== Reproduction (Python, independent of authors' R) ==
all cases 2254 verified 1678
unexp x logIT   r=-0.126 p=3.6e-05 rho=-0.133 N=1075
natural x logIT r=0.132 p=3.4e-06 rho=0.134 N=1235
age x logIT     r=0.042 p=0.16 rho=0.030 N=1157
joint OLS: b_unexp=-0.0306 p=0.027 | b_nat=0.0637 p=0.115 | N=1053
             median  size
death_unexp              
0.0            21.0   114
1.0            20.0    69
2.0            17.0   108
3.0            17.0    76
4.0            12.0   708
drop 0->4: 43%

== Robustness checks (not in authors' script) ==
IT range months: -3.0 1800.0 | negative IT n = 1 | IT>20y n = 37
excl. negative IT         r=-0.124 p=4.3e-05 rho=-0.132 N=1074
excl. IT > 20 years       r=-0.113 p=0.00026 rho=-0.126 N=1038
excl. same family (4)     r=-0.118 p=0.0018 rho=-0.127 N=695
same family only          r=-0.113 p=0.028 rho=-0.118 N=380
no family link (0) only   r=-0.170 p=0.0049 rho=-0.153 N=272
log constant +4             r=-0.126 p=3.6e-05 rho=-0.133 N=1075
log constant +13            r=-0.117 p=0.00012 rho=-0.133 N=1075
Mann-Whitney IT, rating 4 vs 0-3: U=109763 p=2.9e-05, medians 12 vs 18
bootstrap 95% CI for median difference (0-3 minus 4): [3.0, 11.0] months
median regression slope: -2.67 months per rating step, p=1.92e-08

== Coding/missingness checks ==
IT by whether unexpectedness was coded:
                median  size
unexp_missing              
False            14.0  1075
True             24.0   244
unexpectedness rating distribution (%):
 death_unexp
0.0    10.0
1.0     6.0
2.0    10.0
3.0     7.0
4.0    68.0
Name: proportion, dtype: float64
crosstab naturalness x unexpectedness:
 death_unexp    0.0  1.0  2.0  3.0  4.0
NATURAL_DEATH                         
0.0              9    8   35   58  728
1.0            119   68   89   30  134