# When the method was written, and how to check it

`METHOD.md` says the method was committed before anything was run, and that git
history is the proof. **On `main` that proof is not visible**, because the work
was squash merged: `METHOD.md` and `RESULTS.md` arrive in a single commit
(`f6aa128a`, PR #1237). Anyone checking the claim on `main` finds nothing
supporting it. The unsquashed commits still exist on the pull request ref:

```sh
git fetch origin 'refs/pull/1237/head:refs/remotes/pr1237'
git log --format='%ad %h %s' --date=iso refs/remotes/pr1237 -- docs/distribution/study/METHOD.md
git log --format='%ad %h %s' --date=iso refs/remotes/pr1237 -- docs/distribution/study/RESULTS.md
```

Measured 2026-09-19:

| Commit | When | What |
|---|---|---|
| `31bc2149` | 2026-09-18 10:40:05 -0700 | study: commit the method before running anything |
| `df7c3c2a` | 2026-09-18 10:45:00 -0700 | study: 97 out of 100 diaries with nothing in them got a pattern anyway |

**Five minutes apart, and say so.** The ordering is real, and ordering is what
the claim rests on. The gap is also small enough that a reader is entitled to
read it as one sitting rather than a commitment made well in advance. Claim the
ordering; never dress it up as more than that.

**Why this file exists.** Messages I sent on 19 September 2026 carry the
preregistration claim. A reader who checked `main` would have found the claim
unsupported and stopped reading there.

**Added 23 September 2026.** The recorded run started 93 seconds after the
method commit (`generatedAt` in `results.json`), and the study script was
committed with the results, not before. Read the method as a record made in
one sitting, not a preregistration. See `ERRATA.md`.