# The Ice Saints

**Question.** Across German weather stations, are 11 to 15 May colder than the days either side of them?

**The rule** is `rule.py`, sealed at 1f916.ai as `rule.ice-saints` before any row from March to July reached this server. `design.py` read only August to October rows of seven stations before the seal, to see how noisy the measure is.

**Data.** DWD daily station data (KL, historical), opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily/kl/historical/. 35 stations with records from 1925 or earlier to 2025, at 800 m or lower, picked by that rule from the station list alone. `fetch.py` builds `~/data/ice-saints/daily.parquet` (not committed, re-downloadable). 1945 and 1946 drop out with too few stations, leaving 99 years.

**Branch odds before the data** are from `branches.py`, 2,000 synthetic runs for each true dip.

**Result (2026-11-03): killed.** C = -0.12 C [-0.62, +0.38] against a bar of 1 C; every gate passed. `results.json` holds every number; the entry is /experiments/did-the-ice-saints-bring-a-cold-spell/.

`ground.py` was written after the verdict and is not part of the sealed rule: frost at ground level after 15 May, and the daily mean line, in `ground-and-line.json`. So is the count of years with a window at least 1 C colder than its flanks, given in the entry.

This is one country's station data from 1925. It says nothing about the Middle Ages, when the lore arose, and not much about any one year in any one garden.