{
  "name": "back-cast, life expectancy, crisis-robust: the same question as backcast-e0.json with two changes",
  "registered": "2026-10-03",
  "honesty": "Designed AFTER results-backcast-e0.json was seen (its errors were HIV/AIDS recoveries and wars). The changes were chosen on an earlier origin only (fit up to 1984, scored on 2003, by develop-e0.ts: MAE 3.38 -> 3.20 women, 3.43 -> 3.13 men) and frozen before this run. Its 2019 score is therefore not a clean held-out test of the idea; it is reported as what the frozen version does.",
  "data": "as backcast-e0.json",
  "method": "e0.ts with robust: true. (1) The own effect is the median of the country's last 15 yearly residuals, leaving out residuals beyond 3 median absolute deviations of all countries' residuals, shrunk by 15/30 and halving every 10 years. (2) A deficit below the country's own peak of the previous 20 years is regained at a yearly rate learned from every country's history up to the cut-off: the median yearly share regained over the ten years after a deficit of at least 2 years.",
  "outcomes": {
    "primary": "mean absolute error of e0 in 2019 and 2009, per sex",
    "decision": "target 2 needs at most 1.0 year for both sexes in 2019, and intervals (not produced by this version); so this run can only show how far the gap closes"
  },
  "seed": 20261003
}