05 - SUPERTRACK E: Dependency-Safe Construction Order Review (S00-S19)

Checks

check	result
creation order matches FK/deferred-ALTER dependencies	PASS - mirrors approved RP-03 creation order; 4 forward child FKs + cycle FKs deferred to S08
RP-03 expected-constraint set authored after final object set	PASS - #20 rows seeded S09 after S08 deferred constraints; both-EXCEPT at S11
runtime-evidence tables available before hash contracts that reference them	PASS-after-fix - made explicit (XHO-01): #10 source_relation regclass requires S04/S07 relations to exist before S10 seeding
code_catalog/root exists before dependent manifests	PASS - S02 catalog root before S03 anchors before S04 children
H04/H05/H02 dependencies ordered correctly	PASS - runtime-evidence (S07) + signoff_binding before H02/H04/H05 hash rows (S10); G-H04-SCOPE/G-HASHDET gate S10/S11
control_epoch appears before activation-sensitive evidence	PASS - control_epoch columns created with manifest_set/activation (S03/S06) and runtime-evidence (S07), all before activation S14; G-EPOCH-TOCTOU
owner/ACL cutover only after read preflight + rollback snapshot	PASS - S16 after S15; G-DIRECTUS-READ + verified snapshot precondition
no Stage 2.6B / apply step accidentally included	PASS - S00-S19 contain no permit/REAL_RUN/QT001-apply/2.6B step
no step both T1-authorable and production-mutating	PASS - who column cleanly splits T1 (author/read-only) vs OP (LIVE/SEAL/ACTIVATE/REVOKE/FREEZE)

Findings

XHO-01 (P2) - implicit regclass ordering + wrong S19 precondition

• (a) hash_component_manifest #10 has source_relation regclass NOT NULL; a regclass literal requires the referenced relation to exist at insert time. The order S04/S07 < S10 already satisfied this, but the dependency was not stated - an implementer could reorder #10 seeding and fail. Same for storage_class/adapter regprocedure columns.
• (b) doc 04 S19 ("post-cutover verification") listed precondition S16, but S17/S18 (freeze/ deprecate) run between S16 and final verification, and S19 must run the full suite including G-LEGACY-FROZEN (which needs S17). doc 07 PKG-I correctly preconditions on PKG-H (=S17/S18); the doc 04 cell was inconsistent.
• Fix: doc 04 dependency notes now state the regclass/regprocedure binding-order constraint explicitly; S19 precondition corrected to S18; S19 proof now cites the full 30-guard suite incl. G-DOT-NOOVERWRITE/G-LEGACY-FROZEN.

Cross-impact checked

• The S19 precondition change aligns doc 04 with doc 07 PKG-I (PKG-H predecessor) - no remaining order inconsistency.
• The regclass note does not change the step sequence; it documents a constraint already honored.

Verdict

CONSTRUCTION_ORDER_PASS_AFTER_FIX - order is dependency-safe, cycles deferred, no apply/2.6B step, no T1/production blending. Two ordering clarifications applied.