OUTPUT of run 12 of at most 60:
```
IDENTITY COUPLINGS WITH DIGIT-EXCLUSION CONTROLS
For each category: normalized B-A coupling *1e4 / entry-sign-null score.

 L16H3
 digits_0_to_9   :   +0.3458 /   +7.30
 digits_1_to_9   :   +0.0595 /   +1.24
 digits_1_to_8   :   +0.0712 /   +1.43
 number_words    :   +0.0629 /   +1.54
 ABCD            :   +0.0409 /   +1.04
 boolean         :   +0.0554 /   +1.48

 L17H2
 digits_0_to_9   :   +0.0990 /   +3.94
 digits_1_to_9   :   +0.1117 /   +4.25
 digits_1_to_8   :   +0.1128 /   +4.18
 number_words    :   +0.0555 /   +3.01
 ABCD            :   +0.0852 /   +3.98
 boolean         :   +0.0290 /   +1.50

 L16MLP_ungated
 digits_0_to_9   :   -0.1451 /   -2.88
 digits_1_to_9   :   -0.2082 /   -3.85
 digits_1_to_8   :   -0.1984 /   -3.54
 number_words    :   -0.1396 /   -3.37
 ABCD            :   +0.0560 /   +1.37
 boolean         :   +0.0067 /   +0.17

 L17MLP_ungated
 digits_0_to_9   :   +0.4087 /   +7.06
 digits_1_to_9   :   +0.5146 /   +8.20
 digits_1_to_8   :   +0.5849 /   +9.00
 number_words    :   +0.1807 /   +3.66
 ABCD            :   +0.0701 /   +1.66
 boolean         :   -0.0109 /   -0.23

NUMERIC DIFFERENCE-MATRIX FACTORS
These describe static weight coupling between digit identities.

 L16H3 digit order: ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
B-A diagonal: [11.5949  1.5801  0.8645 -2.1269 -0.0729  1.1387 -0.0198  1.5073 -0.0404  0.4276]
First three singular energy fractions: [0.79   0.1082 0.0492]
 mode1 input factor: [ 0.7634  0.2541  0.2821 -0.1808 -0.1207 -0.2007 -0.2085 -0.2726 -0.0641 -0.2523]
 mode1 output factor: [ 0.7738 -0.1535  0.2453  0.0968  0.0001 -0.2056  0.0269 -0.2317 -0.4479 -0.1042] trace contribution: 13.95863409331019
 mode2 input factor: [ 0.1485 -0.5039  0.0964  0.2132  0.4442  0.261   0.053   0.0751 -0.1872 -0.6003]
 mode2 output factor: [ 0.3585 -0.6216 -0.1601 -0.341   0.0912  0.2107 -0.2072  0.2051  0.4531  0.0113] trace contribution: 1.9554137706813326

 L17H2 digit order: ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
B-A diagonal: [ 0.5015  1.3062  0.8117  0.4548  0.2596  0.2507 -0.1563  0.2198  0.4223  0.1837]
First three singular energy fractions: [0.7726 0.1055 0.0586]
 mode1 input factor: [ 0.3176  0.5489  0.5204 -0.0399 -0.1691 -0.2855 -0.1557 -0.2888 -0.285  -0.1627]
 mode1 output factor: [ 0.3955  0.5358  0.4642 -0.0863 -0.205  -0.2843 -0.0471 -0.2702 -0.319  -0.1837] trace contribution: 3.5912111334106345
 mode2 input factor: [ 0.3374  0.1125 -0.1512 -0.6863 -0.457   0.0964  0.085   0.1741  0.1978  0.2913]
 mode2 output factor: [-0.1625  0.0688  0.0358 -0.478   0.0174 -0.4735  0.3557 -0.1863  0.3248  0.4978] trace contribution: 0.5770491065697656

 L16MLP_ungated digit order: ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
B-A diagonal: [ 0.5915  3.3136 -3.3681 -0.0192 -0.5361 -1.2147 -1.8856 -0.8944 -1.6119 -0.6079]
First three singular energy fractions: [0.5487 0.3251 0.0707]
 mode1 input factor: [ 0.1727  0.5706  0.506   0.0719 -0.0429 -0.3011 -0.3956 -0.2444 -0.2631 -0.0741]
 mode1 output factor: [-0.4993  0.2724 -0.5973 -0.1926  0.04    0.1118  0.3999  0.2694  0.0058  0.1898] trace contribution: -7.291789911295797
 mode2 input factor: [ 0.7025  0.199  -0.466  -0.3135 -0.1316 -0.0358 -0.1878 -0.048   0.3076 -0.0265]
 mode2 output factor: [ 0.1558  0.7103  0.037   0.0136 -0.0564  0.1779  0.0137 -0.1921 -0.5482 -0.3117] trace contribution: 0.8236671770674305

 L17MLP_ungated digit order: ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']
B-A diagonal: [ 0.4757  9.3635  1.4514  1.5914 -0.739  -0.7829  1.2992  3.1535  2.2231 -0.4812]
First three singular energy fractions: [0.7656 0.1117 0.093 ]
 mode1 input factor: [ 0.5622  0.3816  0.2095  0.2517 -0.0182 -0.22   -0.4195 -0.3638 -0.2063 -0.1773]
 mode1 output factor: [-0.1244  0.8781  0.1404 -0.1055 -0.074   0.0198 -0.0362 -0.2235 -0.3173 -0.1575] trace contribution: 13.76233273431307
 mode2 input factor: [ 0.4358 -0.0993 -0.7554  0.0859  0.0516  0.1625 -0.088   0.1318  0.3249 -0.2498]
 mode2 output factor: [ 0.8037  0.1429 -0.281  -0.1896 -0.4175 -0.1406 -0.0961  0.0125  0.0547  0.1109] trace contribution: 5.63683895940046
Complete B-A digit matrix:
 [[  0.4757  -0.8697  -8.693    0.6266   0.3739   2.6681   0.1339   1.7005   4.5287  -0.9447]
 [ 16.3306   9.3635   4.8529   6.3546  -0.3763  -5.9994 -10.9771  -8.8582  -5.0301  -5.6605]
 [ -2.2086   2.6259   1.4514   5.5077  -0.4058   0.0041  -3.6803  -3.1907  -2.2771   2.1735]
 [ -5.267    0.0643  -0.2943   1.5914   1.3509   1.134    0.23     0.126   -0.4035   1.4681]
 [ -3.4531   1.3075   3.1778  -2.552   -0.739   -1.2538   1.8715   0.114    0.3536   1.1734]
 [  1.0198   0.741    2.1117  -2.6803  -0.1109  -0.7829   1.2508  -0.2032  -0.4112  -0.9349]
 [  1.4268  -2.0804   1.7659  -2.939   -0.4653   0.3831   1.2992   0.0616  -0.2571   0.8053]
 [ -1.7369  -3.4859  -1.0477  -3.229   -1.0202   1.0528   3.7979   3.1535   0.8179   1.6976]
 [ -5.2053  -5.4652  -1.6295  -0.7669   0.3288   2.0226   3.3033   4.4858   2.2231   0.7034]
 [ -1.382   -2.201   -1.6952  -1.913    1.0639   0.7714   2.7707   2.6108   0.4555  -0.4812]]

FINAL MLP: TOP POSITIVE AND NEGATIVE DIGIT-CONNECTION UNITS
Coefficients below are separate projections of weight rows/columns.
No gate and up projections are multiplied together.
Digit order: ['0', '1', '2', '3', '4', '5', '6', '7', '8', '9']

unit 1180: mean matched-digit coupling A=-7.72332, B=-7.49270, delta=+0.23062
 down-change contribution: 0.23669103762595137 up-change contribution: -0.006075186824516789
 mean up-row projections: [ 0.2654  0.6155  0.0501 -0.1622 -0.2326 -0.1687 -0.1499 -0.1369 -0.0795 -0.0011]
 mean gate-row projections: [-0.3355 -1.2254 -0.5041  0.0768  0.4666  0.3933  0.3136  0.3319  0.367   0.116 ]
 mean output-column projections: [ 0.2898 -2.971   0.468   0.662   0.2441  0.3373  0.2431  0.1526  0.2274  0.3467]
 B-A output-column projections: [ 0.0373  0.0629 -0.0353 -0.0147 -0.0563  0.0063  0.0113 -0.0209 -0.0075  0.0169]

unit 1905: mean matched-digit coupling A=+0.68198, B=+0.85556, delta=+0.17358
 down-change contribution: 0.17319108707501552 up-change contribution: 0.00038798886207569666
 mean up-row projections: [-0.1761 -0.2557 -0.1068  0.0317  0.0946  0.1074  0.1246  0.1276  0.0381  0.0147]
 mean gate-row projections: [ 0.0855  0.0738  0.0693 -0.0515 -0.0014 -0.007  -0.0446 -0.0984 -0.0403  0.0144]
 mean output-column projections: [-0.1368 -0.1985 -0.4171  0.0855 -0.3314  0.2932  0.5335  0.1152  0.0145  0.0419]
 B-A output-column projections: [-0.014  -0.0869 -0.0685 -0.0209  0.0183  0.0268  0.0201  0.0436  0.0374  0.0442]

unit 1712: mean matched-digit coupling A=+2.13462, B=+2.30292, delta=+0.16830
 down-change contribution: 0.15116034822635951 up-change contribution: 0.01713560234335997
 mean up-row projections: [ 0.2746  0.2243  0.2345 -0.0082 -0.0234 -0.0935 -0.1496 -0.2099 -0.1664 -0.0825]
 mean gate-row projections: [ 0.0662  0.0373  0.007  -0.0422 -0.0654 -0.0273 -0.0035 -0.0076  0.0467 -0.0114]
 mean output-column projections: [ 0.6694  0.2859  0.5954 -0.161   0.0353 -0.2636 -0.2261 -0.3307 -0.2941 -0.3104]
 B-A output-column projections: [ 0.025   0.0492  0.0344  0.035  -0.0297 -0.0199 -0.0175 -0.0107 -0.0205 -0.0451]

unit 1309: mean matched-digit coupling A=-14.41574, B=-14.72979, delta=-0.31405
 down-change contribution: -0.2919001394316708 up-change contribution: -0.022147147493003127
 mean up-row projections: [ 0.9043  0.1692  0.118  -0.205  -0.169  -0.241  -0.1132 -0.2574 -0.1075 -0.0983]
 mean gate-row projections: [-1.1226 -0.7053 -0.4513  0.2029  0.301   0.3969  0.268   0.4525  0.3518  0.3062]
 mean output-column projections: [-4.1399  0.5647  1.3351  0.7542  0.6616  0.4119  0.2131 -0.1376  0.0805  0.2563]
 B-A output-column projections: [-0.0812  0.0051  0.033   0.02   -0.0017  0.0054  0.0141  0.011  -0.0011 -0.0045]

unit 1192: mean matched-digit coupling A=-2.73377, B=-2.91274, delta=-0.17897
 down-change contribution: -0.16689333749732246 up-change contribution: -0.012076108514208626
 mean up-row projections: [-0.0765  0.097   0.3534  0.0749 -0.0484 -0.0971 -0.0528 -0.0932 -0.0747 -0.0826]
 mean gate-row projections: [-0.4547 -0.1321  0.4232  0.2633  0.1255 -0.0026 -0.0358 -0.0688 -0.0499 -0.068 ]
 mean output-column projections: [ 0.1198 -0.105  -1.6608 -0.1546  0.3639  0.2985  0.2536  0.2631  0.2991  0.3224]
 B-A output-column projections: [ 0.0037 -0.0838 -0.0814  0.0362  0.005   0.0114  0.0077  0.018   0.0428  0.0404]

unit 412: mean matched-digit coupling A=+0.16888, B=+0.00751, delta=-0.16137
 down-change contribution: -0.1616480518211045 up-change contribution: 0.00028186570830932613
 mean up-row projections: [ 0.1908  0.2103  0.2075 -0.0515 -0.0559 -0.0564 -0.0925 -0.1412 -0.0965 -0.1147]
 mean gate-row projections: [-0.0225 -0.0109  0.0824  0.0793 -0.0219 -0.0699  0.0059  0.0079 -0.06    0.0098]
 mean output-column projections: [-0.0294  0.0781  0.0104 -0.0089  0.0627 -0.0154  0.1927  0.0594 -0.0689 -0.2806]
 B-A output-column projections: [-0.0408 -0.0587 -0.0479  0.0129  0.0204  0.0296  0.0144  0.0191  0.023   0.0281]

SEPARATE HEAD-TO-GATE WEIGHT PATHS
Transfer = final-MLP gate row @ sender output/value matrices.
Columns: belief-question, belief-fact. Values are geometric weight couplings.
Attention selection and intervening nonlinear paths are NOT included.

Sender L0H3
 unit 1180: A=[0.0127 0.066 ], B=[0.0121 0.0654], 1e4*(B-A)=[-6.7573 -5.5251]
 unit 1905: A=[-0.0156 -0.0205], B=[-0.0152 -0.0205], 1e4*(B-A)=[3.9305 0.1554]
 unit 1712: A=[-0.0168  0.0222], B=[-0.0161  0.022 ], 1e4*(B-A)=[ 6.9616 -2.2529]
 unit 1309: A=[0.0456 0.0039], B=[0.0446 0.0038], 1e4*(B-A)=[-9.6935 -0.9278]
 unit 1192: A=[-0.0094  0.021 ], B=[-0.0096  0.0211], 1e4*(B-A)=[-2.5064  0.8286]
 unit 412: A=[0.0196 0.0043], B=[0.0195 0.0047], 1e4*(B-A)=[-0.0789  4.0271]

Sender L1H1
 unit 1180: A=[-0.0338  0.0407], B=[-0.0359  0.0405], 1e4*(B-A)=[-21.5339  -1.677 ]
 unit 1905: A=[-0.0195 -0.0221], B=[-0.0188 -0.0224], 1e4*(B-A)=[ 7.0562 -3.9733]
 unit 1712: A=[0.0769 0.0276], B=[0.0758 0.0277], 1e4*(B-A)=[-11.3323   0.7759]
 unit 1309: A=[0.0257 0.004 ], B=[0.0264 0.0032], 1e4*(B-A)=[ 6.9002 -8.6266]
 unit 1192: A=[ 0.0241 -0.0069], B=[ 0.0229 -0.0074], 1e4*(B-A)=[-11.6421  -4.9403]
 unit 412: A=[-0.0404 -0.0001], B=[-0.0387  0.0003], 1e4*(B-A)=[16.0728  3.7742]

Sender L5H0
 unit 1180: A=[ 0.0086 -0.0531], B=[ 0.0081 -0.0524], 1e4*(B-A)=[-4.4827  6.7712]
 unit 1905: A=[-0.0023  0.0453], B=[-0.0016  0.0456], 1e4*(B-A)=[6.7605 3.4326]
 unit 1712: A=[-0.0319 -0.0075], B=[-0.0316 -0.0082], 1e4*(B-A)=[ 3.2984 -7.4737]
 unit 1309: A=[ 0.0012 -0.0729], B=[ 0.0009 -0.0733], 1e4*(B-A)=[-2.9902 -3.8548]
 unit 1192: A=[-0.0113  0.0483], B=[-0.0112  0.0482], 1e4*(B-A)=[ 0.9601 -1.8073]
 unit 412: A=[-0.0222 -0.0276], B=[-0.0228 -0.0265], 1e4*(B-A)=[-5.7852 10.784 ]

Sender L6H0
 unit 1180: A=[0.0047 0.0381], B=[0.0054 0.0382], 1e4*(B-A)=[7.1784 0.5937]
 unit 1905: A=[ 0.0292 -0.0183], B=[ 0.0283 -0.0196], 1e4*(B-A)=[ -9.811  -13.0772]
 unit 1712: A=[0.01   0.0849], B=[0.0088 0.0842], 1e4*(B-A)=[-11.7029  -7.5141]
 unit 1309: A=[0.0176 0.0328], B=[0.0184 0.0325], 1e4*(B-A)=[ 7.8871 -3.2407]
 unit 1192: A=[-0.0003  0.0147], B=[0.0004 0.0146], 1e4*(B-A)=[ 7.1953 -1.3441]
 unit 412: A=[0.0305 0.0116], B=[0.0312 0.0119], 1e4*(B-A)=[6.997  2.8716]

```