Repository navigation
Research: sharp four/five-source dirty-connector boundary (no new kappa) - #943
Conversation
… kappa)
Add one new complete call type DUO_{j,k}: z += B_j x_j + B_k x_k to the frozen
universal-connector alphabet (the only changed premise is 'no combined
queries'; the call is charged in full). Exact finite result: the five-source
universal source word has unique Pareto profile (10 calls, 12 toggle-adds)
versus the frozen (12, 12), so the all-31-row batch runs 72 root queries at 74
toggle-adds instead of 74 at 74 - the support bound 72, unattainable in the
singleton alphabet, is attained exactly here. k=9 is excluded by a complete
zero-cover census of all 7,888,725 nine-point sets; W>=12 by parity plus the
reproduced 945-pairing control. n=3 (5,6) and an n=4 (6,8) side row included.
Predecessor credit: Ross Richardson's research/universal-connector-boundary
(#943 / #934).
Comparison: #1340 and
#1370 mutable-source shear (Arham Tahir),
discussion 1352; #1185 f(6)=16. Prepared by
Navinor with help from opencode. Apache-2.0.
|
DONE / 2026-10-15T17:04:03.599612+00:00 / priority revision5. Independently reviewed local module published as #2680 at0504da6f76bd50c048db0ec3021f0e1df86947e6. Exact all-field paid-repair rank theorem, complete binary oracle and fully restored priced pilot; native general costs and selected consumer remain open. Claim released. Original claim retained below: CLAIM — LOCAL ONLY, expires 2026-10-15 18:10 UTC. Starting from freshly fetched canonical main0fdd7d82, priority4, task connector-alphabet-change (picker fc48de8a055a955bfdc7a38a0308396d; the prescribed945-pairing control passed). Object: the fixed-map dirty connector chart interface in woklerx's retained The chart and correction are paid operations, not free observations. Any low-rank factorization still needs an actual implementation and complete charge. This targets a scoped adapter invariant, not a minimum over all circuits or a selected-supplier application. The equal-kernel theorem, dirty cancellation and standard row-space argument retain prior credit. No new exponent, all-size physical compiler or unrestricted tariff optimum is claimed. Overlap: all817 open PR records at16:28UTC and1279 Discussion records at16:32UTC plus local package notes searched for chart transport, residual/rank repair and kernel mismatch; bounded metadata coverage is not a novelty guarantee. Stop after one proof, independent finite certificate and priced pilot, or a counterexample. Source inputs will be pinned to main Git blobs/SHA256; no frozen predecessor edits. Prepared by ywz65 with substantial OpenAI Codex assistance. Status/artifact will be updated here. Discussion posting is currently API-rate-limited; this comment is the canonical claim and will receive the final artifact/update. The predecessor chart criterion is retained at main path research/dirty-connector-chart-transport. |
Recursive-interface research construction and scoped negative result; no new kappa.
Carrying partial source sums in one dirty binary connector avoids repeated restoration between output stars. This package proves a universal four-source construction for arbitrarily many outputs and identifies the first obstruction: all 945 canonical five-source paths using each source twice cover at most 31 of 32 possible readout differences. An explicit twelve-toggle path covers all 32. For the batch containing all 31 nonzero five-source output patterns, the complete optimum is 74 queries in the stated straight-line alphabet; the 72-query support bound is unattainable and parity excludes 73.
Two literal repairs and their full recursive lowerings are supplied. The varying-factor fixture uses 182 primitive calls versus 278 for separate column stars, with the same four borrowed dirty rows. The proof prices source/coefficient access, every child and inverse, scratch, frames, preparation, dispatch and tape work. This is an abstract query-model result, not selected-supplier integration, a measured speedup or a general circuit lower bound.
Only
research/universal-connector-boundary/is added. Its README, exact certificate, source pins, claim, validation and attribution form the review package. Discovery base:91a2094f2e022930069e8cfdc810a3ea1b81cd09; refreshed submission base:47481c89fd30a43f8f882eb4b391e89c29273fd0. Main's new PR890 selection at conditional κ = 0.001661957774647029 was followed to its construction/review and remains unchanged.OS assignment seed 52, recursive-interface track. PR934 is the direct binary-query predecessor, pinned at
f3100aec6ce6a7b061987875d5e9a8b93801ac0b, with original README/NOTICE retained. Credit also goes to PR889's fanout interface and Chafik Boukhalfa's PR628 dirty commutator. Concurrent PR938 counts one-source equality cases; this adds a multi-source universal-path criterion and exact failure boundary. The full queue refresh reached #939 and the immediate check included #940/#941. No pinned coordination issue existed, so the claim and result link are local. Prepared by Ross Richardson with substantial OpenAI Codex assistance, under Apache-2.0.Validation on macOS / CPython 3.14.6:
Expected:
PASS_UNIVERSAL_CONNECTOR_BOUNDARY pairings=945 max_coverage=31 queries=74 leaves=182andPASS_UNIVERSAL_CONNECTOR_PACKAGE relocation=1 regeneration=2 corruptions=6.make selected-record-check, all five existing selected-result tests and staged whitespace checks pass on the final commit. The checkout is clean.No full repository regeneration, selected cold replay, hosted-CI result, proof-assistant theorem or independent human-review pass is claimed. Physical supplier realization, common labels/frames/owners, whole-row charging, all-size compilation, tape/precision and outer analytic obligations remain open.