Add general command failure parity matrix

docs/architecture/general-command-differential-testing.md

@@ -2,14 +2,18 @@
 
 ## 상태
 
-- 문서 상태: 구현 전 승인 가능한 설계
+- 문서 상태: 단계적 구현 중
 - 비교 기준: `ref/sam`의 `ng_compare` 브랜치
 - 대상: 휴식과 `cr_건국`을 포함한 일반 장수 예약 명령 55개
 - 범위: 명령 결과, RNG 소비, 로그, 예약 턴 lifecycle, DB 영속화
 
-이 문서는 테스트 구현 자체가 아니다. 아래의 파일, 실행기, 격리 스택과
-fixture가 구현되고 완료 기준을 통과하기 전까지 55개 명령의 동적 호환 상태를
-`확인`으로 올리지 않는다.
+현재 ref MariaDB와 core memory를 잇는 공통 runner, 성공 경로 55개,
+실행 중 확률 실패 9개가 구현됐다. 실패 9개는 내정 critical
+`주민선정/정착장려/상업투자/기술연구/물자조달`과 모략
+`화계/선동/파괴/탈취`이며 RNG 전체 trace, semantic state delta와 실패
+로그 본문을 비교한다. 나머지 제약 실패·값 경계·alternative와 전체
+core PostgreSQL 재조회가 완료 기준을 통과하기 전까지 55개 명령 전체의
+동적 호환 상태를 `확인`으로 올리지 않는다.
 
 ## 결정 요약
 

docs/architecture/turn-state-differential-testing.md

@@ -169,13 +169,18 @@ Compatibility is established per case only when:
 4. live sortie also passes the battle trace comparison;
 5. any ignored path is documented in the case evidence.
 
-As of 2026-07-26, 54 general matrix cases, 35 nation matrix cases, declaration
-and live sortie pass this boundary. Live sortie is the remaining general
-command key and covers battle entry, conquest, defeated-general neutralization
-and last-city nation collapse. Thus all 55 general command keys have a
-completed success-path comparison. This is 91 executable comparison cases;
-it is not yet a claim that failure/boundary paths or all 38 nation commands
-have been dynamically compared.
+As of 2026-07-26, 54 general matrix cases, 35 reserved nation matrix cases,
+declaration and live sortie pass this boundary. Live sortie is the remaining
+general command key and covers battle entry, conquest, defeated-general
+neutralization and last-city nation collapse. Thus all 55 general command keys
+have a completed success-path comparison. The three instant diplomatic nation
+responses pass a separate reference trace and core resolver/API boundary.
+
+Nine general in-action failure cases now also pass the common differential:
+five domestic critical failures and four sabotage failures. They compare the
+full command RNG trace, semantic state delta and the exact action-log body.
+This is not yet a claim that the remaining constraint, clamp, alternative and
+persistence boundaries have been dynamically compared.
 
 The fixture runner also reports whether the requested legacy command reached
 its completed execution path. For multi-turn commands this is derived from the