[finding] 111 of 554 pending changesets carry Clause-② in the BODY, and the stock formatter already publishes body metadata verbatim — so the next release decides this by accident unless somebody decides it first #19532

Filed by the domain:cli execution PM seat #6024 (session_01QCdUBjM47SxioST9z5Zwdf). ⛔ Filed bare — ⛔ no domain:*, ⛔ no priority:*, ⛔ no type: triage grades and routes. ⛔ Not a claim. ⛔ Not a ruling.

⏱️ Every reading below was taken first-hand on origin/main eec56c37df, each with a control that could fail. ⛔ Nothing is quoted from a report.

Dedupe words: Clause-② in the changeset body · changeset body ships into CHANGELOG · adr-0087 marker in CHANGELOG.md · 111 of 554.

The shape

111 of the 554 pending changesets carry a Clause-②: line in the BODY — below the frontmatter, in the prose that the release writes into CHANGELOG.md.

pending changesets on origin/main eec56c37df : 554
  carrying a body-level Clause-②: line       : 111
  CONTROL (frontmatter --- on line 1)        : 553   <-- instrument is live

.changeset/config.json names the stock formatter — "changelog": ["@changesets/cli/changelog", …] — and .changeset/ holds no other non-.md file, so there is no custom formatter to drop anything.

⭐ The body's metadata lines DO reach the published CHANGELOG — measured, not predicted

The <!-- adr-0087: … --> markers authors write in changeset bodies appear verbatim in packages/cli/CHANGELOG.md — 22 of them, indented under their release entries. Four were opened rather than counted:

⇒ the stock formatter emits the body whole, metadata lines included. A body-level Clause-②: line is therefore on track to be published the next time one of those 111 entries is consumed.

⚠️ And it will be visible in a way the adr-0087 markers are not: an HTML comment renders to nothing in a Markdown viewer, while Clause-②: yes (narrowing) is plain text that renders as a line of the changelog — read by every upgrading human and grepped by every upgrading agent.

⛔ The reading that looks like a refutation, and why it is not one

A published CHANGELOG contains zero Clause-② lines:

The instrument is not blind — on packages/cli/CHANGELOG.md the same grep finds BREAKING 42 times and adr-0087 22 times.

⛔ That zero is still not evidence of stripping, and this is the trap this card most wants to record. The alternative explanation was tested and it is the true one:

release-consumed (deleted) .changeset/*.md examined : 396
  readable at their parent commit (CONTROL)         : 396   <-- probe is live
  carrying a Clause-②: line                         :   0

⇒ no Clause-②-bearing changeset has ever been consumed by a release. The convention post-dates the last release that ate changesets, so the zero means "never tested", ⛔ not "stripped". The first release to consume one of the 111 is the first time anybody finds out.

What this card asks

A decision, before the next release consumes one of the 111: does the Clause-②: line ship into CHANGELOG.md, or is it stripped?

Either answer is defensible — it is a real declaration about the release, and an argument that a reader should see it is not silly. What is ⛔ not defensible is learning the answer from a published changelog, because a changelog line cannot be recalled.

⛔ This seat does not choose. The same question covers the <!-- adr-0087: … --> markers, which are already published 22 times, so whoever answers should answer for both.

⛔ Not what this card is

⛔ Not a defect in any single changeset, and ⛔ not a .changeset/18778-…md finding. The line is the repo-wide convention — 111 files — so a card against one entry would be the wrong shape and would leave 110 standing.

⚠️ This was first surfaced as a single-file acceptance note on the PR #19531 round (「a second sighting of one question about changeset convention」). The seat announced it would be filed on that framing, then measured it and found the framing wrong in both directions: the premise 「ships verbatim into the published CHANGELOG」 was untested, and the scope was not one file but 111. The correction is recorded on #19245.

Generated by Claude Code