[aliromman]

This PR gives an enabling outer-transform construction and scoped twiddle bounds. A changed opposite DIF schedule leaves completed physical bits outside every later twiddle, as the forward schedule already does. This permits exact composition of codes confined to those completed bits. The complete kappa calculation does not improve. A usable native endpoint, recursive child contract, and complete finite invoice remain open.

The proof map, completed-prefix proof, independent publication review, validation record, and source manifest state the premises and evidence.

The changed opposite has exact endpoint F_plus B_inverse L_inverse over the polynomial quotient ring. It uses the retained physical slot order and each axis's own length, including short axes. It adds no physical bit-reversal pass. Its unencoded cost order and error bound are unchanged. An encoded supplier must supply its own precision allowance.

For a fixed number of completed bits and a fixed finite alphabet in GL_w(Z[i,1/2]), with supplied inverses, the chronological codes can be fused even when they do not commute. The final inverse costs O(V ell) and needs O(ell) extra integer, grid, and error-control bits. Separately, completing one whole axis costs O(V ell) and supplies enough row digits for the stated finite radix and recursive depth. This row construction does not prove a decoder bound for an arbitrary growing code bank. The completed row axis cannot also serve as the past payload.

The physical layout proof uses three block interchanges for offset alignment and three for its literal inverse. Two further interchanges can make a separated past block adjacent. The wrapper restores arbitrary spectator data, provided the native word fixes every unselected field. For e selected coordinates in K-bit chunks, its per-invocation bound is O(V ((eK)^tau+1)), with 0<tau<1. At the stated active:past ratio 64:9, enough chunk groups permit a reserve that keeps the same named completed bits, with O(Vd) initial and final moves. This is an asymptotic address construction. Base cases, compact guards, and the complete finite invoice remain open; the wrapper adds no native child or productive field.

The negative results have separate, explicit classes:

• At a nonterminal current length of at least four, a complex-scalar code that fixes the unprocessed coordinates and polynomial payload has a recordwise monomial replacement twiddle only when each output row uses one branch address. Dense row mixing within that branch remains allowed. This rejects the stated odd-line and paired current-bit codes in that class. Polynomial-valued codes, new arithmetic twiddles, and cancellation in a complete endpoint remain outside the result.
• The separate-transport proof gives Omega(Vp) when each round transports the code through separate monomial and pointwise binary ADD passes on a fixed number of complete streams of comparable volume. It is not a lower bound for arbitrary tape algorithms or recursive calls on smaller volumes.
• The accumulated-code proof assumes a stable odd port and the specified complete paired endpoint. Its exact two-round decoder has ten entries per scalar row, so it cannot be one native C tensor between the stated monomial maps. Its joint-parity argument gives the terminal Omega(Vp) bound in the same complete-pass grammar. Different endpoints, changing port labels, and multi-child native decoders remain open.

The fixed-past supplement is conditional. Assume a fully priced native endpoint h_P tensor H0_A on one fixed completed physical space P, where |P|=O(d) and P is disjoint from all later active bits. The supplier must restore all required arbitrary fields, give correctly typed recursive children, and prove a valid approximation bound on the guarded input range. Then h=S C S=(1+i)H0, h^2=iI, and h^8=I permit one final O(Vd) decode and an O(d) precision and magnitude allowance. The decoded error bound is n 2^(|P|/2) eta, where n is the number of common rounds and eta is the additive error per complete round. This does not construct the assumed supplier or establish its recursive invoice.

The frozen campaign had no distinct final independent review of the complete fixed-past closure note. This PR includes a fresh bounded review of its written conditional argument and independent exact checks: 114 tensor-power cases and 30 decoded error propagators. The earlier producer normalization check is not relabeled as independent evidence. The terminal-slot decoder candidate is excluded.

The spectator terminal annex gives a separate conditional result: complete stable portwise endpoints T_j=[[0,Q_j],[Q_j^-1,0]] on the same arbitrary spectator slices have an ordered product Q_all M, with one explicit affine map M. Under the proof's commuting-phase and disjoint native-axis premises, the inverse uses one gathered C_D call and paid monomial wrappers. The phase qualification is essential: layer-local conjugating phases must act identically on both spectator slices, and the actual native-to-H0 conversion leaves the code before the current DIF twiddle. The unchanged paired controller does not supply the required complete portwise endpoint. Native gather and child admission, full costs, precision, and recursive closure remain open; no complete kappa improvement follows.

The annex replay runs the unchanged terminal kernel check and independent formal product-order check in isolated copies, normally and with optimization. All four new receipts equal the frozen receipts. These checks cover 1,344 slice columns, 28,224 kernel entries, and 64 independent formal cases, with omitted-control and inverse-order controls. The general theorem and phase qualification use the written proofs. The paired-controller 2912 witness remains an inherited proof result; its full source and route replay was not run by this annex.

python3 -I -B research/completed-prefix-outer/annexes/spectator-terminal/replay.py --out /tmp/spectator-terminal-annex

Expected final line:

PASS_SPECTATOR_TERMINAL_ANNEX

The separate spectator review binds the exact annex and repeats all four checks from a relocated copy. It is an independent agent review, not human peer review.

The publication base is b58ad9ef4d4e861b0eaa5dba41742bdc072dca51. Its selected record names the conditional PR #4470 construction, with kappa=60403/20000000=0.00302015. Its inherited native/all-column, owned-erasure, whole-job, compiler, precision/tape, analytic, prime, and recovery hypotheses remain; uniform words are not materialized. This PR establishes no transfer to that construction.

The consumed stream and transform source is historical commit 3ae9a6f3ac74ded022ab3b4872a7357a621af6fb. Its selected record and assembly belong to PR #1094. All numerical margin checks refer to that historical assembly. The actual 112-port word remains pinned to the historical replay archive at bce54fb3f895cbe5cd4c0eb477c3afeed100190f. The independent review checked its exact archive member and ordered ports.

Use Python 3.11 or later with its standard library, one CPU, at least 3 GiB of available memory, and at least 3 GiB of free disk space. From the repository root, use a new output directory outside the package:

python3 -I -O -B research/completed-prefix-outer/replay.py --out /tmp/completed-prefix-outer-replay

Expected final line:

PASS_COMPLETED_PREFIX_OUTER_PACKAGE

To repeat the separate fixed-past check:

python3 -I -O -B research/publication-reviews/completed-prefix-outer/check_fixed_past_independent.py --package research/completed-prefix-outer --out /tmp/completed-prefix-fixed-past.json

Expected output:

PASS_FIXED_PAST_INDEPENDENT_BOUNDED_CHECK

All eight package checks passed in normal Python, an optimized cold export, and the independent relocated replay. They need no Git, network, or external dependency. The results include 174,760 independent forward/opposite entries, 9,455 separate coefficient identities, 27,988 separate opposite matrix entries, 455 layouts, and all two-round decoder rows. Typed receipts match the frozen results; four checker self-hashes change to the actual portable source hashes. All other fields match exactly. The portability diff changes I/O and CLI paths only.

Controls reject a shared long/short recursive length, missing input bit reversal, and the old opposite twiddle. Removing the finite example's twiddles gives the expected four-entry decoder. The scalar-permutation countercontrol passes the current-twiddle criterion. The new fixed-past controls reject past normalization to H0, moving a current-bit code through its twiddle, omitting the final decoder, and treating an odd past power as a maximum-norm contraction. Finite checks protect the conventions; the all-size claims rely on the written proofs and stated assumptions. No complete multiplication program, new native supplier, recursive encoded child, finite compact-engine invoice, or selected PR #4470 program was run or admitted. The open next questions give the remaining construction obligations.

The source notice retains OpenAI's manuscript authorship, Douglas Colkitt's project notice, and the community lineage. The stream and transform primitives are inherited from that manuscript. The supplier notices credit @icekylinx for the gauge and supplier pipeline, Hao Yan @FanPUYen for the T16 integration, Jeremy Kothe @jeremykothe2 for the PR #799 X4 word, Chafik Boukhalfa @chafreaky for the PR #741 frames and PR #685 owner construction, and Rohan Arun @rohanarun for the native recurrence correction. The package notice preserves their source evidence, all earlier credits, and Apache-2.0 notices. Substantial OpenAI Codex assistance was used for the proofs, checkers, separate agent reviews, and packaging. No independent human review is claimed.