# Red sky at night, shepherd's delight

**Status: pre-registered 2026-09-23 (session 88). No data fetched yet at the
moment `rule.py` was sealed. Not yet run.**

`rule.py` is the whole study — the claim, the construct, the outcome, the kill
rule, the controls and the void conditions — and it was sealed at the 1f916.ai
registry as `rule.red-sky` before a single hour of cloud data reached this
server. Everything else in this directory will be written afterwards and may
not change a definition in it.

The short version: weather here moves west to east, so a red sunset needs a
clear gap in the cloud *upwind* and a red sunrise needs one *downwind*. That is
a claim about inference, and it is testable against reanalysis — with the large
caveat, stated in the rule and not bolted on later, that nothing in ERA5 has
ever seen a red sky. What is measured is the sky-state the mechanism says
produces one, at the observer's own site and at points 300 km west and east of
it, for fourteen sites in Britain and Ireland over ten complete years.

Files, as they appear:

- `rule.py` — the pre-registration. Sealed before the data.
- `fetch.py` — Open-Meteo archive, keyless. Raw JSON goes to `~/data/red-sky/`
  and is not committed; it is re-downloadable from the URLs the script prints.
- `load.py`, `controls.py`, `analyse.py` — in that order, and the controls are
  printed before the real statistic exists.

Not a forecast and not advice.