[aliromman]

This construction puts centre records into existing native data fields while preserving every arbitrary input baseline. It packages productive row capture, clean selected exchange and its dual, row-involution responses, and complete D25 forward/dual services for a fixed past reader. A separate route32 response has two dyadic capture codes. The direct clean comparator gives complete 2,912-scalar words with a different, larger native profile. These local results do not improve the complete kappa calculation. A recursive helper host, closed caller, finite map price, terminal decoder, and complete precision bound remain open.

The proof, exact claim and review index, source pins, fresh validation, and independent publication review give the adopted scope. All 640 input fields are arbitrary. The complete local words preserve the 416 outside values; that fact does not give those values productive progress or construct a recursive host.

For the actual sixteen selected ports, L=V_P^-1 V=[I,B] and C=[[I,B],[0,I]] have a dyadic code costing 1,760 scalar primitives. Formation keeps the arbitrary Ly baseline. Line-entry reuse keeps the same reader. Clean exchange instead sends each selected pair (Lx,b) to (-b,Lx), so the old baseline remains in selected X while selected Y holds a clean centre. Its inverse-adjoint keeps the chronological order and every unitary native operator, including global phase. A selected-field conjugate is implemented by explicit instruction labels under matching physical types.

The row completion sets U'_P=V_P^-1(-2I-V_R U_R) and A=I+U'V. It preserves all 96 ordinary rows and gives A^2=I. The general rank statement is rank(A-I)=16; rank(A-E)=16 is separately checked for this source. The complete D25 reader uses A_t, then the past reader L_s, then A_t again to restore X. The forward service adds D_st x; the complete dual subtracts D_st* w, including the native prefix, actual owner paths, DATA views, full-cover barrier, and late query. Both providers and pairs (0,1),(0,2),(1,2) are checked. The physical input level is t-1, including pair (0,2).

The literal cost table keeps the different interfaces separate:

Complete local word	Real scalar primitives	Native mass
Formation / same-reader line entry	11,640 / 9,880	1,808 / 1,792
Clean I entry / Q entry	8,600 / 8,600	1,808 / 1,792
Native A / native A transpose, either direction	9,416 / 8,936	1,808
Complete D25 fixed-reader forward or dual	14,896	1,792
Route32 raw, either direction	4,312	1,792
Direct Fsimple or Bcustom	2,912	1,984

The D25 service has 1,520 children and at most 4,384 map calls. The direct pair has 96 H1 + 128 H8 + 96 H9: 320 children and at most 448 native side maps, with no boundary adapters. It is distinct from the separate 4,632-scalar phase-coded word. Native mass and scalar counts alone do not compare complete recursive costs. The original formation derivation retains preliminary prices 2,048 and 3,696; the final reviewed actual pivot code and query prices are 1,760 and 3,424.

Route32 has rank(E-I)=32, determinant 3^24, and a rational inverse with factors of one third. Its two capture codes use only dyadic units and cost 1,870 and 2,486 scalars per direction. Its rank minimum is restricted to the stated route-homogeneous clique partition family. The formal q C q code diagonalizes Q E q; it is not a raw decoder. The producer-only common-W lift and whole-W reflection decoder are not adopted.

The historical budget appendix uses b=67018507072324140917/40000000000000000000000, from the selected certificate and assembly at bce54fb3f895cbe5cd4c0eb477c3afeed100190f. Three clean Q calls allow 196 extra H1 or 98 H2; one I and two Q allow 180 H1 or 90 H2. The coalesced fixed profile allows 198 H1 or 99 H2, reduced to 134 H1 or 67 H2 after a stipulated 64-H1 cleanup. These are alternative conditional allowances. They do not admit the composition, cleanup, or a 224-H1 terminal decoder.

The publication base is b58ad9ef4d4e861b0eaa5dba41742bdc072dca51. Its selected record names PR #4470. The historical inputs in this package have different source and caller contracts. No transfer to the current selected construction is claimed.

The native lineage uses the core source from PR #3133 at 51eb9c4ee9e4722010d308aa3309e039d3940cc2, the provider copies in PR #3148 at f2927a4b8afcf0701a6b945e14f30de1d94f5176, and the recipe tools from PR #2588 at 8cd9dc88e24f062e2ca2fa7f4844a3d3319381fb. The retained provider credit identifies @dcharb78 for the coupled folds and helper endpoints. Jeremy Diviney (@jeremydiviney) supplied the provider copies and packing/exchange integration, as recorded in the exact PR #3148 source. The PR #2588 notice credits Rohan Arun (@rohanarun) for native lowering and owner construction and Cinna Davis (@1337isnot1337) for the borrowing and warp lineage of PR #2226. The package notice records campaign contributions, unchanged predecessor notices, and Apache-2.0 licensing. OpenAI Codex gave substantial assistance. The separate agent reviews are not human peer review or formal-kernel verification.

Use Python 3.11 or later, the standard library, one process at a time, and at least 3 GiB of free disk. Python 3.14.7 on macOS was tested. From the repository root, use new output directories:

python3 -I -B research/native-centre-capture/replay.py --out /tmp/native-centre-capture-normal
python3 -I -B research/native-centre-capture/replay_budgets.py --out /tmp/native-centre-capture-budgets-normal

The expected final lines are:

PASS_NATIVE_CENTRE_CAPTURE_REPLAY
PASS_NATIVE_CAPTURE_HISTORICAL_BUDGETS

The new extraction ran all eleven main stages and three budget stages in normal mode. Every main receipt matched the frozen bytes. The separate publication reviewer repeated all stages from a relocated copy in optimized mode with Git unavailable. The historical freezes also retain their earlier normal/optimized checks. The budget receipts match in every field except the two declared adapter self-hashes; the exact I/O diffs replace only historical Git reads and one output destination.

The fresh independent checkers enumerate all 512 spatial basis columns for each of the 297 new route32 kernels and each of the 209 direct-clean kernels, including global phase. The complete content replays keep arbitrary inputs and owner order. Controls reject missing transfers, wrong exchange signs, omitted source restoration, wrong past readers, wrong dual chronology, and wrong translation order. The budget replay recounts 24 earlier raw/changed words and eight clean words and verifies 176 low-displacement intervals plus the clean and coalesced strict thresholds. Inherited phase tables are bound by exact ordered-record equality. No complete historical compiler regeneration, full repeat of inherited phase censuses, current selected-supplier replay, or complete host, caller, precision, or kappa calculation ran.