Construct two-site caller cleanup and certify fixed-word retiming floors

[RossRichardson]

Summary

Two existing unary replacements and deletion of the whole U-coordinate tail recovery improve the complete q2/q3 J callers from stock+8 to stock+4. The actual profiles are H2=36,H4=1 (mass76, stock72) and H2=64,H4=1 (mass132, stock128). Every original field retains its literal F_k J output gauge.

The construction attains the isolated whole-bank Clifford rank-distance bound r(U)+r(W)≥|r(L)−r(R)|: exact Gaussian Bruhat factors make U a physical monomial and W one native2f child. Quadratic Z phases and whole-field Gaussian units are retained. Offline L/R factorizations are not executed as uncharged native queries; monomial/ADD/scalar scans remain paid.

Four new exact crossed-path certificates also prove rank mass at least stock for arbitrary Clifford-frame retiming of the named repaired and two-site q2/q3 words. Every holonomy is full ambient rank, giving floors72/128. This maximizes the path-certificate objective; it does not prove source masses76/132 or80/136 optimal.

Scope and limits

The actual callers cover q2/q3 and every positive f, with every independent original field, both serialized inverse orders and all selected-native recipes. The general rank-distance lemma has its own isolated-interface premises. The retiming theorem fixes each unary/common-frame ADD graph and literal physical boundaries; gauges may mix tensor coordinates. Changed responses, helper couplings, scalar rewrites across native sites or endpoint contracts need a new certificate. No universal-J or arbitrary same-matrix caller bound follows.

The post-mass76 census covers all7 nonempty remaining-tail subsets and703 existing prefix pairs each. All4,921 cases fail the declared single-bank/two-sided-isolation premises. This same-engine reproduction has independent scope controls and is not a global two-site optimum.

Scientific base 20ecf2891d80b5acf18d980b2164fb44ea7faa90, priority6, remains pinned. Canonical main advanced to bb4bccbf42844d2801487450727c2f63e10d1d0f with only an unrelated sparse-baseline helper/test change; guidance, selected record and catalogue are unchanged. The explicit donor continuation changes the joint-query-controller card's scope; the broad original card is not solved. PR4074 at 5758f0f3042b740ce872b5b0ae974917bd07016d is the direct source; PR4069 retains fixed-core/native ancestry. Fresh bounded overlap and exact related heads are recorded; coverage is partial.

No favorable supplier, all-q caller, global optimum, physical owner/controller/precision/tape admission, compatible BIT, formal-kernel proof, new κ or selected-frontier change is claimed. TASK.json records the next questions and blockers. Reopening requires a changed graph/response/contract with a complete original-input and paid invoice.

Test plan

Exact reviewed head: 7b4594222a4b3b97404f9ef1d0b03b71ae2c6768.

• Registered Python3.12 normal/-O: python3.12 -B scripts/research_package.py --package research/two-site-caller-fusion.
• Metadata/pins, exact committed bytes and source history: python3.12 -B scripts/check_research_commit.py --package research/two-site-caller-fusion --check-source-history.
• Official export/integrity and read-only Git-free normal/-O replay with empty PATH; source bytes/modes/inodes unchanged.
• Complete actual admitted reconstruction; separate independent source transport/gluing proof, every used native recipe/embedding/Gaussian phase, full serialized inverses and all current counters.
• Every field of all four author and independent path-certificate fixtures, including edges, source chronology, signed holonomies, phases and literal inverses. No proposer search or cold witness synthesis rerun.
• Wrong pointer, f2-only phase, omitted monomial/inverse edge, reversed noncommuting order, wrong routing and missing BANK_SWAP controls reject; equivalent phase gauges and bank-permuted two-sided isolation pass.

README and REPLAY-MAP separate fresh checks, fixed witnesses and archival searches. Transport is0.99MB; logical aliases total24.47MB expanded or18.22MB by unique payload. Compression/aliasing does not reduce mathematical review scope. All original and executed bytes are authenticated, without network, executable data or bytecode-cache dependence. Separate Codex reviews are AI review, not human/formal acceptance.

The full natural Registered research packages workflow39269926756 is completed/success at exact head 7b4594222a4b3b97404f9ef1d0b03b71ae2c6768. Full workflow state/conclusion and source-head SHA were read separately from local checks. No manual check/rerun or merge was requested.

Sources and credit

Ross Richardson with substantial OpenAI Codex assistance in construction, proof, separate independent audit and packaging; Apache-2.0. Preserve icekylinx/PR531, Rohan Arun/kernel92 and the complete retained caller/compiler lineage; Chris R/Clear1nsight3987 and ToikY4002 phase/lift credit, and Clear1nsight's distinct3989/4045 interface/volume work. Original notices, complete source freezes and exact adapter/function maps are retained. Earlier3630/3632/3645/3649/3655 and separate BIT4021/4023 remain outside this contribution. Coordination stays in Discussion3606.

[RossRichardson]

PR4081, exact f303d156a886d4ee866c926ec8439fde8d1dd63d, supplies the separate arbitrary-support skew two-site tariff under the unchanged scalar/interface/paid-inventory premises, with neutral Pauli/odd-rank examples, and an independent whole-bank Gaussian cubic quadratic-response trace gate. It neither broadens this PR's four fixed-word certificates to arbitrary J nor proves76/132 optimal. Coupled helpers, changed graphs and three-power callers retain explicit open contracts. Canonical finding separates local/export checks from natural CI. Ross Richardson with substantial OpenAI Codex assistance. Sharp odd-kernel fixed-word floors66/120 now supplement the neutral examples at actual parents11/15 versus ambient12/16, with full independent graph/phase/inverse admission; selected-new-native/owner/volume gates remain open.

Current natural workflow39275778821 is completed/success at exact f303d156a886d4ee866c926ec8439fde8d1dd63d, read separately from full current-head local/export checks. Historical bbfc54c36abfd290310e9751a160f6f7c085d31c passed workflow39272321663; it is not the current-head CI receipt. No manual check/rerun or merge was requested.

[RossRichardson]

PR4082, exact ff277e8a958829954f371c85bdc447672266edbd, supplies a new-topology coupled cleanup floor at the exact saved final-independent-a entrance: R_f+S_f≥6f/8f with every binary full-bank ADD counted and arbitrary dirty banks restored. Complete q2/q3 local matrices/recipes/phases/inverses and conditional actual full-scan/equal-volume limiting moment are independently reviewed. This does not prove local6/8 optimality or a global caller gate, and retains earlier-cut/partial-scan/volume escape premises. Canonical finding separates exact local/export passes from natural CI. Ross Richardson with substantial OpenAI Codex assistance. Earlier cut58 appendix supplies the complete current-input matrix and a rank-neutral22f factorization after fixed54f; hypothetical16f remains OPEN, with no selected schema or native saving claim.

Current natural workflow39276431732 is completed/success at exact ff277e8a958829954f371c85bdc447672266edbd, read separately from full current-head local/export checks. Historical 3e6d682c278cdd0979b7f78d4fe18f304bde37da passed workflow39274489295; it is not the current-head CI receipt. No manual check/rerun or merge was requested.