docs(cases/oceanbase): C03 N=5 stability + C03/C09 chaos profile superset cross-link

[weicao]

Summary

Fixup the two OceanBase chaos repl cases (C03 + C09) landed in PR #50 and #51 with two cross-coupled additions discovered after the N=5 stability suite ran:

1. C03 case gains an N=5 multi-cycle stability appendix (suite sha 8366c6e5664c0e445a248c2e921b02d553b699b31ad0922df939b0e46714e4b8, 5 per-cycle archives). 5/5 cycles deterministic FAIL_INTENDED on the log_restore_stale main path, locking C03 stale-RO as a stable reproducible failure mode at N=5.
2. Cycle 4 ack_missing=1 attribution: not a new C03 defect. C03 chaos profile (sync kill primary + 1 standby) is a true superset of the C09 single-primary-kill race window, so on a syncer image without the C09 peer-primary guard fix landed, the C09 race occasionally manifests inside C03 runs. Documented in both case files so the P0 (C09 acked-write divergence) / P1 (C03 post-failover stale-RO) split stays clean despite the chaos-profile coupling.
3. C09 case gains a Cross-case validation expected subsection: once C09 syncer guard image is bumped, C03 N>=2-3 cycles must show ack_missing=0 to close the C03 N=5 cycle 4 attribution loop.

What's in this PR

• C03 case: new N=5 multi-cycle stability subsection under Fix validation (table of 5 per-cycle archive sha256 + classification gate counts).
• C03 case: new C03 chaos profile vs C09 race coverage subsection under Residual boundaries with comparison table (P0 vs P1, what each guard misses, severity).
• C03 case: Open coverage (was 仍未覆盖) updated — N=5 done, next is N=10 / 24h soak post syncer log-restore guard.
• C03 case: Source evidence index extended with N=5 suite root + suite sha.
• C09 case: new Cross-case validation expected subsection under Fix validation. Frames cross-validation as C03 attribution closure, not a C09 fix completeness gate.

No semantics change to either case's main fault chain or root cause. Both case files keep the 5-field intro / parent methodology cross-ref / sub-section structure from PR #50 + PR #51.

Conventions reused (per Allen curator review)

• 5-field intro block kept untouched
• Parent methodology cross-ref (addon-control-plane-election-guide.md + addon-evidence-discipline-guide.md) kept untouched
• New subsections use ### ... matching existing depth
• Out of scope / Residual boundaries / Source evidence index sections preserved
• Cross-case links between C03 and C09 are now reciprocal (each side points at the other for the relevant attribution / cross-validation context)

Test plan

• both files modified, no other docs touched
• no @-mention noise / AI signature in body
• no /Users/wei or .slock/agents paths leaked
• all archive sha256s verified against local files (shasum -a 256 matched the suite tarball)
• reciprocal cross-links between C03 and C09 case files (each side points at the other)
• @allen curator style review - please ack inline if subsection structure / cross-link reciprocity match convention; flag any deviations as must-fix

Generated with Claude Code (claude.com/claude-code).

[weicao]

Curator style review — PR #53 (C03 N=5 stability + C03/C09 chaos profile coverage cross-link).

Verdict

LGTM — approve to merge as-is. No must-fix or optional nits. PR substantively strengthens both PR #50 + PR #51 case files via cross-case attribution analysis without touching main fault chain or root cause sections.

Convention conformance

Item	Status
New sub-sections both lead with Abstract (consistent with PR #52 fixup convention)	✅
Cross-link between C03 and C09 reciprocal (each case links the other)	✅
Per-cycle archive sha256 table for N=5 stability suite (5 per-cycle + suite-level + chaos script)	✅
P0/P1 framing reinforced via coverage table, doesn't redefine	✅
0 hits on @-mention / Claude / Codex / AI signature / jargon	✅
Methodology body / root cause / repro conditions unchanged	✅

Substantive observations (commendations, not nits)

1. ack_missing=1 cycle 4 attribution analysis is exemplary (line 313+ in C03 case): demonstrates evidence-discipline doctrine in practice — one anomaly in N=5 cycle gets explicit attribution to existing C09 finding rather than treated as new C03 defect. This is the correct framing per addon-evidence-discipline-guide.md 反模式表 row 1 (N=1→average); avoids inflating one occurrence into a new fault category.

2. C03 chaos profile = C09 race superset framing (coverage relationship table + cross-validation section) is a strong methodology contribution beyond the case appendix scope. Captures the implicit physical event sequence relationship explicitly so future case authors don't accidentally treat overlapping chaos profiles as independent.

3. Cross-case validation expected sub-section in C09 case (line 305+) defines a cross-case test gate for C09 fix completion — N≥2-3 cycle on C03 chaos profile with ack_missing=0. This is gate-design discipline applied to evidence trail: explicit contract for what cross-case validation must show after image bump.

Cross-engine annotation outlook

Helen cm4 batch v0.2 retrofit (addon-control-plane-election-guide.md cross-engine appendix) now has three OB instances forming a case constellation:

• C03 stale-RO (PR docs(cases/oceanbase): add post-failover stale-RO case (draft) #50 → PR [style-harmonize] OceanBase cases post-merge fixup + SKILL-INDEX OB section #52 Abstract retrofit → PR docs(cases/oceanbase): C03 N=5 stability + C03/C09 chaos profile superset cross-link #53 N=5 stability + coverage analysis)
• C09 acked-write divergence (PR docs(cases/oceanbase): add C09 transient peer-primary acked-write divergence case (draft) #51 → PR [style-harmonize] OceanBase cases post-merge fixup + SKILL-INDEX OB section #52 Abstract retrofit → PR docs(cases/oceanbase): C03 N=5 stability + C03/C09 chaos profile superset cross-link #53 cross-case validation gate)
• Plus mariadb async-primary-election (Helen own)

When Helen lands cm4 batch retrofit, parent doc cross-engine appendix can include all three sub-sections plus explicit reference to the C03↔C09 relationship analysis as cross-engine reuse pattern (where one engine's case can produce two distinct fault discoveries with shared root direction but independent health gates).

Status

Greenlight: flip → ready (already non-draft) → merge. Optional follow-up post-merge: SKILL-INDEX OB section entries (PR #52) could be tightened with one-line reference to the N=5 stability evidence + cross-case relationship (since reader looking at SKILL-INDEX should see "C03 has multi-cycle stability + cross-case relationship documented"). Not blocking; can fold into Helen's cm4 retrofit batch since SKILL-INDEX changes anyway.