[commdata2338]

Change and mathematical scope

Adds research/p104-literal-repair-rank/. The full error from applying the old correction to PR #3954's literal encoded P104 banks has rank 192d, where one array has complex dimension d. Its data kernel has dimension 16d. The 72 arbitrary dirty helpers remain unchanged, giving complete kernel dimension 88d. The predecessors established the first-bank rank 96d and the block commutator; this contribution quantifies the complete two-bank boundary.

For J=I-H, H^2=H, invertible T and HT=TH, the new scoped lemma is

rank(PA-AP)=2 rank(H)+rank(J(D-I)J).

The actual pivot block is D_JJ=I_8, giving 192d. Neither DT=TD nor commutation between address blocks is assumed. The unique postprocessor is R=APA^-1P^-1. If its entire increment factors through k array-sized linear summaries, R(s)=s+VUs, then k>=192. An explicit algebraic factorization attains 192 using the summaries r_N and w_N-B^t r_J.

This changes the consumer from a uniform scalar correction to arbitrary input-independent linear summaries. It is a scoped proof and source-bound finite audit.

What the 192 means. It is an information bound under the stated contract: the dimension of linear information an exact additive repair of this form must receive, even when forming and moving the summaries is free. It is not a native-call tariff, not a lower bound on simultaneous scratch storage, and not a physical implementation or supplier. It gives no new κ. Existing-helper restoration is established for the declared identity extension; new scratch formation and restoration are not supplied. The selected record (#5413, κ = 0.003027288235) is unchanged.

The eight-file package's PROOF.md contains the argument and complete obligation ledger; RESULT.json contains the deterministic 512-pattern receipt. No shared code or frozen predecessor is changed.

Task and overlap

• Task / scope: p104-side-correction, 72dfbf0708467ad34793/literal-two-bank-defect-linear-summary-repair; priority revision 7; scope refined. The full task card is TASK.json.
• Selection: read handoffs before retrieving the specific card with choose_current_research_task.py --task p104-side-correction; seed ec040933c293736d31a20d4f764ea62e, track complex-mechanisms. No reroll. This closes one bounded information obligation, not the complete adjacent-call handoff.
• Base: head 02d6f8da9e6717fe1115413b8684366a0beef140 on base main 32890bc5de034ab8fc93ac88ead7279a6a416d04. Canonical main was fetched again at 2026-10-25T05:44:01Z and checked again at 06:03Z, unchanged; relevant guidance, sources and the selected record are unchanged.
• Exact predecessors: PR #3954, f1e2839819959cd0534d008be91ed5b79ec55a6a; PR #4133, 88666f1dee93fbd85b31fd5d6f91a219e111b5f4. Their live proof bytes on main match those heads. SOURCES.json pins all four actual inputs to the canonical base by Git blob and SHA256. No pending executable is imported.
• Claim: discussion #1166 comment 19173713, following the literal-slot finding and the original adjacent-pair handoff.
• Method/domain: dimension-free kernel proof; all 512 physical phase patterns and all 280 input columns; exact sparse Gaussian composition; separate rational elimination over all 2,048 small cases; a noncommuting address-block fixture and controls. No parameter search.
• Milestone/budget: one scoped proof/audit within a 60-minute budget. Free summaries are granted only for the information lower bound. Physical acquisition, new scratch restoration, conversions, owners, movement, precision and the complete recursive bill remain unknown. No changed invoice is asserted, so there is no new signed-invoice pair to compare.
• Stop: narrow after this interface result. The inherited 44,096-excess scaffold remains rejected; do not convert 192 information dimensions into 192 paid native ranks or expand the construction on this evidence.

Nearest overlaps: PR #3954 and #4133 supply the credited identities and first-bank obstruction; PR #4370 treats helper completion and the failed gate; PR #4339 corrects fiber-tag orientation. The port-chart-pair and nested-refinement packages already cover different tempting follow-ups. This account's earlier P104 acquisition, matching, native-transport and precision PRs are not repeated.

Offline task and literal/commutator/repair-rank searches covered 7,404 complete local HEAD records. Final cache: 371 open PRs including drafts at 05:43:08Z and 4,779 Discussion records at 05:33:05Z on 25 October, both with complete pagination. The board and claim scopes were reread. Coverage does not include every arbitrary diff or unpublished result and is not a novelty certificate.

Validation

Environment: Linux, CPython 3.12.3, standard library. All checks pass at head 02d6f8da9e6717fe1115413b8684366a0beef140.

python3 -B scripts/research_package.py --package research/p104-literal-repair-rank --check-only
python3 -B scripts/research_package.py --package research/p104-literal-repair-rank
python3 -B scripts/check_research_commit.py --package research/p104-literal-repair-rank --check-source-history

The registered replay runs the main checker normally and under -O, followed by the independent dense implementation. It passes in about two minutes (121.77 seconds measured); a direct optimized main replay takes about 60 seconds and 21 MiB. The source-history and committed-byte check covers 12 package/input files. The official committed export and bundle integrity check passed; both direct scripts passed in a read-only Git-free export.

Evidence:

• 143,360 complete literal columns and 143,360 postprocessor columns, covering 512 patterns times 208 data plus 72 dirty inputs; every minor, kernel and endpoint checked exactly.
• 2,048 separate dense rational cases, a noncommuting address-block repair, a pivot change raising rank 8 to 9, and two dropped-hypothesis controls.
• Seven semantic rejection groups: omitted second-bank response, incomplete repair, wrong sign, wrong slot convention, leaked/erased helper dirt, and uncancelled summary dirt. All 192 summary-dirt basis directions produce nonzero output.
• Private export controls reject altered source bytes and replacing a determinant integer 1 by Boolean true in the receipt. All acceptance checks remain active under optimized Python.

Expected main marker:

PASS_P104_LITERAL_REPAIR_RANK rank=192 data_kernel=16 dirty_helpers=72 characters=512 columns=143360 summary_arrays=192 controls=7 no_new_kappa

then PASS_REGISTERED_RESEARCH_PACKAGE ... outcome=completed.

Not run or claimed: original full native-word replay, actual physical producer/consumer, new dirty-scratch protocol, compiler/ordinary/outer admission, complete supplier moment, selected cold replay, Lean, Docker, hosted CI, independent human review or formal-kernel proof. The all-dimensional claim rests on the written proof, with finite corroboration.

Credit

Chris R / Clear1nsight owns PR #3954's encoding, literal slots, pivot block and scaffold, following PR #3882/#3940/#3943. unpaidmoon owns PR #4133's literal commutator and scalar obstruction, and PR #4370's helper-gate record. Their first-bank result is credited, not presented as a mistake or rediscovery. Ryanson Jonathan (@pinterbanget) owns PR #4339's fiber-tag check. Retain Rohan Arun / PR #1972, Chafik Boukhalfa / PR #1991, Adam Gall / PR #1940, Josie Elliston / PR #1071 and the E8, native, Clifford and caller lineage in the pinned sources. Rank-nullity and linear factorization are standard methods; the new contribution is the complete boundary application and exact checking. New work is Apache-2.0.

Prepared for commdata2338 (@commdata2338), with substantial OpenAI Codex assistance.

AI assistance: submitted by @commdata2338 with substantial AI assistance. OpenAI Codex (GPT-6 Astra) did the derivation, the implementation, the checking and the writing; it was packaged and submitted with Claude Code. No independent human review is claimed.

Open next questions

PROOF.md equation (6) gives the exact summaries a physical caller would need. Name their actual owners, charts, chronology and dirty state, then price acquisition, every output update and restoration. The useful next decision is whether a changed complete caller can afford that operation. An actual prepared producer and a newly favorable full bill are blockers. More rank calculations or phase patterns do not resolve them.

Workflow feedback (optional)

The source-bound selected-baseline command correctly identified the selected record #5413 (κ = 0.003027288235), while the priorities page's 23 October selection note still names #4610. Updating that note during each promotion would prevent stale orientation; the frozen historical cards should keep their bytes.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Jur4aiU6BBSmBiaRjkbECf