design blocks 42, completed 42, missing 0

=== reader gpt-6-astra: 42 blocks, 21 problems, infra-touched blocks 0, P1 failed (not correct or <1x) 5 | metric score
arm means (mean over problems of per-problem rep means): bare 1.800, show_all 1.626, bare2 2.830, attempt_then_consult 2.548, consult_on_failure 2.854, audit_then_consult 2.483, prior_then_consult 2.907
  PRIMARY attempt_then_consult - bare2             -0.282 [-0.670, +0.054]  problems up/down 5/14
  audit_then_consult - bare2                       -0.347 [-0.802, +0.011]  problems up/down 4/16
  consult_on_failure - bare2                       +0.024 [-0.075, +0.157]  problems up/down 2/2
  attempt_then_consult - prior_then_consult        -0.359 [-0.625, -0.140]  problems up/down 4/15
  audit_then_consult - attempt_then_consult        -0.065 [-0.312, +0.199]  problems up/down 9/11
  prior_then_consult - bare2                       +0.076 [-0.162, +0.320]  problems up/down 12/8
  show_all - bare                                  -0.174 [-0.482, +0.076]  problems up/down 9/12
  bare2 - bare                                     +1.031 [+0.557, +1.551]  problems up/down 17/1
  attempt_then_consult - bare                      +0.749 [+0.371, +1.201]  problems up/down 15/3
  audit_then_consult - bare                        +0.683 [+0.278, +1.152]  problems up/down 14/3
  consult_on_failure - bare                        +1.055 [+0.572, +1.592]  problems up/down 18/0
  prior_then_consult - bare                        +1.107 [+0.632, +1.630]  problems up/down 18/1
  attempt_then_consult - show_all                  +0.922 [+0.482, +1.423]  problems up/down 15/6
  consult_on_failure - prior_then_consult          -0.052 [-0.320, +0.206]  problems up/down 8/12
  audit_then_consult - prior_then_consult          -0.424 [-0.802, -0.111]  problems up/down 4/16
  diag P2_consult - P2_refine (raw second pass)    -0.244 [-0.670, +0.148]  n_problems 21
  diag P2_audit - P2_refine (raw second pass)      -0.337 [-0.810, +0.064]  n_problems 21
  diag P2_prior - P2_refine (raw second pass)      -0.063 [-0.375, +0.229]  n_problems 21
  diag S - P1 (raw pass evals)                     -0.191 [-0.501, +0.061]  n_problems 21
  selection (which pass each two-pass arm kept): bare2: {'P2_refine': 29, 'P1': 13}; attempt_then_consult: {'P1': 14, 'P2_consult': 28}; audit_then_consult: {'P2_audit': 26, 'P1': 16}; prior_then_consult: {'P1': 13, 'P2_prior': 29}