FINDING / HANDOFF: selected complex compact adapters with sequential repair

[jmbarrancoml]

FINDING / HANDOFF. I have released this claim after opening PR2250, head d404b48a1e7b147556b309497d55089c837dddb2, as jmbarrancoml.

I implemented sequential compact dirty controls and their exact exceptional-record repair: extraction, inverse destination keys, stable binary radix passes, reinsertion and cleanup. I also compiled the actual selected R480/T16 monomial catalogue into columnwise address operations and Gaussian unit phases, with inverse execution.

The written composition retains the existing algebra, sufficient guard and long-record domain. Given a completed ordinary interchange with cost O(V b^tau), it gives O(V((f log p)^tau+1)) adapter time, with control preparation and exceptional repair charged before absorption. Tests use a completed elementary interchange and record its calls; pending PR2234 can supply the faster written contract. I claim no new guard or selected exponent.

Evidence: all 3,474 distinct local monomials on all 512 columns, totaling 1,778,688 exact checks; complete small payload and Gaussian forward/inverse cases; and a complete 8,388,608-record repair with 6,291,456 exceptional records, 2,097,152 good records and 23 radix passes. Python 3.12 normal/optimized bounded replays and the full large replay passed. The full experiment also passed on Python 3.14. Its raw S input and expected T output use address oracles; extraction, sorting and reinsertion execute through the sequential tape routines. The default CI manifest keeps the bounded replays, with the large command documented separately.

The compiler distinguishes unary tensor phases from role scalars. Residual 0:1000.right sends two zero columns to (245,245) with phase zero. The role SCALE coefficient remains -1/2. I also follow the strict-upper quadratic convention despite redundant entries in raw recipe matrices. Both selected 24,767-event local programs match the frozen source catalogue.

The next concrete task is the complete complex caller: bind each lifted child to its actual owner packet and lifetime, preserve every DATA-PHASES translation, and prove row/work restoration and recursive numerical precision. A reviewer can start with PROOF sections 3–6 and the separate completed ordinary-callee contract. This package does not establish the complete complex supplier or multiplication theorem.

Canonical main remained 0fdd7d82de03e43c9e5a1000dfbcd380d1b687d2, priority revision 4. The final overlap refresh at 01:59:46 UTC covered 477 open PRs, 952 Discussions and 1,538 local records, with complete remote pagination and no warnings. I found no competing physical-adapter claim; the search excludes arbitrary code diffs. The original claim expired at 03:00 UTC and I finished within its scope.

I retain Douglas Colkitt's compact-control construction, the OpenAI framework, prior carry/guard audits, Ali Romman's source loader and the native Clifford/gauge44/T16 lineage in NOTICE.md and SOURCES.json. Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance; no independent human or proof-assistant verification claimed. I leave the selected frontier unchanged.

[jmbarrancoml]

DONE / RELEASED / 2026-10-16 21:00 UTC / priority revision 6.

I published PR3055, head 9dc841c9a101435fdd15e0f7623bf16b6d0363e9, as jmbarrancoml. I implement the complete Compact operation through encoded Count controls, fixed physical roles and actual B1 binary calls, including both source orders, constant toggles, exceptional repair and the highest selected bit. I preserve borrowed inputs and descriptors, restore dirty fields and concatenate arbitrary positive outer rows in order.

Under the pinned ordinary-callee arguments, I obtain the retained compact bound in the long-record domain. I charge the per-row B1 calls, encoded preparation, exceptional keys, copies and cleanup before absorption. I separate blank work on return from backing arrays retained after larger prior calls. See the proof and replay manifest.

I passed all eight registered cold commands on Python 3.12.15/macOS arm64: 151 encoded control states and 904 offsets; complete early/late/toggle rows with odd-prime padding; two complete 4,096-record two-digit cases; an 8,192-record mixed-guard case; a 1,024-column fitted binary handoff; and semantic/source/ownership rejection controls. I checked each input bit's destination, disabled the indexed reference oracle during physical execution, and verified owned-work erasure. I also checked all 36 package/input files against the publication commit. Hosted CI and independent review remain unverified.

I retain PR2250's raw identity, guard and repair method, PR2899's named C0 dependencies, ywz65's PR2534 execution fixture, and the earlier formal phase/repair results. The small complete Compact cases use B1's elementary branch; the separate handoff forces a declared reduced fitted body. I have not executed a selected GL64/O(73,2) payload, Gaussian monomial wrapper, complete recursive complex caller or formal refinement. I leave the selected frontier unchanged and do not transfer B1's coarse rate to PR2979's higher candidate claim.

I refreshed canonical main d8e0f83f00191df62174b641e9fff055c357f641 at 20:52 and checked new overlap at 20:53. I release this claim. A next contribution can bind encoded phase accumulation and signed-record traversal to the actual selected monomial recipe, preserving forward phase-before-affine and inverse affine-before-phase order, before joining the recursive caller.

Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance, including analysis subagents. I preserve predecessor attribution and immutable source notices.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Encoded fixed-role complete Compact operation with B1 binary calls, dirty-field repair and arbitrary outer rows","base":"d8e0f83f00191df62174b641e9fff055c357f641","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3055"}

[jmbarrancoml]

DONE / claim released / 2026-10-16 23:06 UTC / priority revision 6.

I published PR3087, head e7e49fb8176772c71ad7006920a407109a89e440, as jmbarrancoml. I completed the scoped physical Gaussian phase/monomial boundary from PR3055. DONE records my scoped result; I claim no independent verification or scientific admission.

I accumulate the repeated quadratic phase modulo four from encoded addresses, traverse each fixed-width signed coefficient field, and compose the affine word through one reusable Compact inventory. Forward execution phases the input address; inverse execution recovers that address before phasing. I enforce the phase-specific most-negative-word guard, preserve borrowed payload/descriptors, close owned scratch and permit a valid retry after the named signed rejection. At one column I avoid scanning unused G, while counting its borrowed storage.

I give the quantified endpoint and complete cost argument in PROOF.md. For fixed schema/recipe and f>1, I retain

S*sum_rows T_B1(v0,(f-1)G) + O((g+1)(V+M*A^6)+g*delta*M*A*(R+A)).

I charge component traversal, phase preparation, Compact repair and returned-port copies. I separate occupied contents, borrowed descriptors and backing retained from earlier calls. I preserve the predecessor universal B1/C0 contracts as explicit assumptions.

I passed all eight registered cold commands on Python 3.12.13/macOS arm64 at 23:04 UTC: 1,344 signed pairs, 580 encoded phase cases, complete generic and selected packets, a complete 2,048-record K=2/rho=1 rectangle, later-field rejection/reuse, a 4,097-digit unused G probe and eight failure controls. Arithmetic and failure controls also passed with assertions disabled. I verified 43 publication files against the exact commit. I have not claimed a hosted CI pass.

python3.12 scripts/research_package.py --package research/physical-gaussian-monomial

Remaining boundary. I leave complete recursive coefficient arithmetic, proper-width children, private lifetimes, root precision admission and formal machine refinement open. A concrete next consumer is stage 0, replica 0, pair 0, events 9394–9395: directed child 7 on L224, then addition from L0. PR2534 already supplies that interval and its earlier fixture. A physical successor must implement its exact halving/addition and signed/grid guarantees before claiming the interval. I leave the selected frontier unchanged.

Credit and freshness. I retain Douglas Colkitt's Compact/affine work through PR2250/2287, AAAmirinu's PR2378 signed/coordinate refinement, Gareth Davidson/bitplane's PR2422 record composition, ywz65's PR2534 selected-child fixture, and the physical ordinary/Compact chain in PR2899/3055. I preserve frozen notices and pin 20 live inputs. I fetched canonical main and checked five newly updated all-state PRs, complete 2238/2253/2417 and the latest 20 hub1166 comments at 23:00 UTC. Main remains d8e0f83f00191df62174b641e9fff055c357f641; I found no competing encoded phase implementation or source objection in that coverage.

Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance, including analysis subagents. Apache-2.0 with retained source notices.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Physical encoded Gaussian phase and monomial wrapper through one completed Compact inventory","base":"d8e0f83f00191df62174b641e9fff055c357f641","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3087"}

[jmbarrancoml]

DONE / claim released / 2026-10-16 23:38 UTC / priority revision 6.

I published PR3091, head d546815f4b116a79bd0ecec6661d2b2b3de29377, as jmbarrancoml. I completed the scoped coefficient boundary after PR3087. DONE records my checked result; independent scientific review remains open.

I implement fused C/inverse-C components with a four-state signed carry, exact halving and necessary/sufficient output-width checks. I traverse complete Gaussian record packets with encoded Counts and one fixed inventory, and implement aligned signed ADD/subtraction while preserving both borrowed operands. I also check nested caller handoffs with live parent Counts and copied protected returns.

I give the quantified endpoint, carry, cost, space and named-rejection arguments in PROOF.md. I charge O(V) per complete C axis and aligned ADD, and O(fV) for a direct f-axis leaf, including encoded setup, movement, rewinds and cleanup. I retain the direct-leaf stopping predicate. I do not replace a faster nonleaf child with this tariff.

At three bits I accept the exact endpoint C(3+3i,3+3i), despite the staged sum 6+6i. I avoid storing that complete numerator as an h-bit signed coefficient. I preserve the older staged program's stronger premises as sufficient; I do not claim that its audit was wrong.

I passed all eight registered cold commands on Python 3.12.13/macOS arm64 by 23:36 UTC: 37,640 C-row cases, 1,124 signed sums/differences, complete spaced/tensor packets, credited sharp-reserve impulses at one through eight axes, later-field rejection/reuse, nested parent handoffs and eight failure controls. Word/failure checks also passed under -O. I verified 39 publication files against the exact commit. I have not claimed a hosted CI pass.

python3.12 scripts/research_package.py --package research/physical-gaussian-leaf

Attribution. Alejandro Zarzuelo Urdiales's PR45 already proves Gaussian parity, the sharp ceil(f/2) C reserve and the component norm. I realize those credited bounds in the physical leaf and retain the separate f-bit normalized-Hadamard requirement. I also retain James Chang's PR1869 ripple adder, AAAmirinu's PR2327 staged prefix audit, ywz65's PR2534 selected-child fixture, and Nick Petalas's PR2772/3089 prefix/capacity work. PR2287 supplies the operator/grid convention; PR3055/3087 and their retained sources supply the physical tape framework. I preserve their notices and pin 20 live inputs.

Remaining work and freshness. I leave actual child/caller composition, general dyadic scalar preparation, normalization, private lifetimes and proper-width nonleaf recursion open. ADD requires aligned operands. I leave the selected frontier unchanged. At 23:35:48 UTC I refreshed canonical main and live overlap, including the sole later all-state update, PR3090 at 47a2bcec53667f490df6d40550e14a8ae465cba4. Its scalar-capacity result remains complementary and requests a physical bit caller. Main remains d8e0f83f00191df62174b641e9fff055c357f641; I found no competing fused physical C implementation or source objection in the checked coverage.

Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance, including analysis subagents. I claim no independent human or formal verification of this program.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Fused physical Gaussian C coefficient arithmetic, encoded complete-axis traversal and aligned signed ADD","base":"d8e0f83f00191df62174b641e9fff055c357f641","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3091"}

[jmbarrancoml]

DONE / claim released / 2026-10-17 00:24 UTC / priority revision 6.

I published PR3098, head ae54061980ef6a956d1276c7775aa833c1f669ca, as jmbarrancoml. I completed the coordinated sequential packet-access and local-lifetime boundary around the inherited scalar recipes. I retain Nick Petalas's PR3094 Boolean caller as a separate contribution.

I process each complete Gaussian component with fixed local target/source/scratch words, execute the literal ADD/SCALE recipe, check source and logical-zero scratch return, append the target result and erase local work. Source walks modify a charged private component copy. I preserve borrowed inputs and descriptors, including named rejection late in the packet and nested caller return.

In PROOF.md, I give the quantified unit/prefix and component-loop refinement. For V=2hN and I<=32 instructions per component, I obtain O((I+1)V) time and O(h+log N) local scratch, plus protected output and borrowed streams. I include loads, private copies, zero writes/checks, result replacement, counters, rewind and disposal. The implementation reaches nine live leases within 128 preallocated scratch tapes.

I passed all eight registered cold commands on Python 3.12.13/macOS arm64 by 00:20 UTC, using the updated main runner. Evidence includes two-provider source compilation with inherited hashes/counts; all 31 literal templates under three policies on 84-bit Gaussian packets; 4,440 small signed prefix cases; 612 HALF/DOUBLE cases; later-field rejection/reuse; nested handoffs and eight fault controls. Prefix/failure checks also passed under -O. I verified 36 publication files against the exact commit. I have not claimed a hosted CI pass.

python3.12 scripts/research_package.py --package research/physical-scalar-stream

The event-109/110 fixture composes the actual literal programs on aligned test packets. I do not reconstruct the original 632-input prelude values there. PR3090's seven-fractional-bit, P+20 capacity remains conditional on its correlated input/frame hypotheses. The 7,072 saving remains a logical real-instruction count; I have not proved a corresponding bit-time saving.

I leave actual family/frame alignment, routing, residual adapters, native/recursive child interiors, private centres and the complete caller invoice open. A next caller step can bind one named scalar entry and its physical owner descriptors, with gathering and return costs.

I retain Nick Petalas's PR3078 recipes and PR3090 capacity, Sagar Thakrar's PR696 lowering, James Chang's PR1869 ripple interface, PR3091's signed tape arithmetic and the full native/geometry/tape attribution in the unchanged archives. I preserve 15 transitive live source pins. I began at d8e0f83 and advanced to current main 18aacb215fc3717b06be8ceb9fca62aeca0d3e99 before publication; guidance, priority6, selection and pinned scientific inputs remained unchanged. The updated runner's path checks passed.

Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance, including analysis subagents. DONE records my scoped result; I claim no independent human or formal verification, supplier admission or new exponent.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Count-driven sequential scalar packet access, fixed word roles and paid source/scratch return around inherited recipes","base":"18aacb215fc3717b06be8ceb9fca62aeca0d3e99","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3098"}

[jmbarrancoml]

DONE / 2026-10-17 01:06 UTC / priority revision 6.

I published PR3106 at 1a9da866330d31d7b6edc5f514802a3ab2176f88 as jmbarrancoml. I close the selected call 0 raw1/raw2 common-chart and complete-bank access boundary. I preserve the selected frontier.

I derive the actual common chart from stage-zero data routes rho=I73, zero Y entrance translations and native frame 0 for owners 119/169. I bind the first scheduled cover sweep and its literal events. Using 704 distinct owners and bijective cover routes, I factor the two-event prefix before the cover tails:

F119 <- (-F119 + F169)/2, with the other families unchanged.

I preserve SCALE -1/2; ADD +1/2 and their intermediate guards. I implement complete-family acquisition from the 14,336-family bank, protected scalar returns, replacement-bank construction and caller-owned commit. I charge O(B+v) time and O(B+v+h+log B) additional transient storage. The caller must close the temporary bank before a recursive tail cut.

Evidence. I passed seven registered cold checks on Python 3.12.13/macOS arm64: selected source/chart binding with six source controls; a disjoint-route factorization model; encoded 15 Gamma extent; complete finite banks at N=1/3; later-component headroom/divisibility rejection, valid reuse and nested parent commit. I also checked source/error cases under -O. After adding the first-schedule binding I reran both affected source checks. GitHub CI was queued when I published this update.

Reproduce from the repository root:

python3.12 scripts/research_package.py --package research/selected-first-prefix

Limits and next consumer. I give the all-N argument in PROOF.md. The finite banks do not instantiate the full O(73,2) payload. I retain the selected route certificate and PR2287's mixed-radix row specification; I do not independently replay the full helper-chart census. The raw3 tail still needs its adapters, gathered packets, children and complete invoice. A next contributor can connect the committed bank to that tail and check closure before its first recursive cut. I claim no global speedup, exponent gain or full multiplication theorem.

I credit James Chang's PR2370 reference semantics/common-chart premise, Ali Romman's PR1969 cover/data-phase analysis, Gareth Davidson's PR1255 chart/phase discipline, ywz65's PR2591 prepared half-ADD, PR2287's row allocation and PR3098's scalar engine. The frozen PR3098 archive retains Nick Petalas's recipe/capacity work, Sagar Thakrar's dyadic lowering, James Chang's arithmetic and the earlier physical-tape/compact notices. SOURCES.json records the exact heads and attribution.

I refreshed canonical main at 01:02:22 UTC and checked live overlap through 01:03:03 UTC. Main remained 18aacb215fc3717b06be8ceb9fca62aeca0d3e99; I found no matching claim or source objection in the complete 2238/2612 threads, latest 20 hub comments, PR delta and named predecessor feedback. Earlier focused searches supplement that coverage. I do not claim coverage of unpublished work.

Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance, including analysis subagents. I claim no independent human or proof-assistant verification. Apache-2.0.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Selected call0 raw1/raw2 source-bound common chart, prefix factorization and complete-bank physical access/commit","base":"18aacb215fc3717b06be8ceb9fca62aeca0d3e99","expiry":null,"artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3106"}

[tekuuu]

DONE / claim released / 2026-10-17 05:15 UTC / complex track, OS seed 34.

Published PR3127, head 2d685e0dbdcc8c9a370bc77c2f36ccaa7636e37d, on base 997e6a9ca7b4f95d50e366f61412fd640d72a88a. I implement b^D, b=(1-i)/2, with D supplied as an encoded Count, exact low-bit rejection, finite signed carry, ten fixed scratch roles, preserved borrows and complete cleanup. For h>=2, N>=1, 0<=D<=h, the scoped argument gives O(V) time and O(h+log(N+1)) local scratch, with protected output paid separately.

The independent Gaussian-integer oracle passed 7,728 exhaustive small cases and 1,022 larger boundary/sample cases. Named rejection/reuse, nested parent handoff, four arithmetic mutations, source drift and typed-receipt controls passed. The same-input D=34 comparator costs 2,584,607 tape events for D successive degree-one calls versus 66,930 for one pass (38.62x); this is a controlled D-pass baseline, not PR2287's already fused high-level recipe or end-to-end multiplication speedup.

python3.12 scripts/research_package.py --package research/gaussian-normalization-stream

Both registered commands passed on Python 3.12.14/Linux, including the optimized read-only Git-free export. Publication-byte checks passed for all 37 package/input files. Hosted CI and independent scientific review remain pending. Main, the latest PR metadata through #3126, the open issue and this discussion were refreshed before submission; no overlapping normalization claim appeared.

Credit Jose Manuel's PR2287 identity/specification and PR3091/3087 physical framework, related PR3098 scalar work, Alejandro Zarzuelo Urdiales's Gaussian grid results, James Chang's ripple-addition lineage, and all retained source-chain notices. Prepared by Tekalign with substantial OpenAI Codex assistance; Apache-2.0.

Released next boundary: compose the S_D phase sweeps and normalizer around the fast recursive C supplier with caller grid and D<=h admission, return copying, restoration and final truncation. This implements the scalar normalization boundary only; it claims no new kappa, whole recursive supplier or formal multiplication theorem.

[jmbarrancoml]

DONE: whole-call physical complex controller

I published PR3385, head 7e5a187214126d31afc241a8f89f4a652352b58c, after merging current main 6f100d4bb5083f8c35d96fd95a539811c62f245c into the research branch. I completed the implementation scope in this claim: the full selected local/global dispatcher, all 349 scalar literals, complete owner-bank replacement, physical recursive continuations and normalized root return.

I passed all three registered cold commands, covering 16 component checks on Python 3.12.13/macOS arm64. The additional two-level fixture passed 110 child calls and returns over 49,152 payload bits. Metadata/pin, committed-byte and source-history preflights pass. I record exact receipts and the interrupted optional exponent sweep in EVIDENCE.md. Hosted CI also passes, including the package replay.

I supply the genuine selected ordinary factory at q*=213*2^80+1 and check its fixed rational admission across 2,395 residual pairs and 111 port/routing charts. I count the actual fixed tape inventory and charge retained input/replacement copies. In the proof I also show that the existing fast-root gate implies the Compact repair inequality, so this join needs no additional execution cutoff.

I retain PR2287's conditional O(V d^(599/600)) bound and its spaced-axis/long-record hypotheses. I have not executed the complete O73/GL64 payload or the production guard width, obtained independent refinement validation, or closed the outer multiplication transfer. I claim no new exponent or supplier admission.

Review handoff: examine the source-relative return induction and common-view directions, then the fixed-prime factory and complete cost argument. The reduced fixtures give concrete execution evidence; reviewers must assess the all-size refinement and retained native/hosting premises at their stated scope.

I retain the PR2287/2809/2899/3055/3087/3091/3098/3106 lineage, Tekalign's PR3127 normalizer, Nick Petalas's lowering/capacity work, Gareth Davidson's helper charts and the remaining source authors. Sagar Thakrar's PR3377 supplies complementary formal recipe dispatch; I do not claim the first source-bound dispatcher. Jose Manuel (jmbarrancoml), with substantial OpenAI Codex assistance including source-analysis agents. Apache-2.0.

{"schema":"integer-mult-coordination/v1","kind":"DONE","task":"uniform-transfer-audit","scope":"Complete source-relative physical recursive complex controller and normalized root, with fixed-prime admission, explicit inventory and reduced complete executions","base":"463113d675af7410ac91d53a4fed51ec6494eb9a","artifact":"https://github.com/CrocSwap/integer-mult-bounds/pull/3385"}

[thakrarsagar]

DONE / RELEASED — independent conditional parent-return refinement for the original PR #3385 physical complex controller. No active reservation remains under this comment.

Public artifact: PR #3464, exact head 56b856a8c6dbd44c4a15bde694c08bbaff6f3e2f, package research/complex-parent-return-refinement.

Result: under explicit complete-callee, primitive/bank-scratch, dense-role and immutable-binding premises, prepare_child, enter_child, return_node and the actual parking codec restore every non-target parent body and descriptor, consume the saved header, resume at the exact next PC, and install only the returned target packet. The proof covers arbitrary finite nesting and derives the child packet extent identity. The original selected inventory has 732 parked roles, hence 4B+733 encoded symbols per frame holding B body symbols; this is not a complete invoice.

Evidence: unchanged source methods and real Tape/Count; independent initial owner slices; rank-one/private and rank-two/bank cuts with 4,096/16,384-bit packets and nonzero spectators; every restored body/head and borrowed input; twelve nested codec frames covering every admitted body symbol; wrong-after, wrong-target and alternate-matching-PC source mutants; header, delimiter and source-corruption controls. Both registered CPython 3.12.15 commands pass, including isolated optimized warnings-as-errors (12.7 seconds total). Eighteen committed package files plus 36 declared inputs pass the committed-byte/source-history check; all eleven frozen files were separately compared to their exact original Git objects. Same-session proof and adversarial executable reviews found no blocker. Canonical main and fork main were synced and the branch rebased before submission.

Limits are explicit: unused origin markers are normalized to LEFT and trailing blank allocation is not logical data; saved-header checking does not authenticate arbitrary corruption. The theorem concerns successful admitted returns, not rollback. Child arithmetic, complete factory/root refinement, production guard, cost invoice, selected O73/GL64 execution and outer multiplication transfer remain separate obligations. The active T74 adaptation is not reviewed, and no selected frontier changes.

Attribution: Jose Manuel (jmbarrancoml) supplies PR #3385 and its credited source chain. All retained files are byte-identical and preserve notices. Sagar Thakrar supplies this independent conditional proof/replay with substantial OpenAI Codex assistance; same-session reviews are not independent human acceptance or formal kernel proof. Next useful original-controller reviews are exact owner/view direction, fixed-prime factory and root normalization under their retained hypotheses.

[thakrarsagar]

DONE / RELEASED — independent owner/common-view direction audit of the original PR3385 controller. No active reservation remains under this comment.

Artifact: PR3603, exact head dd2796bf53d0db44a1e1050c18f3f0bad7141267, package research/complex-owner-view-refinement.

Result: independently reconstructed all 101,376 owner occurrences in 144 original local calls, including 10,002 emitted helper charts. The source's row/column convention correctly gives physical owner d sigma^-1; the proof derives the entry/exit conjugations and all-cover packet partition under explicit retained endpoints. No scoped source discrepancy was found. A real call0 helper distinguishes the correct order (e0 -> 602) from the reversed order (e0 -> 518).

Evidence: all 144 unchanged group constructions checked with a declared exact-matrix cover adapter; every selector and affine envelope checked by independent column-vector arithmetic; complete compiled local middle words preserved; two separate literal ordinal round trips; six semantic source mutations, including both entrance and exit directions, plus counterfeit-source rejection. Both registered CPython 3.12.15 commands passed, including isolated optimized warnings-as-errors. Metadata and committed-byte/source-history checks passed for 15 package files and one declared live input; all eight frozen source files separately matched their original Git bytes. Canonical main and fork were synchronized, and the branch rebased before submission. Hosted CI is not claimed here.

Limits: original R480/T16 PR3385 only. This is a written conditional refinement with finite source binding, not a formal kernel certificate, full O73/GL64 physical execution, complete cover-encoder validation, native/local arithmetic proof, uniform invoice, changed-T74 review or outer multiplication theorem. No selected-frontier change. Fixed-prime factory/root refinement and source-bound transfer to changed callers remain separate questions.

Attribution: Jose Manuel's PR3385 and retained source chain; PR3106's first-prefix/packet-partition result; PR2569's binary owner/common routing. The original check already has a nonidentity terminal-translation test; this contribution adds noncommuting owner-composition and complete owner-ledger coverage, not a new routing identity. Sagar Thakrar, with substantial OpenAI Codex assistance and separate same-session review. Archival publication is not independent scientific acceptance.

[thakrarsagar]

DONE / RELEASED — independent original PR3385 root address-partition refinement. No active reservation remains under this comment.

Artifact: PR3607, exact head 8f656674e660ce67abd8ae7d23f0d1068ea6d410, package research/complex-root-address-refinement.

Result: the source's two reserved leaf calls and recursive core partition exactly all original spaced selected axes. The written conditional proof derives complete prefix/row/address serialization and explains padding removal from the completed rowwise homogeneous endpoint, allowing temporary internal row mixing. It retains leaf/recursive arithmetic, admissibility and physical ownership as premises. The original proof already gives the tensor split and padding-factor argument; this contribution supplies independent source binding and combined layout coverage.

Evidence: unchanged reserved_leaves, apply_leaf, accept, fast_core, real Tape/Count, row padding and transfers; declared permutation endpoint doubles; spacing two/residue one, nonzero spectators, two fields/two outer rows, sixteen rows padded to eighteen per prefix. Two fixtures route 40,960 distinct coordinate records and check all 14 recursive descriptors. Six negative controls reject source drift, wrong residue/width/prefix count, an omitted leaf and nonzero padded output. Adversarial review caught a reversed-residue test-oracle convention; corrected receipts include actual inherited Layout.bit_position checks for both leaves, core and full address, distinguishing the opposite residue.

Both registered CPython 3.12.15 commands pass, including isolated optimized warnings-as-errors. Metadata and committed-byte/source-history checks pass for seven package files and 47 declared live inputs. All eleven controller inputs separately match original PR3385 Git bytes. Canonical/fork main were synchronized and the branch rebased before publication. Hosted CI is not claimed here.

Limits: manual address-consistent plans, not production RootPlan.compute; permutation endpoints, not C arithmetic. No complete root run, phases/normalization/truncation proof, production O73/GL64 execution, uniform invoice, changed-T74 correctness or outer multiplication theorem. Selected frontier unchanged. Remaining original recursive/factory implications and source-bound changed-caller transfer are separate questions.

Credit Jose Manuel's PR3385 and retained source chain, Nick Petalas's PR3493 outer-format adapter, and Sagar Thakrar's PR3464 cohort setup/cleanup. Prepared by Sagar Thakrar with substantial OpenAI Codex assistance and separate same-session proof, implementation and adversarial review. No independent human acceptance or formal kernel proof claimed.