$ python followup_report.py
SIX-HOP PATH / sixteen nodes / 2,048 test inputs
Exact complete path, percent correct
training seed    loop 6    loop 16    loop 32    loop 64
fixed      11      99.41      35.94      12.65       8.79
fixed      22     100.00      99.32      89.50      64.50
fixed      33      99.90      56.93       7.96       1.46
mixed      11      99.85      99.46      97.66      92.87
mixed      22      99.95      99.46      96.39      89.55
mixed      33     100.00      99.90      98.54      90.09

16-BIT PAYLOAD / neighbour-only attention
Training curriculum: 4, 8, 16, 32 links; 2 loops per link
512 distinct held-out payloads; complete answers, percent correct
seed links      32      48      64      96     128     256
  11    16    0.59    0.78    1.56    0.20    0.00    0.00
  11    32    0.00    0.00    9.77    0.00    0.00    0.00
  11    64    0.00    0.00    0.00    0.00    0.00    0.00
  22    16    0.00    0.39    0.39    0.00    0.00    0.00
  22    32    0.00    0.00    0.39    0.00    0.00    0.00
  22    64    0.00    0.00    0.00    0.00    0.20    0.00
  33    16    0.98    2.15    7.42    0.00    0.00    0.00
  33    32    0.00    0.00   79.30    0.00    0.00    0.00
  33    64    0.00    0.00    0.00    0.00    0.00    0.00

32-link trace / seed 11 / item 2
expected: 0100010011000011
loops  answer             matching bits
   16  0111011100000010    9/16
   32  0011011100000010    8/16
   48  0011001110000110    7/16
   64  0100010011000011   16/16
   96  0100010011001000   13/16
  128  1100111010001000    9/16
  256  0001000000110101    7/16
PASS: stored predictions, labels, causal boundary and checkpoint replay
