The landing page says "open core" now, and this repo finally has a LICENSE

[bkd-dotcom]

The landing page told visitors Signetry was "Source-available" with a bare © 2026, and this repo had no LICENSE file at all — which LICENSING.md explicitly calls a bug.

What changed on the page

	before	after
§ III heading	Source-available. Installed from source.	Open core. Installed from source.
§ III lead	(no licence named)	names Apache-2.0, BUSL-1.1, and the 2030-08-31 conversion date
install tag	© 2026 · not on PyPI	BUSL-1.1 · not on PyPI
footer	© 2026 Binay Dalai	links to LICENSING.md; states contributors keep the same licence rights as everyone else

LICENSE is Apache-2.0. It is a static page — there is nothing here to restrict.

A styling bug caught before it shipped

.foot .rights had no a rule, so the new footer link would have rendered in default browser blue against a warm ink/wax palette. Added a rule matching .foot nav a rather than papering over it with an inline style.

README.md — the repo page was blank

Zero-dependency single-file site, so: how to preview it (python3 -m http.server), what belongs here (presentation only — governance behaviour goes upstream to core), and the two things this page is meant to survive that are easy to break by accident — no JavaScript, and a narrow screen.

Verification

Not just grepped — rendered. Headless Chrome at 1440×9000, then cropped and read back both changed regions:

• the footer, confirming "Open core" is underlined in the muted link colour and not blue
• § III, confirming the lead text and the BUSL-1.1 · not on PyPI tag

Plus an HTMLParser pass over the file: unbalanced: none, unclosed at EOF: none, and the new CSS rule confirmed inside <style>.

Not fixed here — reported instead

The page shows core@v0.6.0 in the hero ref line, the install command, and the registry table. Core is at v0.7.0 (the Homebrew formula pins v0.7.0). Out of scope for a relicense PR, but worth its own one-line change.

⚠️ Merge order

The new footer link points at LICENSING.md, which does not exist on main yet — it lands in Signetry/signetry#31. Merge that first.

[bkd-dotcom]

Pushed two more commits while auditing the page for this relicense.

The page was making four stale claims (a2cbdc1)

	page said	actually
core	v0.6.0	v0.7.0 — in the registry row, the hero ref line, and the install command
reviewer	v0.1.2	v0.2.0
plugins	—	v0.2.2

The install command is the one that cost something real: anyone who copied it got a release behind.

And plugins printed an em dash. Everywhere else in this project an em dash means not measured — here it was standing in for "we forgot". That substitution is precisely the one Signetry exists to prevent, so it should not have been sitting on our own front page.

action @v1 and eval v0.2.3 were already correct. action prints a moving major pin instead of its newest semver (v0.4.0) on purpose, because that's how the Marketplace consumes it.

Nothing was checking them (1e7f630)

tools/check_versions.py + a versions workflow. Stdlib only, so the job can't break on an unrelated dependency resolution.

The schedule is the actual check, not the push trigger. Drift here is almost never caused by a commit to this repo — it's caused by another repo cutting a release, which fires no event here at all. A push-only check would have sat green through all four of these. Push/PR runs only catch a bad hand-edit.

Two things worth flagging, because both are the failure mode this project cares about:

The first version of the checker was quietly broken. The row parser used href=...(.*?)<span class="rt"> with DOTALL. A Signetry/signetry link above the table paired with core's cell, shifting every repo/version pairing by one and dropping core's row entirely — so it reported three drifts while silently not checking the fourth. A checker that passes while checking nothing is worse than no checker. Rows are now matched within a single anchor.

Because that failure was silent, the row count is now asserted rather than assumed. If the markup changes so a row stops matching, the run errors instead of checking fewer claims. Both guards were confirmed against deliberately broken input before I trusted the green: a moved href trips the count guard, a stray link outside the table no longer shifts the pairings, and a row with no version cell gets reported.

One decision for you

I set reviewer to v0.2.0 to match the convention the rest of the table follows. But reviewer also carries a v1 moving tag and an action.yml, so it's consumable as uses: Signetry/reviewer@v1 — which would make @v1 the more useful thing to print, exactly as action does.

Both pass the checker (@vN is validated for existence, not currency). Say which you want and I'll switch it; I didn't want to quietly change what the page advertises as reviewer's integration path.

Also note this repo had no CI at all before this — versions is its first workflow.