{
 "name": "back-cast, honest intervals for both sexes twenty years ahead, scored on 2023, which no run has read",
 "registered": "2026-10-05",
 "why": "The published shocks test (backcast-e0-shocks.json) met its bar for women in 2019 but men's intervals covered 65%, their centre too low. This is a NEW test on a year no run has scored; the 2019 verdicts stand as published.",
 "honesty": "The men's method was chosen in develop-shocks-men.ts and develop-shocks-men2.ts on earlier origins only (fit to 1979 and 1984, scored 1998 and 2003; nothing after 2003 read). There, every interval wide enough to cover about 80% lost the paired SIGN test against the narrow blocks arm (best: 118 of 310 countries better), because blocks wins calm countries by a little while honest intervals win crisis countries by a lot (mean interval score 17.34 against 22.18). The primary therefore compares mean interval scores with a paired signed-rank test, which weighs how much better each country is; the original sign test is reported beside it, and the development data predicts it will fail for men. Choosing this test after seeing that is disclosed here. Women use the published method unchanged, as a replication. Frozen before any 2023 value is read.",
 "data": "WPP 2024 life expectancy at birth by sex, the 155 countries of 1M+ people in 2000 (wpp.ts), years up to 2003 for fitting, 2023 for scoring",
 "methods": {
  "women": "the published shocks arm of backcast-e0-shocks.json (shocks.ts, k 2.5, episodes of 10, calm runs of 10), around women's crisis-robust centre (e0.ts robust: true)",
  "men": "centre: men as women minus a gap (gap.ts, hinge 75, trim, own), iterated along women's crisis-robust forecast; spread: women's shock paths (as above) about their own centre, the lower half widened by 3 and the upper half by 1.25: lo = centre + 3 x (women's 10th percentile - women's centre), hi = centre + 1.25 x (women's 90th percentile - women's centre)",
  "baseline": "each sex's own centre plus 5-year runs of pooled-gain residuals (the blocks arm of backcast-e0-shocks.json)"
 },
 "paths": "1,000 per country and sex, seed 20261004; horizon 20 (2003 to 2023)",
 "outcomes": {
  "primary": "for each sex: 80% intervals cover between 75% and 85% of countries in 2023, AND the mean interval score (width plus 10 times any miss) is lower than the baseline's, paired by country, one-sided Wilcoxon signed-rank test p < 0.05",
  "also": "the original paired sign test; misses below and above; coverage by level of life expectancy in 2003; the same for 2013 (ten years ahead)",
  "decision": "'honest intervals for both sexes' is supported only if the primary holds for both; reported whatever happens"
 }
}