Target: the unrestricted x weight ≤ 6 board cell, via a periodic bivariate-bicycle (BB) code on Z_9 x Z_13, found by an automated continuous evolutionary/mutation search over BB exponent sets (l, m, A, B) (lane w6-gap, proposal kernel shared-binomial-factor). This submission independently rebuilds and re-verifies the resulting [[234,78,3]] candidate using only this repository's own trusted tools before submitting.
The source campaign's search over Z_9 x Z_13 exponent pairs used lane w6-gap with proposal kernel shared-binomial-factor (internal candidate id 6423); screening was by a randomized lightest-logical surrogate promoted through increasing trial budgets, the same style of screen described in research/AUTORESEARCH.md step 2-3 (CSS commutation, k, then a randomized distance upper bound) but run in the external campaign above. NumPy surrogate stage: trials=8000, seed=196597815. Bit-packed accelerated confirmation stage: rung 60k, 60000 trials, seed=196597832, pair_depth=16, tightest witness on side X at weight 3.
(H_X, H_Z) from the polynomials below using this repository's own research/kit/bb.py:build_bb, and confirmed the reconstruction reproduces the campaign's own check supports exactly (same X/Z check sets, as sets of qubit supports) before packaging.research/kit/css.py:verify_css / compute_k -- CSS commutation holds, k = 78 exactly (GF(2) rank), n = 234, max check weight 6 on X and 6 on Z.verify/qldpc_verify.py (verify(doc, refute=True)): schema-valid, connected Tanner graph and stabilizer group, CSS commutation, both witnesses valid (in the kernel of the opposite side's checks, outside the rowspace of their own), and the refutation search found nothing lighter.verify/certify.py), run as part of an internal autoresearch campaign on branch certification/periodic-bb-exact-64core of this same repository. That branch is not merged to main at the time of this submission, so nothing here relies on it being externally checkable. Both sides closed exact (no logical lighter than the claimed weight exists on either side); this is reported only as the basis for the exact confidence self-declared below -- per CONTRIBUTING.md's confidence-tier policy this claim will display as an upper bound on the board until a maintainer independently re-runs verify/certify.py and confirms it.d = 3.Not applicable -- this submission mines a single already-vetted candidate from an external campaign's sweep rather than running a new search in this repository; the campaign's own dead ends (screened-and-collapsed candidates) are not part of this submission's own evidence trail.
This candidate's discovery and screening ran as an internal autoresearch campaign on branch certification/periodic-bb-exact-64core of this same repository: an automated (non-LLM) exponent-set mutation search with MILP-based exact certification. That branch is not merged to main at the time of this submission, so nothing here relies on it being externally checkable. No provenance.model is set because no generative model produced this code. This submission's independent reconstruction, re-verification, and packaging in this repository was done separately by Claude Sonnet 5, via Claude Code, operating on this repository's research/kit and verify/ modules only (no edits to verify/). Compute: seconds to a few minutes for the reconstruction and verifier pass.
A = 1 + x^6y^0, B = 1 + x^2y^2 + x^5y^2 + x^6y^0 on Z_9 x Z_13 (n = 2·l·m = 234):
from bb import build_bb HX, HZ = build_bb(l=9, m=13, A_terms=[(0,0), (6,0)], B_terms=[(0,0), (2,2), (5,2), (6,0)])