Repository navigation
Research: reverse F2 transfers attain sharp six-pass connector - #2065
Merged
Merged
Conversation
This was referenced Oct 14, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Result
Adding the reverse F2 root transfer
D:z1+=z0lowers the exact universal five-source/31-output connector threshold from PR1615's one-way 8 transfers to 6, at exactly two calls per B and A operator. The new complete word has 78 root calls: 10 B + 62 A + 3 C + 3 D.The witness uses the restoring
(CD)^3loop in GL(2,F2); four sources change physical ports between their paired calls. This answers PR1615's explicit reverse-transfer open question rather than extending the old one-way census.Evidence
Scope and complete price
This is an abstract incidence-minimal F2 connector result, not a new physical supplier or kappa. Every child port, both transfer directions, scratch restoration and parent routing/admission/dispatch remain charged. Under equal complete child/transfer prices the favorable condition against the old74-query single-root repair is
6*sigma + Delta H < 2*c. Unit root-call prices alone do not improve that repair.Credit Ross Richardson/PR943, woklerx/PR1438, Gareth Davidson/PR1498, Hanson Sun/PR1615, Cinna Davis/Discussion1412 and Sagar Thakrar/seed58. Characteristic-zero signed3x3 PR1675/1682/1983 are different targets/rings. No selected record, frozen source or shared index changes.
Exercised verification
Observed:
No full-suite, formal-kernel or GitHubCI-pass claim.
Independent seed
628ce1c7a4abbbac17914cb067826ef4, recursive-interfaces / connector-alphabet-change, exact basee112d71c8d7bb32435debcc54df932593dd8223a. Claim: #research-claim-and-completion-record .Prepared by Adam Gall with substantial OpenAI Codex assistance; Apache-2.0. Independent implementation paths are not independent human review.
Research claim and completion record
Status: completed. The claim record is preserved here so this PR is self-contained. Historical coordination links in the committed research metadata refer to this record; the separate coordination discussion is being removed.
DONE — PR #2065 , full commit2721b1e31d718a6e70266a99521f629ce2e1d259. Adding reverse F2 root XOR lowers PR1615 universal five-source/31-output incidence-minimal threshold from8 to exactly6 transfers:78 complete root calls,10B+62A+3C+3D. Four sources change physical ports. Moving-frame source/readout proof reduces the four-transfer exclusion to175 word/partition pairs; two exact engines agree across2,102,100 reduced candidates (max31/32). A literal (CD)^3 witness attains32/32. Full180-factor+6-transfer arbitrary-dirty noncommutative replay,522 dense cases and162 semantic mutations pass. Focused optimized CPython3.12 replay and registered package replay both PASS; parent independently replayed and reviewed before publication. No physical supplier, paid-moment or kappa improvement claimed; unit transfer/child prices still give78 versus the old74-query repair. Complete unequal-port tariff and all added transfer/restoration/dispatch costs are retained. No GitHubCI-pass claim. Credit PR943,1438,1498,1615, Discussion1412 and seed58; signed characteristic-zero3x3 predecessors/parallel work are distinct.
Original claim: seed628ce1c7a4abbbac17914cb067826ef4, recursive-interfaces / connector-alphabet-change; basee112d71c8d7bb32435debcc54df932593dd8223a. Added exactly D:z1+=z0 to one-way C:z0+=z1; bounded question was a restoring loop below8 with exactly10B+62A and arbitrary dirty roots/generic maps. Claim fulfilled, not expired or extended.