stations kept: 108 / 110

COLD: n=22
  n=22 mean_r=-0.1716 median_r=-0.1662 #neg=18 #pos=4
  t-test vs 0: t_neg=-2.580 p_neg=0.0087 | t_pos=-2.580 p_pos=0.9913
  sign test: p(#neg)=0.0022 p(#pos)=0.9996
  trend mm/dec: OLS mean=-10.11 median=-1.63 (n_pos=11/22) | TS mean=-9.41 median=+3.10

MID: n=51
  n=51 mean_r=-0.4440 median_r=-0.4805 #neg=50 #pos=1
  t-test vs 0: t_neg=-22.323 p_neg=0.0000 | t_pos=-22.323 p_pos=1.0000
  sign test: p(#neg)=0.0000 p(#pos)=1.0000
  trend mm/dec: OLS mean=+1.74 median=+7.32 (n_pos=30/51) | TS mean=-0.20 median=+5.00

WARM: n=35
  n=35 mean_r=-0.3604 median_r=-0.4244 #neg=32 #pos=3
  t-test vs 0: t_neg=-9.520 p_neg=0.0000 | t_pos=-9.520 p_pos=1.0000
  sign test: p(#neg)=0.0000 p(#pos)=1.0000
  trend mm/dec: OLS mean=-9.52 median=-6.18 (n_pos=8/35) | TS mean=-5.98 median=+0.00

SUPPORT RULE CHECK:
  R1 COLD positive: mean_r=-0.1716, t_p=0.9913, binom_p=0.9996 -> FAIL
  R2 WARM negative: mean_r=-0.3604, t_p=0.0000, binom_p=0.0000 -> PASS
  VERDICT: NOT SUPPORTED
  NOTE: pre-registered rule is ambiguous for this outcome. The parenthetical
  '(or vice versa)' under 'Conditional' would also match; the standalone clause
  'Not supported = positive correlation absent in the COLD group' is applied as
  primary. Either label gives the same substantive conclusion: the COLD-group
  prediction fails and the WARM-group prediction holds.
