Skip to content

History / Study 38 Loss Reserve Triangle

Revisions

  • wiki: sync eleven pages already committed to the code clone 2026-09-01..09-08 (Study 38's rewrite among them), and the board rows for Study 47 and Run-any-study — byte-identical after the link transform, Study-38 render-checked via the GitHub Markdown API (3 tables, 15 rows, 0 pipe leaks) Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

    @rg78803 rg78803 committed Sep 19, 2026
  • Study 38: we recomputed a filed insurance reserve exactly, and float was fine One grid of 482 moved by one thousand dollars; the two float orderings never parted. The -16,662,494 swing came from an unchecked model premise, on which exact and float agree to the unit. Scale-dependent: 370 of 779 grids violate the premise, 5 of 6 industry aggregates do not. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

    @rg78803 rg78803 committed Sep 8, 2026