Skip to content

[Research focus] One active deliverable: arbitrary-period matching-root consistency #650

Description

@LightChainr

Current focus — 2026-09-13

The owner asked to stop the proliferation of parallel analyses and choose one mainline after a repository-wide research review. This issue is the current allocation entrypoint. Preserve evidence and correction history; this change is about attention, not a promotion of scientific claims.

One active analytical deliverable: consolidate and independently audit the arbitrary-integer-period square-site matching-root theorem. Use the probability part of #735, read at 9d29d014df28af7c635e6859d98a95ffe2b34d06, as the working proof. Main is still eb89e9422791d9e3c3a78f0e65d56912b815a7bd (#705 merged); do not describe the later PR stack as integrated.

The single question

For the actual NN square-site model on Z^2/Lambda, with shortest nonzero Euclidean period ell(Lambda), does the root of M_Lambda=P2-P0 converge uniformly to the infinite square-site critical probability when ell→infinity, without aspect, area or shear restrictions?

The submitted proof compares the two rare-sector probabilities, not their absolute size. Its critical dependency is the oblique-band argument, yielding P2/P0 <= exp[-kappa(p) N/ell] at each fixed subcritical p, with the complementary bound above the threshold. No CFT label or fitted exponent is needed.

One acceptance package — return to #735, no parallel issue

Produce one self-contained theorem manuscript, with a lemma-indexed independent proof audit and a closest-prior-theorem comparison. The manuscript contains the finite rank dictionary, imported site sharpness and matching-critical relation, lattice reduction, local-arm/Harris lower bound, disjoint oblique-band upper bound, uniform ratio comparison, unique-root conclusion and one root-versus-full-law boundary example.

The independent audit must check: local event support and injectivity; last-entry/first-exit extraction for backtracking cycles; uniform entry-vertex counting at arbitrary tilt including a shortest vector nonprimitive in ambient Z^2; NN+NNN diagonal steps; independence only on disjoint site sets; constants uniform in area and shear; exact prior-art scope. Re-running the author's code or verifying a quotation is not a substitute for this audit.

A finite arithmetic form of the same proof is now in the owner handoff: with S=ell^2, B=4 ceil(sqrt(S)) and a rigorously justified arm bound a<=abar, (B abar)^4 < (1-abar)^S implies P2/P0 <= [(B abar)^4/(1-abar)^S]^(N/S)<1. This is a finite-root sign certificate, NOT an infinite-pc numerical enclosure. No near-critical arm bound has been computed here.

What is retained but not extended this round

Stop and completion conditions

The main theorem is an author-supplied proof with local controls, not yet a new independent publication-level acceptance. #732 audited the narrower rectangular #718 claim, not all of #735. If an oblique-band step fails, report the precise failed implication and retain the narrower valid result; do not evade it by opening another mechanism project. If prior art already proves the full claim, state the overlap and narrow the contribution rather than renaming it.

Do not add L^-4, original-U/Jordan identification, all-width spectral weights, a new numerical pc, or a full birth-law concentration theorem to this deliverable. #276's original task is already closed completed: do not reopen it as though the old proof is missing. #613 retains its own full-law scope.

No branches or raw data are deleted, no running machine job is interrupted, and no PR merge is authorized. Existing external retrievals may return; do not automatically generate another four-way production/peer-check round from each result. Completed results should reduce open claims, not generate new default queues.

Previous allocation, 2026-09-12 — preserved, superseded for new execution

Current allocation — 2026-09-12

Main now includes #702, #703 and #704. Latest integration commit: 6edf775. This replaces the old September 8 dispatch order under the owner's delegation. It is a resource decision, not a ban on unrelated useful exploration. Preserve raw data and correction history; merging is not a scientific claim upgrade.

Completed — do not commission again

#702, merged: correct symplectic orthogonality gives the all-honest-torus rank-one label map, allowed five-cell support and spiral set identity from the existing digital-Alexander theorem. D=r-1 and F=(1+M)/2 are structural. No extra census is needed merely to prove these statements. Degenerate short-period quotients stay outside the stated proof scope. Raw-p C3 and the alleged failure of the two Aff(1) actions to intertwine are corrected. P5's irreducibility/polygon-rank statements and #675's unrestricted operator no-go are corrected.

#703, merged at d84fc48: covariance-support/PSD and saturated-df=0 inference repairs; full N580 covariance recovered from EXISTING delete-one shards, not simulation. Bare aspect is not rejected at the declared nominal 3-sigma cutoff in the retrospective pure-line test. Original frozen two-rung underpowered verdict unchanged. Full CI 34681074153 passed: 1626 tests, 3 existing skips, smoke/compile/C++ checks green. Test dependencies repaired; no tests removed to obtain a pass.

#704, merged at 6edf775: independently bounded rung-specific H8 ratios instead of an assumed common ratio. Seven nominated shape-plus-leakage images still miss one nominal 99.73% mean ellipsoid at B=1; bare aspect intersects at B=0. Executed C2 sensitivity, not new evidence or H8 measurement. Three new mathematical controls, all 24 fixed-bound fits, and threshold endpoints passed local 60-dps optimality checks. A full suite for this final six-file delta has not yet been observed; parent CI must not be presented as child CI. P3 correction addendum and current README are on main; the historical manuscript remains NOT submission-ready.

#575 closed completed: the original replay and subsequent existing-shard reader satisfy the recovery job. Cancel further N580 replay solely for cov(r2,r4).

#583 closed not planned: retire the standalone saturated three-angle pure-H8 acquisition. Any defensible replacement belongs to #589. Raw data, old designs, comments and branches remain preserved.

Two delegated analyses currently worth resources

1. #622 — CPU, existing data only. N145/290/725 same-observable square-site lineage, orientation-wise inverse CDF before spin0 combination, pooled anchored quantiles and aligned delete-one covariance. Report normalized reflection residual A, adjacent Delta Z/Delta A and uncertainty; no free exponent, no site/bond pseudo-ladder. One ordinary CPU machine, no new Monte Carlo or GPU; use the issue's first-delivery cap and measured runtime. Team assigns hardware. Use current main's support policy and the corrected #655 lineage code where needed, recording its source.

2. #681 — bounded primary retrieval/proof. Jacobsen 2015, Mertens–Ziff 2016, Jacobsen 2024 Reply. Give a typed finite-torus closure versus cylinder-spectrum dictionary, exact limit order, and the one missing hypothesis or restricted obstruction. No weighted transfer implementation, large enumeration or generic venue survey. This is the sole continuation channel for #321/#635/#658's bridge question. A missing paper statement is not an impossibility theorem.

Separate theory and reserve design

#275: original-U candidate-map problem remains live. Two named mechanisms must provide same-source normalized raw prediction columns; check represented coordinates and nuisance-profiled model images before another score or acquisition. N580 compatibility does not supply these columns.

#589: reserve design, not a purchase order. Find one identifiable angular/Smith contrast or an overdetermined angular closure test, with all nuisance columns explicit. Existing #205/#595/#596 calibration work is not to be repeated as new. Unanchored model cones both contain zero, so their minimum distance is zero: price source-anchored nonzero signal or fixed SNR, and disclose when a source range includes zero. N650/N1105/N2210 are candidate geometries, not automatic runs.

#636: uncommissioned until #681 provides a distinct scientific target for a restricted local transfer representation. The reason is information value, NOT the withdrawn #675 no-go.

#680/#673: genuinely remaining geometry statements can be pursued as proofs. Noncrossing does not itself imply equality of NN/NNN reachable state classes. Do not enumerate the next size simply to replace an argument.

Resource and interpretation limits

No GPU job or generic large-N campaign is justified by these next questions. Do not turn repeated views of one block into independent evidence. Do not infer a continuum field or pc law from a surviving finite shape, raw curvature sign, or a nominal confidence-region intersection. Hardware helpers, keys and power operations remain with the operators; do not alter credentials or interfere with other runs. Return a result/counterexample/precise missing input through the existing issue channels rather than opening duplicate tasks.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    priority:P1Bounded parallel analysis or a concrete reserve direction; not all run at once.

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions