Head pending notes ## Unreleased and drop the dash from CHANGELOG.md release headers - #55

Independent review (advisory)

Verdict: APPROVE · 0 inline, 0 off the diff

Good

• The diff matches the PR card: 6 header lines change in CHANGELOG.md, and the note text, the ### groups and all other files are untouched (read in the diff).
• Every ## line in CHANGELOG.md is now ## Unreleased or ## AtmosphericModels vX.Y.Z YYYY-MM-DD; a grep of the ## lines on this branch shows no exceptions.
• v0.3.8–v0.3.12 now use the same header form as v0.3.0–v0.3.7, which were already in the convention, so the whole file is consistent.
• The in-repo bin/release still works on this file: its awk matches ^## AtmosphericModels and reads the version from $3, both unaffected by dropping the -, and it skips ## Unreleased (read bin/release:53-70).
• Only headers changed, so the version, manifests and REUSE headers are untouched and no tests or docs are affected.

Not good

• Until the bump PR lands, the in-repo bin/release on main passes over ## Unreleased and picks v0.3.12, which already matches Project.toml. A manual run would post the 0.3.12 notes again instead of failing. The old v0.3.13 (unreleased) header failed the version check and blocked this, so the safety net is gone until the bump replaces the script.
• CLAUDE.md says bin/release "extracts the top section of CHANGELOG.md". With ## Unreleased on top, that section is no longer what it extracts. The doc is slightly stale, though CLAUDE.md is reasonably left out of a headers-only PR.
• The card's second verification line quotes the box's error with a double space (CHANGELOG top is (expected v0.3.12)). That is the tool's real output, but it reads like a typo to someone who hasn't seen the tool.

_{claude, rubric CLEAN_CODE.md. A different lab from the implementer
on purpose: a reviewer sharing its blind spots would not flag its mistakes.}