Releases v0.23.1

[johnnyt]

Release prep for statifier_persistence 0.23.1 (sp-zllw). Bump and changelog promotion, plus the upgrading page's 0.23.1 subsection in a commit of its own ahead of the release commit; the tag follows the merge and is not made here, and the publish is the operator's.

What changes

Commit "Carries the 0.23.1 patch in the upgrading page":

• docs/upgrading.md: the "0.22 to 0.23" section's existing text moves, unchanged, under a new ### 0.23.0 heading, and a ### 0.23.1 subsection follows it saying a host changes nothing, the way "0.14 to 0.15" places 0.15.0 and 0.15.1. Additions only. The page's title, opening pin list and schema sentence are not touched.

Commit "Releases v0.23.1":

• mix.exs: @version 0.23.0 to 0.23.1.
• README.md: not edited. The install pin stays ~> 0.23.0, which already admits a patch; .claude/wurk/release.md says only a major or minor moves it.
• CHANGELOG.md: a new ## [0.23.1] 2026-09-29 section with a lead paragraph and an upgrade note, then the fragment's one Fixed bullet byte for byte; the header's changelog.d/ link moves from v0.23.0 to v0.23.1.
• changelog.d/sp-gbyk.md: deleted, the only fragment on main. changelog.d/README.md stays.

The release commit touches no file outside .claude/wurk/release.md's list. The upgrading edit is outside that list, so it is its own commit.

Why a patch

The fix is in StatifierPersistence.Testing.StorageConformance, which ships in lib/ and which a host runs, so it is host-visible and the patch is cut (ruled by the operator, 2026-09-29). Read in git diff v0.23.0..origin/main -- lib:

• lib/statifier_persistence/testing/storage_conformance.ex: the suite's using block derives @conformance_hash_suffix from the using module's name, and every hash a case retires carries it, including the private own_chart_a/0 for the tombstone-check case. Case names and what each case asserts are unchanged. Fragment sp-gbyk.md.
• lib/statifier_persistence/executions.ex: one comment above batch_one/3 rewords a record citation; no code changes.

Closed-set question: none. No refusal, reason or outcome union, event list, enum or public option gains a member in that diff; Executions.error/0, Storage.error/0, the adapter behaviour's callbacks and telemetry events/0 are unchanged. No schema migration; the statifier floor stays ~> 2.9.

Verification

• Full mix quality green on the release commit's tree, which carries the upgrading commit too: format, compile with warnings as errors, doc links, deps, docs, credo, 1,347 of 1,347 tests at 95.8% coverage, dialyzer; Postgres on 5432. The release commit was made on a staged tree byte-identical (git write-tree) to the tree the gate ran green on, with the repo gate lock and a machine slot held across both. The upgrading commit touches no Elixir code.
• mix hex.build builds statifier_persistence 0.23.1.
• The fragment's bullet is present byte for byte in the new section; the one removed CHANGELOG.md line is the header link moved to v0.23.1; no line of a published section is removed; changelog.d/ holds only its README.

Own review

Checked the lead paragraph and the upgrading subsection against main: the deadlock, its 40P01 answer and the cases it reached are the ones the sp-gbyk.md bullet states; the per-module hashes are @conformance_hash_suffix in storage_conformance.ex; the package's own adapters are unchanged because the only other lib/ change is a comment. The heading has no dash separator and carries the local date; no link-reference block is added.