.24 | 0.0953 | 0 | 4 | 250 |
+| locoL_fullsat_s86750_K1000 | 1.03 | 4.66e-10 | 0.511 | 0.0812 | 0.201 | 0.817 | 0.139 | 0.24 | 0.101 | 5 | 4 | 250 |
+| locoL_lora16_K1000 | 0.757 | 0.0898 | 0.837 | 0.271 | 0.4 | 0.831 | 0.144 | 0.24 | 0.102 | 2 | 4 | 250 |
+| locoL_lora16sat_s20000_K1000 | 0.757 | 0.0898 | 0.837 | 0.271 | 0.4 | 0.831 | 0.144 | 0.24 | 0.102 | 2 | 4 | 250 |
+| locoL_lora16sat_s40000_K1000 | 0.74 | 0.764 | 0.911 | 0.339 | 0.424 | 0.841 | 0.0453 | 0.24 | 0.00734 | 0 | 4 | 250 |
+| locoL_lora64_K1000 | 0.639 | 0.00548 | 0.931 | 0.381 | 0.538 | 0.843 | 0.199 | 0.24 | 0.151 | 2 | 4 | 250 |
+| locoL_lora64sat_s20000_K1000 | 0.639 | 0.00548 | 0.931 | 0.381 | 0.538 | 0.843 | 0.199 | 0.24 | 0.151 | 2 | 4 | 250 |
+| locoL_lora64sat_s40000_K1000 | 0.668 | 0.0307 | 0.948 | 0.387 | 0.553 | 0.847 | 0.177 | 0.24 | 0.129 | 0 | 4 | 250 |
+| locoL_lora64sat_s51750_K1000 | 0.62 | 0.00158 | 0.947 | 0.395 | 0.572 | 0.846 | 0.164 | 0.24 | 0.127 | 0 | 4 | 250 |
+| loco_lora16_K1000 | 0.494 | 0.000602 | 0.845 | 0.313 | 0.521 | 0.836 | 0.0892 | 0.24 | 0.0417 | 9 | 4 | 250 |
+| xcalL_K1000 | 0.565 | 0.000245 | 0.636 | 0.0711 | 0.274 | 0.828 | 0.127 | 0.256 | 0.0897 | 0 | 4 | 250 |
+| zs_K1000 | 0.725 |  | 0.721 | 0.059 | 0.195 | 0.834 | 0.0749 | 0.24 | 0.023 | 0 | 4 | 250 |
+| zs_K2000 | 0.729 | 0.538 | 0.719 | 0.0604 | 0.173 | 0.836 | 0.0756 | 0.24 | 0.0229 | 0 | 4 | 250 |
+| [within-cohort calibration] zs__cal | 0.501 | 0.0011 | 0.74 | 0.106 | 0.347 | 0.839 | 0.142 | 0.206 | 0.113 | 0 | 1 | 250 |
+| [within-cohort LoRA r16] real_lora__raw | 0.504 | 0.00879 | 0.971 | 0.857 | 0.965 | 0.873 | 0.479 | 0.24 | 0.452 | 15 | 4 | 250 |
+

paper-releases/pbisc-diffusion/updates/20261003T105229Z-deck-cross-cohort-loco-k1000/diffusion/dataset/server_1/model8/plans/paired_crosscohort_20261002/scores/aggregate.csv

@@ -0,0 +1,89 @@
+cohort,arm,uses_target_pairs,comp_l1,precision,coverage,precision_dm,coverage_dm,pb_r_median,logfc_r,broadcast_r,partial_r,deg50,deg50_broadcast,deg50_max,p_l1_vs_zs_K1000,p_l1_vs_within_lora
+immicro,locoL_fullsat_s20000_K1000,False,0.2076823216876906,0.7401869158878505,0.38560747663551403,0.8212461059190032,0.7194392523364486,0.8863773039051367,,,,,,,4.28646220899865e-19,7.499747580707642e-07
+immicro,locoL_fullsat_s40000_K1000,False,0.21055421464175067,0.7288473520249221,0.36778816199376946,0.8121495327102803,0.738006230529595,0.883436496463128,,,,,,,3.934224985657432e-18,3.3368425928831524e-07
+immicro,locoL_fullsat_s60000_K1000,False,0.22251011942239113,0.7089719626168224,0.3490342679127726,0.8016199376947041,0.7200623052959502,0.8817591984662231,,,,,,,1.1414042339557987e-16,3.2178143752924207e-08
+immicro,locoL_fullsat_s65000_K1000,False,0.21255123747092092,0.7041744548286605,0.34579439252336447,0.7967601246105919,0.7125233644859813,0.881878378631933,,,,,,,2.6736045710989122e-17,5.099080196244263e-07
+immicro,locoL_lora16_K1000,False,0.16262332160969473,0.8329595015576324,0.45065420560747665,0.8711526479750779,0.7047352024922119,0.8858693369632604,,,,,,,2.80949222844839e-19,0.8301939730942434
+immicro,locoL_lora16sat_s20000_K1000,False,0.16262332160969473,0.8329595015576324,0.45065420560747665,0.8711526479750779,0.7047352024922119,0.8858693369632604,,,,,,,2.80949222844839e-19,0.8301939730942434
+immicro,locoL_lora16sat_s28500_K1000,False,0.17491854684276595,0.8407476635514018,0.45165109034267914,0.883177570093458,0.6854828660436137,0.8873478720732648,,,,,,,3.2351062602524277e-19,0.26318956196551035
+immicro,locoL_lora64sat_s20000_K1000,False,0.19608124284129383,0.8545171339563863,0.437196261682243,0.8965732087227415,0.7131464174454829,0.8777141587697234,,,,,,,6.782738418768735e-18,3.7509756222820777e-06
+immicro,locoL_lora64sat_s39000_K1000,False,0.19858024117626522,0.8349532710280374,0.38174454828660437,0.8785669781931464,0.66398753894081,0.877567489140798,,,,,,,1.4840659597953144e-17,3.768658900124394e-05
+immicro,xcalL_K1000,False,0.17220690476643966,0.901993769470405,0.5724610591900312,0.9343302180685358,0.7378193146417446,0.8888913715424525,,,,,,,5.674071885692667e-19,0.5674072544647283
+immicro,zs_K1000,False,0.4457108278919341,0.6705919003115265,0.1864797507788162,0.6563239875389408,0.2107165109034268,0.8920390135033525,,,,,,,,1.0480605741268486e-18
+immicro,zs_K2000,False,0.463206914700923,0.5517757009345794,0.13177570093457944,0.5385669781931465,0.14728971962616821,0.8886936323693132,,,,,,,1.6915630361025524e-10,8.869637083427336e-19
+immicro,[within-cohort calibration] zs__cal,True,0.19850275515696958,0.9050467289719626,0.527227414330218,0.944423676012461,0.7689719626168224,0.8801553120873381,,,,,,,7.12072465706631e-17,1.930319432195753e-07
+immicro,[within-cohort LoRA r16] real_lora__raw,True,0.16172890350786212,0.9241121495327103,0.7842990654205607,0.9606230529595016,0.9623052959501558,0.8994149755898816,,,,,,,1.0480605741268486e-18,
+immicro,[within-cohort full FT] real_full__raw,True,0.1572206245036138,0.8895327102803738,0.6953894080996885,0.9459190031152648,0.9501557632398754,0.9140523649103616,,,,,,,6.000732255664671e-19,0.24760598984759485
+ye2026_sepsis,locoL_fullsat_s100000_K1000,False,0.5774379174565976,0.818780487804878,0.35317073170731705,0.7932520325203252,0.5532520325203252,0.8771110294275071,0.07776924200901651,0.31594581626202867,0.0059716621665920466,7.0,21.0,250.0,0.05896308746880887,3.4391180557739516e-08
+ye2026_sepsis,locoL_fullsat_s108500_K1000,False,0.5846173784329263,0.8208130081300813,0.3456910569105691,0.7928455284552846,0.5505691056910569,0.8765570684516901,0.07297860484594258,0.31594581626202867,-7.021571424081427e-05,5.0,21.0,250.0,0.0381363876701571,1.1497681302540135e-08
+ye2026_sepsis,locoL_fullsat_s20000_K1000,False,0.4979485447700929,0.8376422764227642,0.32926829268292684,0.7985365853658537,0.5865853658536585,0.876321030085444,0.12050615062254941,0.31594581626202867,0.0502926168036133,10.0,21.0,250.0,0.9723412853463738,5.7153801707671366e-05
+ye2026_sepsis,locoL_fullsat_s40000_K1000,False,0.5300574276228053,0.8350406504065041,0.33471544715447155,0.7960162601626016,0.5895121951219512,0.8761626372026949,0.11242527537940607,0.31594581626202867,0.044959343731627986,7.0,21.0,250.0,0.39884521572626563,4.000113093132899e-06
+ye2026_sepsis,locoL_fullsat_s60000_K1000,False,0.5459371522940706,0.8296747967479675,0.33926829268292685,0.7919512195121952,0.5704878048780487,0.8766076538284067,0.08306109681282589,0.31594581626202867,0.012438320245952872,7.0,21.0,250.0,0.2196764453237604,4.5131528630386724e-07
+ye2026_sepsis,locoL_fullsat_s80000_K1000,False,0.5756093100530909,0.8283739837398374,0.3416260162601626,0.807479674796748,0.577479674796748,0.8778080739176894,0.08726311196922498,0.31594581626202867,0.016341209913177546,6.0,21.0,250.0,0.06476713445259935,2.5046618892523646e-08
+ye2026_sepsis,locoL_lora16_K1000,False,0.5000091994646808,0.8658536585365854,0.3468292682926829,0.8295121951219512,0.5719512195121951,0.8647148730786527,0.12455395226696071,0.31594581626202867,0.07238115840305821,9.0,21.0,250.0,0.8732818777129994,3.8429820644344716e-05
+ye2026_sepsis,locoL_lora16sat_s20000_K1000,False,0.5000091994646808,0.8658536585365854,0.3468292682926829,0.8295121951219512,0.5719512195121951,0.8647148730786527,0.12455395226696071,0.31594581626202867,0.07238115840305821,9.0,21.0,250.0,0.8732818777129994,3.8429820644344716e-05
+ye2026_sepsis,locoL_lora16sat_s26750_K1000,False,0.4913211548128607,0.8611382113821138,0.3413821138211382,0.8254471544715447,0.5679674796747968,0.8655208432760766,0.10171004353585707,0.31594581626202867,0.05041380972761813,8.0,21.0,250.0,0.8010797939807706,6.303558842559553e-05
+ye2026_sepsis,locoL_lora64sat_s20000_K1000,False,0.5500938157804457,0.8356097560975609,0.2948780487804878,0.7879674796747967,0.49365853658536585,0.866523942246353,0.17680441182744347,0.31594581626202867,0.123774850768618,12.0,21.0,250.0,0.09108874909541066,4.096370143419454e-07
+ye2026_sepsis,locoL_lora64sat_s40000_K1000,False,0.5637732142083826,0.8269918699186992,0.2910569105691057,0.7889430894308943,0.489349593495935,0.8647820399736996,0.21467488951638702,0.31594581626202867,0.16248853205684335,4.0,21.0,250.0,0.06819027045646373,1.9115971450977619e-07
+ye2026_sepsis,locoL_lora64sat_s59000_K1000,False,0.5696031573329058,0.828130081300813,0.2891869918699187,0.7696747967479675,0.48577235772357724,0.8623590296603416,0.2279128935245801,0.31594581626202867,0.1762499986107458,4.0,21.0,250.0,0.08933271679629097,4.132099266679222e-08
+ye2026_sepsis,loco_lora16_K1000,False,0.537784206603567,0.8506504065040651,0.28804878048780486,0.7960162601626016,0.5129268292682927,0.8657510412987999,0.09162100245097674,0.31594581626202867,0.03357463198358491,7.0,21.0,250.0,0.18381713324066362,8.810964325972628e-07
+ye2026_sepsis,xcalL_K1000,False,0.4438043638834868,0.8960975609756098,0.4638211382113821,0.8978861788617886,0.651869918699187,0.8655010028906098,0.09668769745340702,0.3172672992140051,0.016469708030032833,13.0,27.0,250.0,0.0027593342725831522,0.0074870419209778424
+ye2026_sepsis,zs_K1000,False,0.5094583598384997,0.6169918699186991,0.07398373983739838,0.5752032520325203,0.07121951219512195,0.8786516011139656,-0.021848803748335324,0.31594581626202867,-0.10302610190502556,9.0,21.0,250.0,,0.000684921410799816
+ye2026_sepsis,zs_K2000,False,0.4961155621237927,0.6156910569105691,0.05723577235772358,0.5948780487804878,0.05544715447154472,0.8778601343608707,-0.03602114990970071,0.31594581626202867,-0.11556092960733022,10.0,21.0,250.0,0.0079929874710464,0.0008803845436323717
+ye2026_sepsis,[within-cohort calibration] zs__cal,True,0.40065324062398555,0.9034146341463415,0.4308130081300813,0.9110569105691056,0.6827642276422764,0.8611254347365214,0.12267970089558222,0.27791460488477016,0.03938214909552812,11.0,22.0,250.0,0.00018919827139893567,0.19313667138032242
+ye2026_sepsis,[within-cohort LoRA r16] real_lora__raw,True,0.3618246783833512,0.9256910569105691,0.829349593495935,0.9334959349593496,0.9490243902439024,0.890814724779387,0.5267612711229187,0.31594581626202867,0.46874787884206276,23.0,21.0,250.0,0.000684921410799816,
+ye2026_sepsis,[within-cohort full FT] real_full__raw,True,0.3716534409830164,0.8664227642276423,0.7946341463414635,0.8782926829268293,0.9273983739837398,0.9013444743771939,0.534366534546327,0.31594581626202867,0.46768653125846193,24.0,21.0,250.0,0.00040186525013656175,0.7132287608306445
+imperial_eskd,locoL_fullsat_s20000_K1000,False,0.6320557687766534,0.7625415211651477,0.6004599267524061,0.8030406268631292,0.692019419129546,0.8774232337126225,0.5321348760714089,0.6606053132085035,0.3253497877510006,98.0,84.0,250.0,2.638498689139773e-13,6.05884420577385e-17
+imperial_eskd,locoL_fullsat_s40000_K1000,False,0.608905739142699,0.7874116344434035,0.6026318030832126,0.8236947449109957,0.7040712034750022,0.87824769514186,0.5027476205152808,0.6606053132085035,0.29334444050906294,99.0,84.0,250.0,4.886513239724087e-13,2.080482876777613e-15
+imperial_eskd,locoL_fullsat_s60000_K1000,False,0.5648876304384981,0.7981006728558044,0.6114470658376628,0.8325525934758539,0.7030917298356187,0.8806915068148377,0.49841549661342743,0.6606053132085035,0.29078020889831985,99.0,84.0,250.0,1.3795383253440688e-15,1.582705982680838e-13
+imperial_eskd,locoL_fullsat_s80000_K1000,False,0.5760579253485651,0.809385912613917,0.6047610936036113,0.8334468954944213,0.701303125798484,0.8798943711569296,0.46572797196536914,0.6606053132085035,0.2508436463606439,93.0,84.0,250.0,2.5704941305972313e-15,2.4498945088365615e-14
+imperial_eskd,locoL_fullsat_s93000_K1000,False,0.5795732789528253,0.8022314964653777,0.6076143429009454,0.8363427306021634,0.7007069244527724,0.879378718090629,0.45945747768103096,0.6606053132085035,0.24406653287984167,97.0,84.0,250.0,3.4524384903074747e-15,4.208149676813619e-14
+imperial_eskd,locoL_lora16_K1000,False,0.6574725585859008,0.8041052721233285,0.682863469891832,0.8409419981262244,0.7504045651988758,0.8721980540395425,0.5267014418562213,0.6606053132085035,0.3160891154874635,96.0,84.0,250.0,8.240504595371867e-15,3.458902313776944e-18
+imperial_eskd,locoL_lora16sat_s20000_K1000,False,0.6574725585859008,0.8041052721233285,0.682863469891832,0.8409419981262244,0.7504045651988758,0.8721980540395425,0.5267014418562213,0.6606053132085035,0.3160891154874635,96.0,84.0,250.0,8.240504595371867e-15,3.458902313776944e-18
+imperial_eskd,locoL_lora16sat_s40000_K1000,False,0.6354385520845358,0.8195639213014224,0.6838003577208074,0.8513329358657695,0.7547909036708969,0.8713917655234306,0.49115540391304124,0.6606053132085035,0.2681664933612403,94.0,84.0,250.0,1.0066597334428257e-14,1.5361107057499805e-16
+imperial_eskd,locoL_lora64sat_s20000_K1000,False,0.6215824758631929,0.8080231666808619,0.6829060557022401,0.8451154075462056,0.760710331317605,0.8703845911453068,0.5106680194292705,0.6606053132085035,0.29693242843172907,101.0,84.0,250.0,4.577247144033568e-16,8.648144888064831e-17
+imperial_eskd,locoL_lora64sat_s40000_K1000,False,0.6034640048005437,0.8191806490077507,0.7111404480027255,0.8568265054083979,0.7818754790903671,0.8737452140146552,0.5353798731445953,0.6606053132085035,0.314018613209171,101.0,84.0,250.0,6.149556620111599e-17,5.78659288288674e-15
+imperial_eskd,locoL_lora64sat_s60000_K1000,False,0.5769841372055678,0.8175623882122477,0.7225960310024699,0.8532492973341282,0.7831956392130143,0.8758184245049896,0.5679152517190105,0.6606053132085035,0.3612044321070582,109.0,84.0,250.0,1.9803830347651483e-18,6.464485470063783e-14
+imperial_eskd,locoL_lora64sat_s69250_K1000,False,0.5944384838913485,0.8223319989779405,0.7228941316753258,0.8549953155608552,0.7787241291201772,0.875865335929421,0.5804458634433078,0.6606053132085035,0.37286527777401546,102.0,84.0,250.0,6.300421879924113e-18,3.3008547839146704e-14
+imperial_eskd,xcalL_K1000,False,0.6511665403698953,0.8453709224086534,0.6370411378928541,0.8699855208244613,0.6638701984498765,0.8649784260819683,0.6892227232868162,0.6767220706910705,0.44492298254048923,108.0,84.0,250.0,1.055623106261569e-16,9.174972659328609e-17
+imperial_eskd,zs_K1000,False,0.8029956289662181,0.4935695426283962,0.11315049825398177,0.5616642534707436,0.1567157823013372,0.8842598187000824,0.6419400899561818,0.6606053132085035,0.4277975641323528,108.0,84.0,250.0,,9.145080649116403e-21
+imperial_eskd,[within-cohort calibration] zs__cal,True,0.4943429956255003,0.8417085427135679,0.6392981858444766,0.8640235073673452,0.6882292820032365,0.8669239796274295,0.6787028713891314,0.6901082987421676,0.41180392471305144,111.0,97.0,250.0,8.137216788076729e-16,7.01268509482979e-11
+imperial_eskd,[within-cohort LoRA r16] real_lora__raw,True,0.37638560875745664,0.847457627118644,0.8500127757431224,0.8881270760582574,0.9371433438378333,0.8801149509776351,0.8187448084273437,0.6606053132085035,0.6899314750539278,143.0,84.0,250.0,9.145080649116403e-21,
+giroux2022,locoL_fullsat_s100000_K1000,False,0.48950182739661546,0.9827777777777778,0.7636111111111111,0.9422222222222222,0.7413888888888889,0.8603229232152234,0.22468780640184455,0.5752659929360795,0.13129069625153494,1.0,5.0,101.0,0.016406655311584473,2.9802322387695312e-06
+giroux2022,locoL_fullsat_s120000_K1000,False,0.46479005295114834,0.9844444444444445,0.7755555555555556,0.9325,0.7313888888888889,0.8687330484163021,0.19614193186548523,0.5752659929360795,0.10728036024779791,2.0,5.0,101.0,0.05264079570770264,2.014636993408203e-05
+giroux2022,locoL_fullsat_s20000_K1000,False,0.5427203589458633,0.985,0.7530555555555556,0.9516666666666667,0.7847222222222222,0.8558387167208481,0.20774513671763944,0.5752659929360795,0.13114806465452292,2.0,5.0,101.0,0.0007423162460327148,2.9802322387695312e-06
+giroux2022,locoL_fullsat_s40000_K1000,False,0.5323483591706378,0.9880555555555556,0.7527777777777778,0.9433333333333334,0.7419444444444444,0.8516461288528285,0.20257937229680234,0.5752659929360795,0.13104442637591107,2.0,5.0,101.0,0.00023925304412841797,5.960464477539062e-07
+giroux2022,locoL_fullsat_s60000_K1000,False,0.4988853291728783,0.9863888888888889,0.7563888888888889,0.9338888888888889,0.7263888888888889,0.8548987160712762,0.21660646011764068,0.5752659929360795,0.14055050699438607,2.0,5.0,101.0,0.008714795112609863,1.6689300537109375e-06
+giroux2022,locoL_fullsat_s80000_K1000,False,0.4661782609721092,0.9838888888888889,0.7644444444444445,0.94,0.7291666666666666,0.858248926902685,0.22820748246226313,0.5752659929360795,0.14030919379463738,2.0,5.0,101.0,0.06910169124603271,3.0159950256347656e-05
+giroux2022,locoL_lora16_K1000,False,0.41567310153778664,0.9913888888888889,0.7930555555555555,0.9722222222222222,0.8188888888888889,0.8452838948254842,0.2956654638808478,0.5752659929360795,0.203719618154743,0.0,5.0,101.0,0.15150094032287598,0.0012379884719848633
+giroux2022,locoL_lora16sat_s20000_K1000,False,0.41567310153778664,0.9913888888888889,0.7930555555555555,0.9722222222222222,0.8188888888888889,0.8452838948254842,0.2956654638808478,0.5752659929360795,0.203719618154743,0.0,5.0,101.0,0.15150094032287598,0.0012379884719848633
+giroux2022,locoL_lora16sat_s35750_K1000,False,0.43086767828477845,0.9947222222222222,0.8188888888888889,0.9727777777777777,0.8352777777777778,0.8441236024372596,0.299296791280024,0.5752659929360795,0.20338552802806034,0.0,5.0,101.0,0.15998530387878418,0.0004944801330566406
+giroux2022,locoL_lora64_K1000,False,0.4914305111532487,0.9936111111111111,0.8025,0.9766666666666667,0.8194444444444444,0.835420984705427,0.26353377324727834,0.5752659929360795,0.17065962242627203,2.0,5.0,101.0,0.02112710475921631,0.0001621904716428196
+giroux2022,locoL_lora64sat_s20000_K1000,False,0.4914305111532487,0.9936111111111111,0.8025,0.9766666666666667,0.8194444444444444,0.835420984705427,0.26353377324727834,0.5752659929360795,0.17065962242627203,2.0,5.0,101.0,0.02112710475921631,0.0001621904716428196
+giroux2022,locoL_lora64sat_s40000_K1000,False,0.4692412988819341,0.9941666666666666,0.8180555555555555,0.9766666666666667,0.835,0.8430871064956877,0.23874546841194885,0.5752659929360795,0.15381449914514383,0.0,5.0,101.0,0.033985137939453125,0.00017547607421875
+giroux2022,locoL_lora64sat_s60000_K1000,False,0.42970541878518415,0.9936111111111111,0.8152777777777778,0.975,0.8447222222222223,0.8469462885794707,0.2724235616940377,0.5752659929360795,0.18688971173452829,1.0,5.0,101.0,0.11396121978759766,0.00023925304412841797
+giroux2022,locoL_lora64sat_s69250_K1000,False,0.4098228307587457,0.9944444444444445,0.8086111111111111,0.9811111111111112,0.8411111111111111,0.8504071851340169,0.27821710171594927,0.5752659929360795,0.1866642090555729,0.0,5.0,101.0,0.22918963432312012,0.0035042762756347656
+giroux2022,loco_lora16_K1000,False,0.34714171317520415,0.9911111111111112,0.8077777777777778,0.9630555555555556,0.8263888888888888,0.8467677060411474,0.29602421545115876,0.5752659929360795,0.18828174123576252,0.0,5.0,101.0,0.8333734273910522,0.15998530387878418
+giroux2022,xcalL_K1000,False,0.20217456364466954,0.9880555555555556,0.8911111111111111,0.9791666666666666,0.8886111111111111,0.885349841361257,0.23805572307713363,0.5685800393550717,0.1152602250055946,0.0,2.0,101.0,0.00014960765838623047,0.013754521768656157
+giroux2022,zs_K1000,False,0.33717201742780784,0.9044444444444445,0.2927777777777778,0.9727777777777777,0.36944444444444446,0.889358629460119,0.33988455976970655,0.5752659929360795,0.1974453132347231,0.0,5.0,101.0,,0.03148162364959717
+giroux2022,zs_K2000,False,0.3224497952055856,0.9169444444444445,0.27666666666666667,0.9797222222222223,0.3436111111111111,0.8891586789821797,0.33929379359060685,0.5752659929360795,0.18877795295343638,0.0,5.0,101.0,0.04245976217928354,0.07379257678985596
+giroux2022,[within-cohort calibration] zs__cal,True,0.2663183045196587,0.9947222222222222,0.9061111111111111,0.9805555555555555,0.9019444444444444,0.865104686301811,0.26862025176520443,0.5643395657145305,0.13339193412452294,0.0,2.0,101.0,0.009575724601745605,0.7898378372192383
+giroux2022,[within-cohort LoRA r16] real_lora__raw,True,0.2580442754044374,0.9980555555555556,0.9280555555555555,0.9841666666666666,0.9744444444444444,0.8654960479659805,0.46094848304079133,0.5752659929360795,0.3230347269696575,5.0,5.0,101.0,0.03148162364959717,
+ji2025_chili,locoL_fullsat_s20000_K1000,False,0.8380828593421132,0.7137373737373738,0.1684848484848485,0.649090909090909,0.39191919191919194,0.8263071018497892,0.1866168577702635,0.2398746224459366,0.13985235513909222,4.0,4.0,250.0,1.0028015822172165e-06,0.0009519711602479219
+ji2025_chili,locoL_fullsat_s40000_K1000,False,1.0064364587606467,0.6105050505050506,0.11212121212121212,0.6103030303030303,0.2816161616161616,0.8176942881361339,0.1487554246967952,0.2398746224459366,0.10514260743112055,2.0,4.0,250.0,4.656612873077393e-10,1.2829434126615524e-05
+ji2025_chili,locoL_fullsat_s60000_K1000,False,1.0320664027955349,0.5414141414141415,0.08363636363636363,0.49414141414141416,0.23030303030303031,0.8144120252926536,0.1160886730996881,0.2398746224459366,0.07874720428226975,0.0,4.0,250.0,4.656612873077393e-10,5.628913640975952e-06
+ji2025_chili,locoL_fullsat_s80000_K1000,False,1.0518949849638783,0.5119191919191919,0.07212121212121213,0.4602020202020202,0.20040404040404042,0.8147569778739214,0.13200511557944813,0.2398746224459366,0.09531932418029626,0.0,4.0,250.0,2.3283064365386963e-10,2.304092049598694e-06
+ji2025_chili,locoL_fullsat_s86750_K1000,False,1.0304931953192484,0.5107070707070707,0.08121212121212121,0.5072727272727273,0.2014141414141414,0.8170674801777872,0.1386792074228318,0.2398746224459366,0.10111202584513725,5.0,4.0,250.0,4.656612873077393e-10,4.976987838745117e-06
+ji2025_chili,locoL_lora16_K1000,False,0.7572057975657671,0.8371717171717171,0.27090909090909093,0.7256565656565657,0.40040404040404043,0.8313748974629823,0.14448522107370879,0.2398746224459366,0.10236673594628912,2.0,4.0,250.0,0.089798514964059,0.0020859958603978157
+ji2025_chili,locoL_lora16sat_s20000_K1000,False,0.7572057975657671,0.8371717171717171,0.27090909090909093,0.7256565656565657,0.40040404040404043,0.8313748974629823,0.14448522107370879,0.2398746224459366,0.10236673594628912,2.0,4.0,250.0,0.089798514964059,0.0020859958603978157
+ji2025_chili,locoL_lora16sat_s40000_K1000,False,0.7395379838267597,0.9105050505050505,0.3387878787878788,0.7658585858585859,0.42404040404040405,0.840793406384541,0.045342224880778284,0.2398746224459366,0.007343118962226443,0.0,4.0,250.0,0.7643533307127655,0.0008192581590265036
+ji2025_chili,locoL_lora64_K1000,False,0.6387148684316141,0.9309090909090909,0.3814141414141414,0.876969696969697,0.5381818181818182,0.8425369225194382,0.1991321409551958,0.2398746224459366,0.15108174654279038,2.0,4.0,250.0,0.005484450375661254,0.04842899041250348
+ji2025_chili,locoL_lora64sat_s20000_K1000,False,0.6387148684316141,0.9309090909090909,0.3814141414141414,0.876969696969697,0.5381818181818182,0.8425369225194382,0.1991321409551958,0.2398746224459366,0.15108174654279038,2.0,4.0,250.0,0.005484450375661254,0.04842899041250348
+ji2025_chili,locoL_lora64sat_s40000_K1000,False,0.6684182614003742,0.9476767676767677,0.38747474747474747,0.8951515151515151,0.5527272727272727,0.8469107999233323,0.17688721837501256,0.2398746224459366,0.1292245973429637,0.0,4.0,250.0,0.03071066946722567,0.029283775947988033
+ji2025_chili,locoL_lora64sat_s51750_K1000,False,0.6199837283023538,0.9470707070707071,0.39515151515151514,0.9183838383838384,0.5721212121212121,0.8462697758784372,0.16414840904557088,0.2398746224459366,0.12651420594700494,0.0,4.0,250.0,0.0015806348528712988,0.05513314809650183
+ji2025_chili,loco_lora16_K1000,False,0.4942971856123803,0.8452525252525253,0.31272727272727274,0.7476767676767677,0.5208080808080808,0.8363609394362492,0.08922685293161287,0.2398746224459366,0.04166517163260728,9.0,4.0,250.0,0.0006023156456649303,0.9578276048414409
+ji2025_chili,xcalL_K1000,False,0.5647530791873953,0.6357575757575757,0.07111111111111111,0.563030303030303,0.27414141414141413,0.8277413131675961,0.12672949998502117,0.25614963059319984,0.08965150980608351,0.0,4.0,250.0,0.0002449189778417349,0.29599490854889154
+ji2025_chili,zs_K1000,False,0.7246556382795132,0.7212121212121212,0.05898989898989899,0.4991919191919192,0.19474747474747475,0.8343238300989376,0.07494970415326974,0.2398746224459366,0.022996024886643866,0.0,4.0,250.0,,0.008788360049948096
+ji2025_chili,zs_K2000,False,0.7294414806659761,0.7185858585858586,0.060404040404040404,0.4391919191919192,0.17272727272727273,0.8359971453700948,0.07560473673816925,0.2398746224459366,0.02285003393821291,0.0,4.0,250.0,0.5375959745066456,0.006567624397575855
+ji2025_chili,[within-cohort calibration] zs__cal,True,0.5005691181619872,0.74,0.10646464646464647,0.6703030303030303,0.3472727272727273,0.8385862483484186,0.14191696206220386,0.20647363470913363,0.11277544159311834,0.0,1.0,250.0,0.0011035518255084753,0.9578276048414409
+ji2025_chili,[within-cohort LoRA r16] real_lora__raw,True,0.5044992874562503,0.9709090909090909,0.8569696969696969,0.9727272727272728,0.9648484848484848,0.8728442542892082,0.47928798340225065,0.2398746224459366,0.45204372964017614,15.0,4.0,250.0,0.008788360049948096,

paper-releases/pbisc-diffusion/updates/20261003T105229Z-deck-cross-cohort-loco-k1000/diffusion/dataset/server_1/model8/plans/paired_crosscohort_20261002/scores/immicro/zs_K2000/README.md

@@ -0,0 +1,60 @@
+# score_ft.py output: immicro, 107 scored donors
+
+Arms: zs_K2000.  Every arm is scored on the same donors against the same observed reference.
+
+## Inputs and fixed contracts
+- Observed reference: `/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/scoring_cache/immicro/reference.npz` (sha c5e2547be3ec), built from datadir:/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_finetune_intradataset_20260927/data/immicro;
+  observed cells sum exactly to the features pseudobulk of every donor. Observed labels: /home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/utility_studies_final6k_20260925/immicro/labels/observed_immicro/labels.parquet.
+- Generated cells: `generated_counts` read out once with Poisson(seed 20260916) over the **whole NPZ in row order**
+  (`poisson_readout.poisson_integerise`), then the scored donors' rows. These integer counts are used for labels, embedding,
+  depth and the contrast. Pseudobulk uses the summed `generated_counts` (rates), as `score_generated.py`.
+- Labels: frozen f01common scANVI (`predict_scanvi.py`, CPU, zero query epochs), six-type gate ['T cell', 'B cell', 'Monocyte', 'NK cell', 'DC', 'Other']; query genes =
+  the 1,006-gene annotation list minus genes the dataset's sc assay did not measure in every sample (1004 genes).
+- Scoring genes: 6k panel genes measured by both assays in every sample of the dataset (5508 genes).
+
+## Definitions
+1. **Composition**: fraction of each donor's cells per six-type label (all generated cells; all observed QC cells).
+   `comp_l1` = sum over the six labels of |generated - observed| (immicro_common.l1). `comp_l1_obs_loo_null` = L1 of the
+   donor's observed composition to the mean of the *other scored donors'* observed compositions. `composition_top1` =
+   fraction of donors whose generated composition is closest (L1) to their own observed composition.
+2. **Pseudobulk**: `pb_r` = Pearson r of CPM-log1p(summed generated_counts) vs CPM-log1p(observed pseudobulk) on the scoring
+   genes (evaluate_realbulk.cpm_log1p / paired_r). `pb_top1` = the donor's generated pseudobulk is closest (Pearson) to its own
+   observed pseudobulk among the 107 scored donors (immicro_common.retrieval = evaluate_realbulk.top1 arithmetic).
+3. **Precision / coverage / density** (stage_b_embed.py functions): per-cell CP10K over the scoring genes + log1p; PCA-50
+   (randomized, random_state 20260925) and k = 10 real-real k-NN balls fitted on **R1 of the scored donors only**.
+   Reference rule (independent of fold and arm): per donor, cells in canonical order (library_id+barcode), one
+   permutation from `default_rng([20260925, uint64(sha256(donor)[:8])])`; R1 = first n, R2 = next n,
+   n = min(150, n_cells // 2) (16050 R1 cells here). Generated query: per donor n cells (the donor's R1 size),
+   without replacement, `default_rng([20260926, same key])`. Precision = fraction of generated cells inside any R1 ball
+   (Kynkaanniemi 2019); coverage = fraction of R1 cells whose ball holds a generated cell; density = Naeem 2020.
+   Per-donor precision = that donor's generated cells; per-donor coverage = that donor's R1 cells (the pooled values are
+   their n-weighted means; plain means when every donor has 150 reference cells). R2 (held-out real cells of the same donors) is the
+   perfect-generator control: precision 0.9634, coverage 0.9999.
+   **Depth matched**: stage_b_embed depth matching run pairwise (R1 vs this arm): per donor and within-donor rank of detected
+   genes, both cells are thinned by UMI sampling without replacement to the smaller detected-gene count; PCA kept from
+   un-thinned R1, radii recomputed on thinned R1; fresh rng(20260929) per arm. R2 control: precision
+   0.9635, coverage 0.9997.
+4. **Monocyte %** = 100 x Monocyte fraction (all cells) per donor; **median UMI per cell** = median over cells of the
+   Poisson-read (generated) / observed counts summed over the scoring genes (per donor, and pooled).
+5. **Group contrast** (T, B, Monocyte, NK): not computed (dataset immicro has no group contrast (use --groups-csv / --case-groups)).
+   Genes per type = scoring genes detected (count > 0) in >= 1 % of the scored donors' observed
+   cells of that type. Primary statistic `mean_cp10k`: per-cell CP10K over the scoring genes; per donor the mean over its
+   cells of the type (donors with >= 1 such cell); group mean = mean over the group's donors (donor-balanced);
+   logFC = log2(case mean + 1) - log2(control mean + 1). Secondary `pseudobulk_log1p_cpm`: the SLE benchmark evaluator's
+   statistic (donor x type summed counts, log1p CPM over the scoring genes, mean case - mean control;
+   evaluate_generated_vs_truth.group_log_fold_change). Both are computed on the generated cells (their frozen labels) and on
+   the observed cells; `r_generated` = Pearson r of the two logFC vectors. Broadcast baseline `r_broadcast`: the same
+   observed vector against the logFC of the input bulk (condition NPZ `bulk_counts` of the arm, scoring genes, per-donor
+   CP10K resp. log1p CPM, group means). `partial_r_generated_given_broadcast`: r of the OLS residuals of observed and
+   generated logFC on the bulk logFC (residual_specificity.py).
+6. **Top-50 DEG overlap** (contrast.csv rows `statistic = welch_top50_deg`; summary `deg_top50`): not computed (contrast off).
+
+## Files
+- `per_donor.csv`: one row per arm x donor (all per-donor quantities above).
+- `summary.json`: per arm the donor means / pooled values, provenance (NPZ sha, receipt check, generation metadata,
+  Poisson nnz of the whole NPZ, annotation contract), and the reference block (PCA, R2 control).
+- `contrast.csv`: per arm x cell type x statistic (`mean_cp10k`, `pseudobulk_log1p_cpm`: logFC r; `welch_top50_deg`:
+  top-50 DEG overlap, T / B / Monocyte / NK / DC; columns of the other kind are empty).
+- `arms/<arm>/labels/` (predict_scanvi.py labels + manifest), `arms/<arm>/generated_cells.parquet` (NPZ row, donor, label,
+  UMI on the scoring genes, P/C subsample flag), `arms/<arm>/deg_top50_lists.csv` (observed, generated and broadcast
+  top-50 gene lists, Ensembl ids, rank order).

paper-releases/pbisc-diffusion/updates/20261003T105229Z-deck-cross-cohort-loco-k1000/diffusion/dataset/server_1/model8/plans/paired_crosscohort_20261002/tools/build_archive_request.py

@@ -0,0 +1,122 @@
+#!/usr/bin/env python3
+"""Archive-maintenance request for the cross-cohort (LOCO) benchmark (PBISC-Diffusion paper, pbisc-diffusion).
+
+Follows paper-release-20260928/maintenance/POLICY.md.  Payloads (completed): code, README, queues, splits / prepare
+manifests, run receipts (RUN.json, history.json, milestones.json, train.log; every fine-tuned checkpoint's sha256 is in
+RUN.json), score tables, aggregates and the LoRA integrity audit.  Not uploaded: fine-tuned checkpoints and generated
+NPZs (location + receipt only).  Generated NPZs are on server_1, or on server_2 at the same path after a sha-checked
+move (tools/offload_scored.sh leaves <name>.moved on server_1).  Queue units not finished yet are registered as planned while claims/.queue_live exists, else as stopped.
+Condition NPZs and Liu inputs were published in earlier deltas; unchanged files are skipped by update.py.
+usage: build_archive_request.py OUT.json "summary text"
+"""
+from __future__ import annotations
+
+import glob
+import json
+import sys
+from pathlib import Path
+
+P = Path("/home/sj_server_1/PBISC/immun-pbmc-fm/model_8_sparse_data_diffusion/plans/paired_crosscohort_20261002")
+EXPERIMENT = "paired_crosscohort_loco_20261002"
+
+
+def files(patterns):
+    out = []
+    for pat in patterns:
+        for f in sorted(glob.glob(str(P / pat), recursive=True)):
+            q = Path(f)
+            if q.is_file() and not q.is_symlink() and "__pycache__" not in q.parts and ".bak" not in q.name \
+                    and not q.name.endswith(".tmp"):
+                out.append(str(q))
+    return sorted(set(out))
+
+
+def main(out: str, summary: str) -> int:
+    assets = []
+
+    def add(paths, kind):
+        for p in paths:
+            assets.append({"host": "server_1", "source": p, "kind": kind, "state": "completed"})
+
+    add(files(["tools/*.py", "tools/*.sh"]), "code")
+    add(files(["README.md"]), "document")
+    add(files(["queues/*.txt"]), "config")
+    add(files(["PREPARE.json", "PREPARE_EXTRA.json", "splits/*.json", "data/*/*.txt", "audit/*.json"]), "metadata")
+    add(files(["runs/**/RUN.json", "runs/**/history.json", "runs/**/milestones.json", "runs/**/train.log",
+               "audit/*.log"]), "log")
+    add(files(["scores/aggregate.csv", "scores/AGGREGATE.md", "scores/k1000_saturated_summary.csv",
+               "scores/*/*/summary.json", "scores/*/*/per_donor.csv", "scores/*/*/contrast.csv", "scores/*/*/README.md",
+               "scores_nosex/*/*/summary.json", "scores_nosex/*/*/per_donor.csv", "scores_nosex/*/*/contrast.csv",
+               "scores_nosex/*/*/README.md"]), "table")
+
+    locs, seen = [], set()
+    for rec in sorted(glob.glob(str(P / "generated" / "**" / "*.log"), recursive=True)):
+        try:
+            r = json.loads(Path(rec).read_text())
+        except Exception:
+            continue
+        if not isinstance(r, dict) or not r.get("output_sha256"):
+            continue
+        npz = rec[:-4] + ".npz"
+        moved = Path(rec[:-4] + ".moved")
+        if Path(npz).is_file():
+            host = "server_1"
+        elif moved.is_file():
+            host = "server_2"
+        else:
+            continue
+        rel = Path(npz).relative_to(P / "generated")
+        item = {"host": host, "path": npz, "state": "completed", "experiment": EXPERIMENT, "cohort": rel.parts[0],
+                "artifact_role": "inference_output", "receipt": rec, "sha256": r["output_sha256"]}
+        for k in ("output_sha256", "checkpoint", "checkpoint_sha256", "input_condition_path", "input_condition_sha256",
+                  "condition_npz", "condition_sha256", "seed", "steps"):
+            if k in r and not isinstance(r[k], (dict, list)):
+                item[k] = r[k]
+        if host == "server_2":
+            item["move_marker_server_1"] = str(moved)
+        locs.append(item)
+        seen.add(npz)
+    # unfinished units of the active queue: planned
+    for q in ("loco_liu_sat_20261003.txt",):
+        for line in (P / "queues" / q).read_text().splitlines():
+            if not line or line.startswith("#"):
+                continue
+            f = line.split("|")
+            if f[0] == "zs":
+                outs, cohort = [P / "generated" / f[1] / f"zs_K{f[2]}.npz"], f[1]
+            else:
+                cohort = json.loads((P / "splits" / f"{f[1]}.json").read_text())["held_out"]
+                if f[0] == "xcal":
+                    outs = [P / "generated" / cohort / f[1] / f"xcal_K{f[2]}.npz"]
+                else:   # milestone outputs are unknown until training ends: register the run's output stem
+                    outs = [P / "generated" / cohort / f[1] / f"K{f[2]}_{f[3]}.npz"]
+                    if glob.glob(str(outs[0])[:-4] + "_s*.log"):
+                        continue
+            key = ("zs_%s_K%s" % (f[1], f[2]) if f[0] == "zs" else "xcal_%s_K%s" % (f[1], f[2]) if f[0] == "xcal"
+                   else "loco_%s_K%s_%s" % (f[1], f[2], f[3]))
+            queue_live = (P / "claims" / ".queue_live").exists()
+            for o in outs:
+                if str(o) in seen:
+                    continue
+                if queue_live:
+                    state, note = "planned", ("queued or running at request time (run_loco_queue.sh, server_1); a "
+                                              "step-saturation run writes one output per milestone, <stem>_s<step>.npz")
+                else:
+                    state, note = "stopped", ("not run: user decision 2026-10-03 19:45, the cross-cohort benchmark is "
+                                              "reported at K = 1000 only; queue stopped (claim " + key + ")")
+                locs.append({"host": "server_1", "path": str(o), "state": state, "experiment": EXPERIMENT,
+                             "cohort": cohort, "artifact_role": "inference_output", "note": note})
+                seen.add(str(o))
+    req = {"project_id": "pbisc-diffusion", "summary": summary, "assets": assets, "inference_locations": locs}
+    Path(out).write_text(json.dumps(req, indent=1))
+    kinds, states = {}, {}
+    for a in assets:
+        kinds[a["kind"]] = kinds.get(a["kind"], 0) + 1
+    for l in locs:
+        states[(l["host"], l["state"])] = states.get((l["host"], l["state"]), 0) + 1
+    print(json.dumps({"assets": len(assets), "by_kind": kinds, "locations": {f"{h}/{s}": n for (h, s), n in states.items()}}))
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main(sys.argv[1], sys.argv[2]))