[aliromman]

This construction package gives exact storage for all 512 centre-query rows in the historical 14,336-owner schedule, a general selected-minor unit-completion theorem, one native masked-query unit, and cleanup at a specified aligned cut. It does not improve the complete kappa calculation. Read the proof, independent publication review, and source pins.

Construction	Exact coverage and local cost
Relative dirty transport	512 rows in 64 windows through distinct old productive owners; original donor progression preserved; 274,432 real scalar primitives; no added child or field.
Clean idle storage	512 rows, 1,024 paid swaps, and 384 bridge read changes; 26,624 real scalar primitives; no added child or field.
Masked unit	All 712 local input columns and 480 helpers; 213,466 real scalar primitives and eight rank-one children beyond clean-J; no added field.
Owner-local cleanup	All 457 latent columns; 25,124 real scalar primitives, 240 maps, one owned C field, and no added child.

Dirty transport proves a relative identity; a later query read still needs to cancel the old donor value. Clean storage keeps the query's original physical frame. Any use of a parked query must restore it before swapback. The original compatible-K theorem requires an invertible selected minor and the stated selected block of K. Its mask requires a supported right inverse; the proof includes the required coordinate projection. Only D=R_U has an emitted native word. Cleanup assumes the exact aligned post-F2 cut and distinct real-family roles. The algebraic inverse does not authorize operations between unequal native frames.

The general-completion annex removes the selected-K compatibility premise. For an onto query, a unit selected minor is necessary and sufficient for a unit completion that retains every other output and stores the full query in the selected outputs. If RK=gI and a unit d-field minor is available, the minimum number of changed complete output fields among all linear unit repairs is zero when 1-g is a unit and d otherwise. The actual Gaussian-dyadic matrix gives eight by reduction modulo 13. This is a field count, with no native-child lower bound or complete kappa improvement.

The source pins, credit, additive manifest, and separate review preserve the original package and state the nonlinear residue premise. The relocated replay matches the complete frozen receipt: 500 F5 cases, including 400 incompatible selected-K cases, both inverse orders, and two converse controls. Run the portable annex replay with:

python3 -I -O -B research/borrowed-centre-storage/annex/general-completion/replay.py --out /tmp/borrowed-centre-general-completion

Use a new output directory. The expected final line is:

PASS_BORROWED_CENTRE_GENERAL_COMPLETION_ANNEX

Native kernels, recursive storage, finite precision, and a complete supplier invoice were not rerun.

The publication base is b58ad9ef4d4e861b0eaa5dba41742bdc072dca51. It selects the changed conditional PR #4470 construction; admission of these historical local words to that construction is not established. The historical schedule archive is consumed at 0a31ad88419a4eebf515d3df4f6732c4e38c0029; the alias source is pinned at bce54fb3f895cbe5cd4c0eb477c3afeed100190f. PR #1862 supplied the input-only source contract. PR #1969 supplied the joint-query matrix and routes. These are consumed predecessors. Their results are not new claims here. The notice preserves the gauge44/t16 contributor credit and licenses and identifies the campaign construction and review sources. The retained source credits @icekylinx for the gauge and pipeline, @FanPUYen for t16, @jeremykothe2 for the retained local word, @chafreaky for the frame and owner constructions, and @rohanarun for the native recurrence correction. Substantial OpenAI Codex assistance was used; no independent human review is claimed.

Use Python 3.10 or later, the standard library, one process, and at least 3 GiB of free space. From the repository root:

python3 -I -B research/borrowed-centre-storage/replay.py --out /tmp/borrowed-centre-storage-check

The output directory must not exist. The expected final line is:

PASS_BORROWED_CENTRE_STORAGE_PACKAGE

The fresh relocated replay passed all nine checks. It reconstructs both complete storage schedules and both local words. Controls reject wrong phases, duplicate owners, a missing swap sign, stale bridge reads, wrong mask order, omitted cleanup, and private C that survives an owner boundary. Evidence links the original proofs and receipts; adapter changes are explicit.

Native child kernels were not rerun. Address, precision, control, cover, and recursive storage costs need separate bills. A changed global masked schedule, target-code retirement, B3 admission, and complete supplier admission remain open. No global storage lower bound or selected-frontier change is claimed.

The full-W component annex adds a borrowed-record construction with 128 old donor families, at most 48 live at once, and no new owned record family. Its proof establishes equality with the same clean-owned-record reference on all 14,336 final old fields, for arbitrary data and donor values. Only the 3,584 protected first-pair target transforms are identified. The other 10,752 target maps remain open. Donors reach their required old-schedule native outputs; this is not literal restoration to their input arrays or independence from their original productive inputs.

The added profile is 3784 H1 + 7 H2, with rank mass 3,798, 7,582 native wrappers, 8,192 cover maps, and a scalar component allowance of 1,264,736. The allowance includes the full replacement bridge bill; it replaces that old bill rather than adding another copy. Retained source calls, old helper and idle jobs, the final cover correction, recursive children and hosting, finite stream and address costs, and precision remain outside this subtotal. There is no complete supplier or kappa improvement. Whole-pair parking, complete helper endpoints, and the ordered gather, return, and three donor adapters are required.

The source pins retain all 66 original files and the exact historical archive. The fresh replay receipt records four serial independent component checks under -I -O -B: composition, actual donor allocation, delayed cleanup, and all-output equivalence. Every complete receipt matches its frozen bytes. The only portable adaptation changes one absolute repository root. The old raw native kernels, local repair, and clean reference remain pinned premises; their large suites were not rerun.

The fresh separate agent review repeats the four checks from a relocated copy. Its independent audit recounts the component and donor leases and checks 1,037,568 exact adapter phase values. Controls reject missing error corrections, wrong return order, wrong views, and incomplete parking. Use Python 3.11 or later, the standard library, one CPU, at least 3 GiB free disk, and a new output directory:

python3 -I -B research/borrowed-centre-storage/annexes/full-w-component/replay.py --out /tmp/full-w-component-check

Expected final line:

PASS_FULL_W_COMPONENT_ANNEX

All accepted parent files and historical records remain unchanged. The annex notice preserves source credit and licenses. Substantial OpenAI Codex assistance was used. This is independent agent review, not human peer review. The historical component is not admitted to the current selected supplier by this PR.