[drsamonline]

Adds the fourth of CLDR's reference links — cocogitto — as a CI-only gate.

What it does. cog check runs on pull_request over the commits the PR adds (base.sha..head.sha), against a cog.toml that declares the types this repo actually uses. licence: (12 commits) and sweep: are in the list alongside the standard set; without them the gate would have been red on history it was never meant to judge.

Advisory, not blocking. continue-on-error: true, matching how rustfmt and clippy are already gated here. Promote it once a few PRs have gone through clean.

Scope, stated plainly: it judges new commits only. The 82 commits already on main are not relitigated — 52 of them use a type outside the standard set, and rewriting history to satisfy a linter is not a change I would make to a shipped repo.

Dependency: cocogitto 7.0.0, pinned, downloaded from its GitHub release (musl x86_64, ~2 MB) and installed into the runner. It never touches the shipped binary: no crate, no runtime dependency, nothing added to Cargo.toml.

Evidence is the Conventional commits job on this PR itself — the first PR it has ever run on. Nothing else in the workflow changed.

[drsamonline]

Negative test, run locally with the same pinned binary and this repo's cog.toml — a gate that only ever passes proves nothing:

$ cog check $BASE..HEAD
Found 1 non compliant commits in 73ad6bf..HEAD:

Errored commit: ce25c76 <t>
	Commit message: 'Fix the thing'
	Error: Missing commit type separator `:`
exit=1

In the same range, feat(units): a conforming subject and licence: PolyForm Noncommercial 1.0.0 (docs/index.html) both pass — which is the point of declaring licence in cog.toml rather than leaving the standard set to reject the repo's own vocabulary.

And on the PR itself, the job in 6s:

cog 7.0.0
cog check "aea2963..7960e89"
No errored commits