10 — Supertrack J: cross-impact and non-regression review

Verdict: CROSS_IMPACT_CLEAN.

Invariant accounting

Check	Result	Basis
net authority surfaces = exactly 27	✅	RP-02→#05, RP-03→#20, RP-04→catalog root, RP-05→dependency_manifest/analyzer_contract_manifest, RP-07→FK on existing root, RP-08→#21. Zero new authority surfaces.
runtime-evidence remains non-authority	✅	11 physical tables; data are facts; object/constraint identities enumerated via typed #20 rows; values never define policy
new readiness gates = 0	✅	existing 14 gates extended in behavior only (OBJECT_AUTHORITY_IMMUTABLE constraint compare; SIGNOFF_AUTHENTIC / CAPABILITY_BEHAVIORAL / NO_BYPASS / DEPENDENCY_TRUTH consume runtime facts)
new hash contracts = 0	✅	7 contracts (H01..H07) unchanged in identity/count; bindings of H01/H02/H04/H05 updated to the now-defined columns (this is a binding completion, not a new contract)
11 runtime tables do not alter authority count	✅	category-(c) non-authority; explicitly outside the 27
retention authority adds no surface	✅	option i, fields on #05
expected-constraint catalog adds no surface	✅	typed rows inside #20
no prior CP-01..CP-09 regressed	✅	see below
no Stage 2.6B opened	✅	design only; all live actions operator-gated/blocked
no permit / apply / REAL_RUN enabled	✅	unchanged; explicitly blocked

Non-regression vs the "do not change" list (T1 handoff doc 12)

• 27 byte-level child contracts — untouched (CP-01 doc 02 rev 10 intact).
• no-policy-CHECK / no-DEFAULT-false — preserved (runtime tables hold facts; pass/blocked are guard-derived, not policy CHECKs).
• typed-operand num_nonnulls=1 + compatibility — preserved; mirrored in capability_measurement (num_nonnulls(...)=1).
• sealed catalog root + one-active index — preserved; RP-04 adds families inside it.
• CP-05 every-threshold-sealed / no-threshold-table — preserved.
• CP-06 canonical hash rules (hex encode, trim_scale, UTC, COLLATE "C", total order, JSON-null, no MD5/delimiter, PG-major gate) — preserved verbatim in doc 07; the additions are the Runtime-Evidence Column Bindings section (which is where P-01/P-02/P-03 live).
• CP-07 path A + real-query preflight — preserved; RP-08 layers on it.
• byte-defined evidence/principal/identity/analyzer registries + ALTER-broken FK cycles — preserved (CP-08 doc 09, CP-09 doc 10 intact).

The one cross-impact caveat

The cross-impact table (Codex doc 08) lists "H01/H02/H04/H05 bindings updated" and "H01/H02/H04/H05 recompute" as a required check. The accounting is clean (no new surface/gate/contract), but the binding update itself is incomplete at the precision level — see Supertrack B (G1–G3) and the proposal package. This does not change any count; it is a completeness defect inside an existing contract, not a cross-impact regression.

Conclusion

Cross-impact and non-regression are clean. The "exactly 27 / no 28th authority surface" invariant holds; nothing from CP-01..CP-09 regressed; no live posture changed.