Ceremony audit: remove the process apparatus that made phase-bc 5x (checks, stack layers, TTL, reviewer sets)

[randlee]

Read-only audit of the repo's process apparatus (develop @ 139a1ed, 2026-09-26) under the just-say-no-to-process-porn-and-ceremony skill, ordered by Rand's hard line: no unnecessary checks; blocked on ceremony = remove it, never add a check. CI slimming already in #1604; ceremony-qa agent in #1603.

The 5x, measured

• Phase bc: 31 PRs for 4 feature layers. 13 were fix/bcN-review-M layers (bc2 reached review-7; docs(bc2): correct cumulative production receipt #1565 is +3/-3 lines with its own worktree, PR, stack link and a 17-check CI run), 7 were fix/bc-phase-end-* layers, plus a triage-closure PR and an evidence PR.
• The stack's stated reason ("CI runs once, merge once") is false: ci.yml has run on fix/, sprint/, feature/* bases since f8059e5 (2026-08-23). Every layer ran the full matrix. gh-stack-guidelines §13 ("stacked PRs receive no CI, open a CI-trigger PR") is stale.
• TTL machinery is maintained but unused: .sprints/BC/events.ttl has 0 Completion events (AL, AM also 0). .triage/ is 2,694 files / 16 MB / 218k lines, larger than the 193k lines of Rust; 969 of those files sit in 12 phases declared unparseable. No product code reads any of it.
• just lint pytests runs 90 files / ~20k lines of tests-of-lint-scripts on every run; release-preflight.yml:274 is literally test -f .github/scripts/release_gate.sh; fmt runs in four places.
• QA-1 runs seven reviewers; three of them (ruthless-boundary-qa, rust-best-practices, rust-service-hardening) are by charter structure-request generators and never converge. Each non-converging finding became a stack layer.

Removal list (highest wall-clock saving first)

1. Per-finding fix layers. Replace "new worktree + PR + gh-stack link + full CI + TTL record + closure edit per QA round" with "fix on the sprint branch until QA PASS, one PR per sprint; stack only when sprint B compiles against unmerged sprint A". Edit CLAUDE.md "Phase Integration Branch Strategy" and gh-stack-guidelines §0; delete §13 (stale), §14 (ten local gates re-run, already in pre-push + CI), §15 (trunk freeze + acks), §5/§7 ledger (no ledger file exists).
2. Reviewer sets. QA-1 = req-qa + rust-qa-agent (+ rust-code-reviewer); ruthless/best-practices/service-hardening advisory only, never blocking; drop plan QA after plan-hardening (a review of a review); phase-end review once with the five reviewers Rand named. Merge boundary-guard into ruthless-boundary-qa.
3. just lint process checks (11 of 38): remove pytests from the default lint order (run only when .just/, scripts/, .claude/ change, ubuntu only), ttl-triage, daemon-signing-coupling, site-generated (generate in pages.yml, stop committing output), function-length, legacy-mailbox-paths, capability-degradation, legacy-transport-removal (all freeze migrations that completed in PR Phase AM: Tokio/Axum transport-stack deletion (deletion-only) #853), merge hermes-adapter + nudge-taxonomy into one retired-identifier grep or drop; drop peer-dial-seam check 3 (verbatim design lines); strip the doc-wording assertions from arch-gates (boundary_enforcement.rs:330, 2406, 2501, 3117-3150) and the .sprints/.triage arms of the merge-marker test; remove the 6 "reserved for future phase" empty modules and their shear allowlist. modules is defined in run_lint.py but in no order list (dead entry).
4. TTL / triage apparatus: retire graph-orchestration (29 files, 3,726 lines), triaging-findings + qa-triage agent, closing-triage, scripts/triage_carry_forward.py, .sprints/, dead .triage/ phases now and the live ones once migrated: one GitHub issue per finding labelled phase-N; sprint-report merges into triage-report and both become gh issue list --label phase-N.
5. Dead or duplicate agents/skills: phase-orchestration + scrum-master (codex-orchestration is the live flow), the 5 sc-git-worktree-* agents + .claude/commands/sc-git-worktree.md (three copies of one capability), registry.yaml semver/depends_on (frozen at 0.1.0 since 04-22), .codex/AY5|AY6-task-list.md, .cursor/agents/publisher.md, the 976-line fuzz planner that does not execute, todo-triage (a grep elevated to a skill), 5 gemini workflows (dispatch: 30 runs, 25 skipped, 5 failed; review: 0 runs).
6. CLAUDE.md: drop the stale "Current Status" paragraph, the per-sprint "Agent-teams review" step 6 (2 postmortems in ~30 phases), and the duplicated Send-To rule (keep agent-conventions.md R8).
7. Kit-frozen, report only: release-preflight.yml:274 check-that-the-check-exists (sc-publish kit is frozen; file upstream if ever unfrozen).

Keep (product invariants or operator rulings): the 25 product-invariant lint checks, lines (RULE-003), site-links, adr-index, the small-fix rule, PR/QA-immediate + CI-merge-gate-only, merge-commit-only, worktree-only, codex-idle rule, task-assign-by-template, team-protocol, agent-conventions, gh-stack §18/§19/§21/§22, raw evidence under site/reports, ci-scope docs fast path, phase-aq-evidence.

Full audit tables are in the session transcript of 2026-09-26; each row names location, class, missing gate element (consumer / gate / defect / retirement), evidence and action.

[github-actions]

🤖 Hi @randlee, I've received your request, and I'm working on it now! You can track my progress in the logs for more details.

[github-actions]

🤖 I'm sorry @randlee, but I was unable to process your request. Please see the logs for more details.