== w1m repair threads=16  CPU-s: median of both copies of each arm (n legs); A/A = worst |X - X2| / min over the reps; verdict clears the A/A floor or is UNRESOLVED
      m |     fold    force     F/T |   aa_F   aa_T  floor | verdict |      wF      wT  wF/wT | foreign max | peak MB force/fold
    320 |   272.48   283.39   0.961 |   3.2%   2.1%   3.2% | fold    |   27.82   27.56  1.010 | 103% (4 legs) | 4750/924
    352 |   280.37   273.53   1.025 |   7.2%   4.2%   7.2% | unres   |   26.89   41.96  0.641 | 25% (4 legs) | 4781/990
    384 |   314.38   288.16   1.091 |   5.7%   3.0%   5.7% | ntt     |   24.79   24.51  1.011 | 16% (4 legs) | 4817/1057
    416 |   342.94   297.38   1.153 |   0.8%   0.4%   0.8% | ntt     |   24.85   23.83  1.043 | 34% (4 legs) | 4852/1125
    448 |   366.37   298.66   1.227 |   0.1%   0.7%   0.7% | ntt     |   26.11   24.17  1.080 | 16% (4 legs) | 4888/1192
  crossover (log-interpolated, median): CPU m ~ 339   wall m ~ 383
  foreign CPU over 20 legs: median 13% of a core, max 103% (after-leg median 12%)