RS5B-CLOSEOUT 04 — Caveat Ledger & Non-Overclaim Boundaries — 2026-06-21

Purpose: carry the five PATCH2 Codex caveats forward exactly, with explicit status, so no successor summary silently over-claims. Each caveat: source · risk if ignored · required handling · blocks closeout? REGISTRATION_HOLD retained · 0 mutations.

---

1. Caveat ledger

C1 — Rollup revision typo

• Source: Codex RS5B-PATCH2 §2 & §9.1 — the PATCH2 macro rollup states "All revision 1," but rs5b-patch2/06-rs5b-patch2-decision-packet is revision 2 (verified in this closeout's source register, file 01 §4).
• Risk if ignored: future summaries quote a stale body sentence instead of AgentData metadata, propagating an inaccurate revision claim; readback integrity erodes; an auditor could mistake the metadata mismatch for a hidden overwrite.
• Required handling: future summaries MUST quote AgentData metadata, not the stale body sentence. PATCH2 directory revisions = 1/1/1/1/1/1/2/1 (index/01/02/03/04/05/06/review-packet); rollup = 1. (Carried into the supersession map as S4.)
• Blocks closeout? No. Non-blocking metadata caveat; the rev-2 decision packet carries the same accepted R1/R2 contract and gate posture, so there is no semantic conflict.

C2 — Oracle totality is classifier-scoped, not runtime totality

• Source: Codex RS5B-PATCH2 §6 & §9.3 — the package's "oracle is total: every input" wording is accepted only for the explicitly defined contract classifier and its ordered predicates.
• Risk if ignored: a later implementer treats "total" as proof that every arbitrary malformed serialization, missing runtime carrier, null value, parser failure, or implementation behavior is already covered — and ships a classifier that fails open on out-of-domain inputs.
• Required handling: treat totality as a property of the declared closed outcome set { AUTHORIZATION_BINDING_MISSING_EFFECT, APPROVAL_NOT_BOUND_TO_EFFECT_IDENTITY, ARTIFACT_HASH_MISMATCH, AUTHORIZATION_SCOPE_MISMATCH, EFFECT_IDENTITY_IMPURE, AUTHORITY_OVERCLAIM, BINDING_CHECK_PASS } over well-formed classifier inputs. Arbitrary malformed/implementation-level inputs require their own guards designed and proven in a future authorized lane. BINDING_CHECK_PASS stays necessary-not-sufficient.
• Blocks closeout? No (R2 partition + order are deterministic), but it forbids promoting the self-check to runtime-totality status.

C3 — Design fixtures are not executed tests

• Source: Codex RS5B-PATCH2 §6 & §9.2 — BI-E1..E7, OP-BI-1..4, and PATCH2-03 pseudocode are design fixtures / design notation, not executed tests; the pseudocode does not itself implement the BI-E7 PASS-level guard or the OP-BI operation-domain guards.
• Risk if ignored: the package is read as a passing test suite; a reviewer infers runtime correctness from unexecuted fixtures and skips real test authoring in P2.
• Required handling: label every fixture DEFINED_NOT_EXECUTED; any executable test is a future P2 deliverable, authored and run only in a separately-authorized lane (and only with RS-VALIDATOR if separately authorized).
• Blocks closeout? No (closeout is a contract/design consolidation, not a test run), but no fixture may be cited as evidence of executed/runtime behavior.

C4 — No live runtime verification (package attestation only)

• Source: Codex RS5B-PATCH2 §7 & §9.4 (and RS5A-PATCH4 §9.4) — "0 runtime mutations" / current-state claims are package attestations; the Codex reviews performed NO_CODEX_LIVE_READ / NO_RUNTIME_VERIFICATION, and this closeout likewise performed no live runtime read.
• Risk if ignored: an attestation is mistaken for independently verified runtime state; a stale "ownership=0 / no register_dot / vocabulary absent" snapshot is treated as a current guarantee.
• Required handling: treat current-state facts (ownership=0, 6 scopes, registry=9, 14 apr_action_types with register_dot absent, assign_governance_owner unimplemented, no canonical_operation vocabulary) as last-attested, not live-verified by this closeout. Any P2 entry gate must re-run a read-only preflight to re-establish current state at that time (see file 05 requirement R4).
• Blocks closeout? No (this package is KB-only by design), but no runtime fact may be asserted as live-verified here.

C5 — No P2 / Chairman execution / runtime authorization

• Source: Codex RS5B-PATCH2 §7, §9.5, §10 "DO NOT IMPLEMENT" — acceptance is narrow: contract wording may carry forward; it does not authorize P2, Chairman execution, canonical-operation vocabulary creation, owner/scope creation, implementation, registration, or activation, and resolves no carried registration prerequisite.
• Risk if ignored: Codex acceptance of PATCH2 (a contract-only ACCEPT) is misread as authorization to open P2 or to execute the founding act — the exact AUTHORITY_OVERCLAIM / P2_OPENED_EARLY failure (file 07 XBI-1, XBI-2).
• Required handling: P2 may begin only after a later independent review accepts the P2 entry-gate design (file 05) and an explicit, recorded, out-of-band Chairman authorization token (packet item 13) exists for the P2 design/build lane. Contract acceptance ≠ execution authorization.
• Blocks closeout? No — but it is the hard boundary of the whole package: closeout consolidates and scopes; it authorizes nothing.

---

2. Non-overclaim boundary statements

This package, and any summary of it, MUST NOT assert any of the following:

1. that RS5B is accepted as runtime authorization (only contract-layer acceptance exists);
2. that Codex acceptance of PATCH2 is P2 authorization;
3. that the BI / OP-BI oracle is total over arbitrary runtime inputs (classifier-scoped only);
4. that BINDING_CHECK_PASS is a registration / authority / activation PASS (necessary-not-sufficient);
5. that the canonical_operation vocabulary exists at runtime (it is REQUIRED_NOT_PRESENT);
6. that current-state facts are live-verified by this closeout (package attestation only);
7. that REGISTRATION_HOLD is cleared or REGISTRATION_CAN_PROCEED is anything but NO.

3. Caveat propagation check

All five PATCH2 caveats are carried with explicit status; none is dropped; none is silently converted into a runtime claim. Therefore the closeout does not stop at RS5B_CLOSEOUT_HOLD_PATCH_CAVEATS_NOT_PROPAGATED. All five are non-blocking for the closeout consolidation and become binding constraints on any future P2 lane (enforced by file 05 and tested by file 07).