RS-TKT-1 (Phase 1) · 18 — Codex-Style Self-Validation of Design

Lane: RS-TKT-1 — Phase 1 TKT Base Design Package (design-only) Date: 2026-06-22 Gate: REGISTRATION_HOLD · REGISTRATION_CAN_PROCEED = NO · 0 runtime mutations Authority: NON_AUTHORITY · may_gate=false · decision_effect=NONE

Adversarial self-validation of the design blueprint (01–08, 16, 17). Codex-style: I do not trust report prose; I located actual governed files; I reconstructed from the KB; I built bad inputs beyond the macro's list; I checked for fail-open.

---

1. The adversarial checklist (each: test question · evidence file · answer · stop state if unsafe)

#	Test question	Evidence file	Answer	Stop state if unsafe
1	Did I trust reports, or locate the actual governed files?	00 (read inventory)	Located actual files; read mandatory Codex report in full (rev 1, 9253 chars).	n/a — done
2	Did I fresh-reconstruct from KB, not local prose?	00	All sources read via agent-data KB; phase1-design/ confirmed empty pre-write.	n/a
3	Did I trust the harness?	00 §7	No — pre-write count verified count=0; readback planned (22).	n/a
4	Did I create bad inputs beyond the macro's list?	17 §2/§4	Yes — added BAD-L3-001 (cross-brick read), BAD-PROP-001/002 (HOLD/N/A), BAD-RS-001 (RS5B promote).	n/a
5	Does the design REJECT invalid input at every level?	02, 04, 06	Yes — every level has a FAIL and a HOLD branch; default = HOLD.	RS_TKT_1_REJECT_FAIL_CLOSED_UNRESOLVED
6	Can invalid input still emit PASS/digest/seal/cert-like output?	04, 17 §6	No — 6-conjunct predicate + structured & unstructured detection + scan-surface HOLD; every BAD-FC fails closed.	RS_TKT_1_REJECT_FAIL_CLOSED_UNRESOLVED
7	If invalid input can emit PASS-like output ⇒ fail-open ⇒ REJECT?	04 §9, 17 §6	Confirmed none can; if any could, the rule is REJECT.	RS_TKT_1_REJECT_FAIL_CLOSED_UNRESOLVED
8	Did I distinguish engineering PASS from authority PASS?	01 §5, 06 §8	Yes — authority_effect/registration_effect always NONE; aggregate advisory.	RS_TKT_1_REJECT_AUTHORITY_OVERCLAIM
9	Did I distinguish design-only PASS from implementation PASS?	01 §5, 15	Yes — no implementation; "design blueprint ≠ construction."	RS_TKT_1_REJECT_RUNTIME_DRIFT
10	Did I distinguish TKT Base PASS from semantic Text-as-Code PASS?	01 §4, 04 §3	Yes — L4/L5/L6 deferred; their tokens are forbidden output.	RS_TKT_1_REJECT_AUTHORITY_OVERCLAIM

2. Structural invariants re-verified (the macro §3.1 "do not report READY if…" list)

P1 fail-closed contract weakened?        NO — 04 inherits PATCH2 verbatim; §9 weakening test passes (floor not shrunk, no exit==0 escape).
L3 recombined into a mega-block?         NO — 05 keeps four independent bricks; 05 §6 anti-recombination clause.
Source authority tiers collapsed?        NO — 00 §6 keeps Tier-1/2/3; MCB-6 OPEN.
L1 can call subject-under-test runtime?  NO — 02 §3 inert-only; BAD-L1-001 ⇒ HOLD_RUNTIME_SURFACE_REQUIRED.
NVSZ root invented?                      NO — 07 §6 designates none; BAD-NVSZ-002 ⇒ ESCROW_E9/ROOT_E4.
RS5B draft rules promoted?               NO — 08 §5 no auto-promotion; BAD-RS-001.
HOLD can become PASS?                     NO — 06 §1/§8; BAD-PROP-001.
N/A can upgrade aggregate?               NO — 06 §3/§5; aggregate has no N/A value; BAD-PROP-002.
Construction blueprint contains executable implementation? NO — checked in 19.

3. Extra adversarial probes (beyond the Agent/macro tests)

A1 "Aggregate PASS smuggles authority" — refuted: 06 §8 + 13 §4; authority_effect is the constant NONE; report malformed otherwise.
A2 "Caveat-§5 token floor could be shrunk to fake-green" — refuted: 04 §3 monotonic config; removing a floor token is a forbidden config change.
A3 "Caveat-§6 removal of aggregate N/A could hide a real unassessable case" — refuted: 06 §3 maps 'nothing assessable' to HOLD (row 2), not a silent pass.
A4 "An RS5A rule firing on a non-RS5A packet looks like a finding" — refuted: 08 §4 it is a CONFIGURATION ERROR, not a finding; scope_class gates it.
A5 "Claim-audit could pass on prose if files unreadable" — refuted: 05 §2 unreadable cited files ⇒ HOLD_OUTPUT_SURFACE_UNAVAILABLE, never PASS.
A6 "Identity brick could quietly allow a new TKT registry" — refuted: 05 §3 L3_NEW_REGISTRY_PROPOSED is a trip; R-TKT-3.
A7 "0 mutations is asserted as live proof" — refuted: 22 states it is a package attestation (Codex §6.7), not live PG/Directus inspection.

4. Self-validation verdict (design)

No fail-open path found in the DESIGN. Every invalid input maps to FAIL or HOLD; authority/registration effects are constant NONE;
the four-brick split, the L1 inert boundary, the NVSZ no-invent rule, the no-auto-promote rule, the HOLD/N/A rules all hold.
This is an ENGINEERING/DESIGN self-check ONLY. It is NOT a Codex PASS, NOT an authority PASS, NOT an implementation PASS,
NOT a runtime PASS, NOT a production PASS, and does NOT clear REGISTRATION_HOLD. Independent Codex review still required.
RESULT: design self-validation PASS (engineering/design only).