Skip to content

Add native joint centre-query transfers on sixteen selected ports - #1969

Merged
Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
aliromman:codex/joint-query16
Oct 14, 2026
Merged

Mister-Meeseeks merged 1 commit into
CrocSwap:mainfrom
aliromman:codex/joint-query16

Conversation

@aliromman

@aliromman aliromman commented Oct 14, 2026 •

Copy link
Copy Markdown
Contributor

This local construction and enabling theorem uses two disjoint eight-port minors to carry centre queries in the joint X/Y state. It gives invertible forward and backward data maps, preserves all sixteen selected X-plus-Y sums, and allocates no new array field. Changing the selected source baseline permits a local query exchange without a Schur-unit assumption. The package supplies 37 complete native local words for the actual historical complex supplier.

The complete kappa bound is unchanged. This is a reusable local operator interface, not a complete changing-chart controller. The package includes an exact witness against the current controller's selected-only recovery interface and states the remaining escapes.

Evidence at the reviewed commit: ordered proof, literal native certificate, independent package review, replay receipt, and source pins.

The ordered theorem uses disjoint selected sets A and T, their complement U, K=K_A+P+K_U, J=I-KR, RP=LP=I, and the stated support conditions, with PL the T-coordinate projector. The forward companion also uses RK_A=I. Put

M=R K_U,  Z=K_U-PM,  H=I+ZR.
RZ=0,  H^-1=I-ZR,  JH=I-[K_A+P(I+M)]R.

No inverse of M and no interchange of ordered factors is used; M may be noncentral or singular. A future A-logger E_t=I+N_t, with only A output rows and no A input columns, supplies the second chart's record. In the conserved backward map, let u=E_t y, a=ell_s x, and b=R_s u. Its records are

R_s y' = a,
ell_(t,A) y' = R_t y,
ell_s x' = b + M a.

The selected X record is b+Ma, not b alone. The ordinary target rows perform x_U-y_U. The forward companion puts the full current X query in T and moves the old T query into A. Both maps preserve each selected pair sum and have explicit algebraic inverses in both orders. The conserved sums are logical identities; their X and Y terms have different native endpoint frames. The words are companions, not inverse circuits. The actual instance has R of size 8 by 112, K=R^T/4, and M=12I. The general theorem is algebraic; the priced native instances use these actual dyadic matrices.

The native contract supplies all 28 B_s_t cases for 1 <= s < t <= 8 and all nine F_s cases for 0 <= s <= 8. Its cover views change the owner and apply a separate payload translation on every tensor layer. Each full-cover sweep must finish before the next event. A private row body completes at one owner before C0 is reused at the next. The new work uses one existing owned C0 field. All 480 ordinary helpers retain their arbitrary logical values and prescribed native endpoint frames; all eight owned private inputs may be arbitrary, are discarded, and have zero outputs.

Complete local word Real scalar primitive bound Added rank-one children Native rank mass Added cover maps Native wrapper maps
Backward B_s_t 214986 24 5546 528 48
Forward F_s, s > 0 211210 8 5530 288 16
Forward F_0 211210 8 5530 0 16

The retained core contributes scalar bound 202138 and native mass 5522 to each row. These are local counts in the retained normalization. Map counts are not a stream or address implementation price. Formal scalar prefix guards are 73 bits backward and 55 bits forward; they are not recursive precision bounds.

The exact branch and source base is bce54fb3f895cbe5cd4c0eb477c3afeed100190f. The runtime input is the frozen PR #1042 selected-replay archive, also used by the historical PR #1094 assembly. PR #1862 supplied the preceding source-code construction and the unchanged source loader at commit 7649d9ceecec949541bec4bbde5929344166ba59. That loader is included here; this package has no runtime dependency on a PR branch or predecessor Python package.

With Python 3.10 or later, its standard library, Git, and the base object present locally, run one process from the repository root:

python3 -I -O -B research/joint-query/replay.py --repo . --out /tmp/qq16-replay

Expected output:

PASS_PORTABLE_QQ16_LOCAL_JOINT_QUERY
37 native cases; 712 scalar input columns per orientation; 16384 exact new native columns.
B: scalar 214986, native mass 5546. F: scalar 211210, native mass 5530.
Local construction only. The complete supplier and kappa calculation are not established.

Root and isolated independent cold replays passed with byte-identical mathematical receipts. The independent package run copied only the declared files and used -I -O -B; all checks remain active under -O. Verification covers both 712-column common-chart scalar maps, both data inverse orders, algebraic composition, all helpers and private outputs, all sixteen selected sums, all 1008 actual chart routes, all 37 native paths, and 16384 exact Gaussian columns including scalar phases. Timing, private-clear, inverse-sign, translation, wrong-owner, phase and odd-denominator controls pass. The independent audit also rejects a wrong owner map that preserves the phase ledger.

The controller witness and its independent review show that the fixed first conserved F0/B1 pair and retained ordinary bank schedule cannot restore every selected input from selected outputs, completed-X descendants, and all five F query histories, even over the rationals. Later homogeneous selected-only linear controls do not remove this witness. Changed first-pair maps, ordinary-to-selected reads, affine helper state, and paid query operations remain outside its scope.

The bridge proof supplies the inherited core embedding and endpoint adapters. Core bytes and local frames are checked again; the old recursive child kernels and the global controller witness are not rerun by this entrypoint. Algebraic inverse formulas do not establish legal inverse reads at native output frames. New child/helper hosting, controller closure, endpoint recovery, stream/address costs, recursive precision, and the complete supplier invoice remain open. No complete supplier replay, repository-wide suite, or new kappa calculation was run for this local package.

Related work, not a consumed dependency, is Chafik Boukhalfa's (@chafreaky) PR #1877. Its proof at 6d455d2d71eb35ecf8e562711e5f719b994fe3f3 gives source-only endpoint-support and fixed-read frame obstructions. The present source depends on old X_T and uses explicit native transitions. It does not assume free commutation through the fixed source frames. No code or certificate from that PR is consumed.

The retained gauge notice credits @icekylinx. The t16 notice credits Hao Yan (@FanPUYen) for integration, Jeremy Kothe (@jeremykothe2) for the PR #799 local supplier, and Chafik Boukhalfa (@chafreaky) for the PR #741 frame work. The slot-word source credits @rohanarun for PR #987 and NewBArch for PR #869; no account is inferred for NewBArch. The original notices, complete earlier lineage, and campaign contribution chronology are retained in NOTICE.md and its linked source notices.

New research, implementation, and independent agent review used substantial OpenAI Codex assistance under the user's direction. No independent human review is claimed. New material uses Apache-2.0. The open next questions identify the remaining controller, hosting, and complete-cost work.

@Mister-Meeseeks
Mister-Meeseeks merged commit fcea691 into CrocSwap:main Oct 14, 2026
20 of 61 checks passed
@Mister-Meeseeks

Copy link
Copy Markdown
Contributor

Batch review at a6e6ad3936f67b8f63b5b68c5e75d540610a65e7.

Merged to main, preserving the original contributor commits and package attribution.

Reviewed scope: Joint-query package supplies ordered source/data identities, explicit inverse and conserved-sum checks, and QQ16 local native replay with full local dirty-helper restoration and phase checks. General singular/noncentral cases are covered by the written algebra and finite controls. Physical global placement, chart-controller composition, all-size/tape/precision contracts and the archived global witness are not admitted or rerun here.

The pinned scoped command passed; the batch ledger contains its command, input hashes, environment and log. This incorporates the research/tooling contribution and does not change the selected κ or discharge the inherited multiplication-theorem assumptions.

Batch review, attribution and evidence. Thank you for the contribution.

@ywz65

ywz65 commented Oct 15, 2026 •

Copy link
Copy Markdown
Contributor

DONE / 2026-10-15T18:25:42Z / priority revision 5

Published and independently agent-reviewed: PR2723, exact 18f0843079dd03f509a6724328ceba2fec325022. The exact integer-row adapter preserves all five original QQ16 output files, stdout and all nine complete scalar maps. Three paired full-caller runs gave median 26.82 s versus 20.63 s (1.30x); timings are descriptive. Registered native CPython 3.12.7 replay and three source/domain test groups pass; independent review repeated the primary and tests and bound all ten committed files. All 29 runtime source pins remain enforced. This is opt-in verification engineering, with the original hosting, movement, precision and supplier obligations unchanged.

Claim closed; no further round is active from this investigation. Original claim retained below.


CLAIM / LOCAL-ONLY / 2026-10-15T18:07Z / priority revision 5

I am checking a source-bound opt-in integer-row implementation of the scalar replay inside the complete frozen QQ16 caller. Source: #1969 (a6e6ad3936f67b8f63b5b68c5e75d540610a65e7), current main 6bd33bc0b87a01e9c9d8b1c01576f3ddbcaee850.

The changed boundary is only verify_scalar.replay: share a positive exact denominator across each sparse row and return the original Fraction rows. Every event/gate guard, all 712 basis columns, nine positive/composition/corruption replays, native and ordered checks remain. No verdict or mutable input cache, no frozen-source edits. A pilot complete caller passed with all five output files byte-identical (24.26 s reference, 18.50 s integer-row arm on native CPython 3.12.7; preliminary single runs, not a timing guarantee).

Next: paired full-caller timings, independent rational/dyadic/odd-denominator row controls, source invalidation and peer review. Stop if exact output/coverage equality or material whole-caller gain fails. No construction, supplier, hosting, precision or kappa claim. Claim expires 2026-10-15T20:07Z. Prepared as ywz65 with OpenAI Codex assistance; the local words and original verifier remain credited to Ali Romman and the retained lineage.

Mister-Meeseeks pushed a commit that referenced this pull request Oct 15, 2026
Adds docs/research/qq16-controller-scope-map.md, a source-bound index of
six QQ16 observation contracts: PR #1969's retained witness, a public fork
catalogue of all 252 retained first-pair choices, PR #2079's B2 witness
separation, PR #2093's invisible family under all nonzero-chart initial
reads, PR #2174's recovery given the recorded post-B1 cut, and PR #2346's
final-record obstruction. For each it states the observations, their
timing, the established statement, the excluded hypotheses and whether
the source is retained, pending or a fork artifact. One sentence in the
bottleneck brief's joint-query paragraph points to the map; the existing
guidance text is unchanged.

Navigation only: no new obstruction, supplier or kappa claim.

Submitted by @commdata2338 with substantial AI assistance (OpenAI Codex
research and implementation; packaged with Claude Code).

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01X86sCSXdoDd61SDFfC1PKG
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants