[LittleCoinCoin]

Closes the docs and records items of the v1 conformance review: table C rows R, K and E5, and
table D row O9 (__reports__/candler-v1-review/00-findings_v0.md on
rdm/m2c-planner-neutrality). Every item was re-checked on f4ffe67 (v0.4.0, after M2a). M2a's
own drift in the same sections is included. The text is fixed, not the code. Where the code is
what departs from the design's intent, R01 keeps its rule and the gap goes to the code session.

What lands

• R01 (__design__/architecture.md). Before anything was edited, a design-intent reviewer
  checked each item against R01's history, the original design reports and the milestone
  rulings. A second pass then attacked the edited sentences.
  • The opening says what is built.
  • The import's manifest table gains treatment, snapshots, inputs and environments.
  • The raw-frames source is the digest of the gzipped legacy file.
  • The writable /opt/env volume (A13) and the candler.role label are now described, and so
    are control containers.
  • The Isolation signatures are the M2b trio, with the long-lived handle M5's.
  • The "plan" ban is scoped to the experiment.
  • The pattern rule now says what the schema test checks.
  • The parity plug-in is htn-creator-eval@0.5.2.
  • Nothing is touched in §Experiment file, §Run manifest or §Plug-ins, which M2c's amendment A2
    owns.
• README: what each crate holds at v0.4.0.
• CONTRIBUTING:
  • the schema line covers every root, and the core line excludes the evaluator helper;
  • the 0.y.z bump rule (features_always_increment_minor);
  • O9: a narrowing schema change is feat(schemas)!/fix(schemas)!, never a plain
    fix(schemas);
  • release-plz 0.4.3, as CI pins it, and uv in the tools;
  • a PR title that fits in 72 characters with GitHub's (#<number>);
  • the new length checks in Enforcement.
• Subject length (the writing-history setup, a blocking hook as Eliott chose):
  • the cog-managed commit-msg hook refuses a subject over 72 characters, counting characters,
    not bytes;
  • CI refuses a PR title that, with GitHub's (#<number>), would make the merge subject
    longer than 72; the limit is computed from the PR number. That suffix is how 049a9ab
    reached 73.
• Records:
  • a dated note on supervision_gates_report_v0 that the Linux half ran (runs 42568786540 and
    42576041947);
  • E5's "imports losslessly" now says what the oracle compares;
  • the wire-schemas campaign's 29 leaf gates are ticked on merged-history evidence, two marked
    superseded by GitHub releases;
  • its Progress rows are written by hand, as Amendment Log rows were. No dirtree-rdm command
    writes them, and every README validates.

After merge

• docs(readme) triggers a v0.4.2 core patch release (main is at v0.4.1).
• Run cog install-hook --all again from the main checkout: the installed hook is a copy.

Handed over, not fixed here

• Code:
  • no agent container record, so the job-end sweep refuses the agent container;
  • the legacy streams are missing from import.sources[] and its blobs;
  • the 37 local #[allow]/#[expect], for the Grit reviewer.
• M2c: the imported treatment has no commit, and imported snapshot digests are not the
  RFC 8785 form.
• Follow-up: M2b's and M2a's leaf gates are still ⬜ after merge (159 and 21).

Checks

• every subject ≤ 72 characters, cog check origin/main..HEAD clean, the trailer on all 11
  commits
• the title limit run by hand for merge(experiment): gate m2c, m4 and m5 on a core free of domain #7 and #100 (67 and 65); wc -m under C.UTF-8 counts
  a→b as 3 on macOS and Ubuntu 24.04
• the hook run by hand on subjects of 72, 73, 72 with a non-ASCII dash, fixup!, and no
  scope: exits 0, 1, 0, 0, 1
• actionlint .github/workflows/ci.yml
• no R01 hunk inside an M2c-owned section; no builds run

Co-Authored-By: Claude Opus 5.5 noreply@anthropic.com

🤖 Generated with Claude Code