Skip to content

Conditional κ ≥ 0.0037260455556 (+0.11194% over #7669) with fifteen banked transfer pivots - #7674

Open
aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/p104-banked-transfer-pivots
Open

aliromman wants to merge 2 commits into
CrocSwap:mainfrom
aliromman:codex/p104-banked-transfer-pivots

Conversation

@aliromman

@aliromman aliromman commented Nov 18, 2026 •

Copy link
Copy Markdown
Contributor

Conditional κ ≥ 0.0037260455556. Fifteen rational changes of sender and receiver basis remove 240 paid recursive calls in the 16-stage construction from SYLforge’s #7669. Every public operator and productive rank count is preserved. The new bound is 0.0000041662072 (0.11194%) above the parent’s directed upper endpoint. That increment is 1.758 times the incremental step in #7663. The cumulative difference from the selected 0.00363005817 is 2.64424% and includes the work of the predecessor authors. Proof · Signed receipt · Parent · Previous step

Quantity #7669 This PR
Paid moves per native call 6,157 6,142
Paid calls in all 16 stages 102,064 101,824
Productive fields 21,152 21,152
Target minus paid rank mass 1,328 1,328
Raw crossing, directed (0.0037218793483, 0.0037218793484) (0.00372604555571, 0.00372604555572)

Construction

A paid transfer into B can have several consumers sending to the same C. At each of fifteen sites, another sender can form one linear combination of those outgoing rows using its own saved fields and side information at C. Subtract that combination from the other outgoing rows, remove the incoming transfer and recover every old read locally at C. Each witness includes an exact equality on arbitrary independent inputs and a legal schedule. Seven edits use the forward circuit and eight use the transposed scalar circuit. The actual implementation executes the final forward graph. Fifteen witnesses · Construction and scheduler

Verification

  • All 1,501 native output rows agree with the public specification. The checker verifies 8,533,666 exact coefficient products. The signed caller returns 3,536 root roles and 17,616 independent helper jobs.
  • The operator replay checks rank, signed Pauli action and absolute phase on 3,024 chart pairs and both adapters on all 104 main ports. All 512 forward and 512 adjoint columns agree on each of the ten distinct new routes at surviving slots.
  • Two directed evaluators bracket the full signed crossing. The compiler and input-lift checks pass at 0.00372604555567. All 54 strict and four functional outer predicates pass, with minimum margin 10⁻¹⁵.
  • All eight registered checks passed on Python 3.12.16 on arm64 macOS, including in a Git-free export of the 29 committed package and input files. Failure controls cover sender sign, causality, coframes, absolute phase and inexact odd ticks. Replay manifest
python3.12 -B scripts/research_package.py --package research/p104-banked-transfer-pivots

Limitations and review request

The exact-grid realization charges 15,290 private full-field homes per native invocation, or 244,640 per raw caller, plus the public fields and compiler/adapter workspace. The physical proof charges full-extent erasure, ordinary arithmetic, bank selection and restoration with fresh finite constants. It uses the selected five assumptions for uniform native realization, owned erasure, compilation, fixed-tape precision and analytic recovery. A materialized full multiplier and independent scientific review remain open. The separate rounded, fixed-tariff implementation in #7669 has its own service and error obligations. A reviewer can start with witness zero, then check the charged bank schedule and common-root precision substitution in sections 3–4. Selected application

Sources and credit

Base e54523517d0f03e58352e647c403a5af3e525a57. The native word ee251f60…b08d and 16-stage caller come from SYLforge’s #7669 at eaef5d1580e5761feaed09262b964871f3e5f553. Source contributions include DreamingOfClouds’ joint native reconstructions, Clear1nsight’s word and helper improvements, Rohan Arun’s signed caller lineage and Jerry Chen (CubeJerry)’s selected complex and BIT constructions. Earlier bank support and destination repair work is credited in the package. The exact private-field argument and checker methods build on #7663 and the selected application. Ali Romman received substantial OpenAI Codex assistance in the new research, implementation, verification and writing. Original source notices · Exact source inventory · Coordination discussion

Hosted verification and shared CI prerequisite

GitHub ran all eight checks on Linux at final head 28b2af322a1fab01b3734365988930920f9dda48. The planning job and the historical replay also passed. A fresh full-history clone passed the committed-byte, executable-mode and source-ancestry preflight for all 29 package and input files. The eight vendored files also match the Git blob IDs at the two published predecessor heads. Hosted scientific replay · Complete CI run

The first hosted attempt stopped during onboarding because the base CI guide still named #6030. A separate commit carries ToikY’s copy of Clear1nsight’s existing repair. Its five file blobs match #7257 exactly. The complete scientific contribution is commit 26b6c469aa6c4aa04322ce25a22150f90e5fb14e. Scientific diff · Shared CI repair · ToikY’s source · Original onboarding failure

aliromman and others added 2 commits November 18, 2026 01:39
Rebuild the sixteen-call parent and check the exact native and caller operators, phase, fresh bank costs and complete conditional kappa application.
Reuse Clear1nsight's exact CrocSwap#7257 guide and historical-source snapshot repair, already validated by the hosted checks on CrocSwap#7644. The research package bytes are unchanged.

(cherry picked from commit 694667c)
(cherry picked from commit c58206c)
(cherry picked from commit 75cdd54)
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.

2 participants