M9: CI-matrix tidy-up (§14.11) — reconcile spec to the actual lanes

[guyo13]

M9 §14.11 — CI-matrix tidy-up (the final M9 build slice)

Docs-only reconciliation. The eight workflows already implement the intended per-PR / nightly / manual matrix — but §14.11 predated the M9 additions (fuzz F1–F4, Miri, !Sync/dir-lock, soak) and no longer described what actually runs. This rewrites §14.11 into a faithful index, with no src/ or workflow behavior change.

The honest outcome of a "tidy-up" slice: the workflows were already right; the spec had drifted. Rather than invent churn, this makes the doc match reality and states plainly what is not yet automated.

What changed

• Per-PR (blocking) row now enumerates the real ci.yml jobs instead of a stale summary:
  • rustfmt + clippy (-D warnings, all targets)
  • build + test = §14.1 codec/CRC vectors + §14.2 (reduced proptest) + §14.7 zero-alloc assertion + cargo bench --no-run + the §14.6 !Sync compile-fail (tests/ui.rs trybuild) & dir-lock / Send tests — all under cargo test
  • MSRV (1.85) cargo check --all-targets --locked
  • Miri (codec subset) — noting this is the whole Miri-executable surface (file-backed paths use foreign syscalls Miri can't run)
  • fuzz smoke (F1/F2/F3/F4) — BLOCKING (a reproducible crash is a real D1–D11 bug and reds the PR; distinct from the nightly time-boxed lane; never gates an H1 dispatch)
  • §14.4d dir-fsync presence (strace)
  • plus the paths-filtered per-PR gates: m8-macos.yml (H4 Half A), lazyfs.yml, m8.yml
• Nightly (scheduled) row names each workflow with its staggered cron — bench.yml 03:17 / fuzz.yml 04:17 / m8-dmflakey.yml 04:23 / soak.yml 05:17, all CONTINGENT + dispatch — and honestly flags that full-iteration §14.2/§14.3 and the §14.9 differential are not yet automated nightly (covered at PR granularity by the reduced proptest + the M6 model_oracle suite). No over-claim.
• FS-matrix honesty note (plan Slice 9): the FS matrix is meaningful only for the durability/metadata-fault gates (§14.4b LazyFS, §14.8 H3-physical, §14.4d); the byte-level codec/CRC/recovery-classifier logic the §14.5 fuzz + §14.6 Miri lanes exercise is filesystem-independent, so those lanes run once and are not multiplied across the matrix. Over-claiming an "all-FS fuzz matrix" would be dishonest.

No gate over-claimed

No long-run gate is marked green: the §14.13 N-CPU-hour fuzz row and the multi-hour soak stay OPEN-pending a dedicated runner; the Miri / !Sync / zero-alloc DoD rows remain DONE.

Verification

Docs-only (CLAUDE.md + docs/wal_design_v6.md). cargo fmt --check, cargo test, cargo clippy --all-targets -D warnings unaffected/green.

M9 status

With this, M9 is feature-complete — F1–F4 ✅, Miri ✅, !Sync/dir-lock ✅, loom removed (characterized) ✅, soak ✅, CI-matrix ✅. The only remaining items are the standing owner / dedicated-runner observations: the F1–F4 N-CPU-hour fuzz gate and the multi-hour soak gate (both infeasible in-session; wired as contingent scheduled/dispatch lanes with loud banners until first observed).

🤖 Generated with Claude Code

---

Generated by Claude Code