# Friday the 13th on British roads

**Question.** Are reported injury collisions on British roads more common on Friday the 13th than on the Fridays a week either side of it in the same month?

**The rule** is `rule.py`, sealed before any British count reached this server. It was designed after the void American study (`/research/friday-13/`) and because of it: the error comes from ordinary Fridays as well as from a bootstrap, and the bar is +10 %.

**Data.** DfT road safety data (STATS19), the collision file from 1979, one row a collision. `fetch.py` builds `~/data/stats19/daily.csv` (not committed, re-downloadable).

**Branch odds before the data** (15 synthetic series each, 650 a day with weekday shape, day-to-day noise and a fixed day-of-month pattern of standard deviation 1.5 %, script's seeds 0 to 14): true 0: killed 14, survived 1. True +5 %: survived 10, inconclusive 2, killed 3. True +10 %: survived 15. True +15 %: survived 15. True -5 %: killed 15. V3 voids 2 or 3 of 15. At a 2.5 % day-of-month pattern V1 and V4 void about half of them: if real calendar structure is that strong, this ends void again, and that would be the answer.

This tests one claim on one country's reported collisions. It says nothing about admissions in one district in 1993, or about the American data.

**Result (2026-11-03): killed.** R = 0.9961 [0.9658, 1.0275] on 75 months; every gate passed. `results.json` holds every number; the entry is /experiments/is-friday-the-13th-worse-on-british-roads/. `fetch.py` read 9,116,617 collisions, with 8 rows without a usable date left out.