== rules only
cases.jsonl: 43/43 pass
== rules + both judges
cases.jsonl: 41/43 pass
  clean           16/18
  wrong_fact      10/10
  advice          6/6
  promissory      4/4
  out_of_scope    2/2
  unknown_fact    2/2
  silent_failure  1/1
  FAIL c06 SEND REVIEW ['unknown_fact'] 
  FAIL c13 SEND REVIEW ['unknown_fact'] 
== rules only
heldout.jsonl: 18/20 pass
== rules + both judges
heldout.jsonl: 20/20 pass
  clean           8/8
  wrong_fact      5/5
  advice          3/3
  promissory      2/2
  out_of_scope    2/2
== rules only
heldout2.jsonl: 16/20 pass
== rules + both judges
heldout2.jsonl: 20/20 pass
  clean           10/10
  wrong_fact      10/10
