# RSP-1 source and scope review v0.1

14 October 2026 UTC. Development — no registry entry. Physical promotion: 0.

Reviewer: Codex `source_program_audit` parallel AI thread. This is a separate AI source/scope inspection, not human peer review, custodian promotion, or an independent physical result. It does not establish CRL-2b eligibility under the attached governance proposal.

Reviewed file: `/workspace/sites/unified-dynamics-public/rsp1.py`.
SHA-256: `8e9db2e432d9defb78dde206ac8c39714f291c0d9ecb8dc63279d0496e9f7ca8`.
The report binds only this exact version and its inspected `response_contract.py` readout; changes require renewed inspection.

## Result

No concrete transition or stopping bug was found in the inspected candidate. The C/H response, cached prior latent receipt, current intact-null mask, reservoir conservation, renewal-only step and full-state fixed-point HALT are consistent under the declared prepared-register assumptions. This is candidate semantic closure, not recovery of an original native source law.

An additional bounded replay checked 7,986 prepared states: all admitted binary T,L,q, stock coordinates 0–2 constrained to fixed integrity, branches C/H, and banks empty, one zero row, or a zero row followed by the current stock vector. Every C run's committed-update duration agreed with `close_duration`, and C exhausted available stock. Each one-step transition commuted with all six pair permutations. This synthetic AI reviewer replay does not score an independent episode and does not prove every future input claim.

## Source alignment

- C0593, `UD_CLOSE_VS_HOLD_GATE_HISTORY_NONIDENTIFIABILITY_THEOREM_v0.1.md`, defines C(q−,T+,L+)=T+L+ and H(q−,T+,L+)=T+(q− OR T+L+). Both satisfy omega+≤q+≤T+ and differ on retained-integrity latent extinction. They are lawful countermodels/candidate response maps; the source does not select H or C as native physics.
- The implementation recomputes I=T(1−L_new). H uses h=I*q_old, so q_new=omega_new+h exactly matches the C0593 hold map. It does not reuse h from a previous intact-null stratum.
- C0926 §§4.5–4.6 defines original HD source R_e=c1,e²(c0,i²+c0,j²), R_p=sum over the opposite pair, L_p=1[R_p>0], and T_p=product of hard edge-integrity bits. RSP's integer stock and r*(1−q_old) witness are separately stipulated candidate objects. The code's module description excludes a native UD claim; supporting presentation must maintain this distinction explicitly.
- C0635 §9 defines original edge integrity as 1[U_e†U_e=I]. RSP's fixed Boolean T is a candidate restriction; it does not derive break/repair dynamics or a physical transport predicate.
- E01's projection obstruction remains intact: a new complete candidate state may distinguish reduced inputs without proving that native Hodge inputs determine the same update.
- E03, `EXACT conditional mathematical criterion: visible Markov closure`, requires hidden state whenever visible projection is not lumpable. RSP retains stock, bank/cursor, cached L and branch. A Theta-only model is not generally closed: for branch C and T=111, L=q=000, r=000 versus r=100 have identical visible Theta but different next Theta. No fitted visible transition law follows.

## Required assumption clarity

1. **Cached receipt semantics.** Initial L is an admitted cached prior receipt, not a recomputation from current stock and q. Arbitrary initial admitted states are allowed. For example, T=000,L=100,q=r=000 yields a response-null inspection that clears cached L before HALT. This is valid under receipt-cache initialization but must not be presented as a simultaneous HD field measurement.
2. **Fixed frame and integrity.** The code works in one persistent labeled pair basis with identity event frame; `relabel` verifies covariance, not a sourced time-varying frame producer. T remains fixed. Neither choice closes the original native frame/integrity debts.
3. **Finite prepared resource.** Bank rows and cursor are complete candidate state. No external inflow occurs, consumed stock is removed, and renewal releases only the next already prepared bank row. The consumption law and queue precedence are candidate choices.
4. **Null is a response tag.** `RESPONSE_NULL` means no q/Theta toggles. It can coexist with cached-L updates, renewal-only stock changes and nonzero hidden-state changes. It is not a physical no-event statement or missing-source DEFER.
5. **Stopping checks full state.** HALT requires `nxt==s`, including cache, stock and cursor. A null response alone does not stop the episode. H can halt with latched-open q, available stranded stock or unconsumed bank entries. This is explicitly a model control outcome.
6. **Attempt versus commit.** `committed_updates` counts changed transitions and excludes the final fixed-point inspection. The terminal row records that final attempt. Duration statements should use the matching convention; update counts are not seconds.

## Remaining source debts

Original HD-to-stock carrier/readout bridge; independent preparation provenance and stock normalization; native response-branch selection; native integrity and frame producer; physical feedback/renewal mechanism; target-hidden positive and null episodes; source-authenticated custody; physical clock. C0593/C0926/E01/E03 do not close these debts. The finite prepared candidate can be replayed and compared while physical promotion remains 0.