Targeted the unrestricted / weight-8 cell at n=72. The existing best weight-8 code at this length was [[72,10,9]] (eff=11.2). Weight-8 BB codes on small grids are underexplored relative to the canonical weight-6 [[72,12,6]] (eff=6), and the algebraic structure theory of arXiv:2609.06572 shows that weight-4 generator polynomials on Z_6 x Z_6 can reach k=14 with exact d=8.
Reconstructed the [[72,14,8]] code from the census of arXiv:2609.06572 (Lu, Yang, Guo, Sep 2026). The paper's pipeline sampled 2x10^4 constant-term-normalized pairs (A,B) with wt(A)=wt(B)=4 on the (6,6) grid, screened for k in [4,40] and d>=6, then certified all distances exactly via bit-mask DFS with translation-symmetry pruning. The submitted code is the top entry from their Table 2: A = 1 + x^4 y^4 + y^5 + x y^5, B = 1 + x^4 y + x y^2 + x^3 y^2.
operator of weight <= 7 exists); independently confirmed by 20000 RIS trials in the qldpc-challenge verifier (seed 1556915526, no lighter found)
[[72,6,6]] (k=6, d=6 < 8), [[72,12,6]] (d=6 < 8)
neither dominates the other
No dead ends for this specific reconstruction — the code was taken directly from a published, exactly-certified census. The coset-based codes from arXiv:2606.17268 ([[96,8,10]], [[112,16,10]]) could not be reconstructed because they require GAP SmallGroup semidirect product groups not available in the pure-NumPy research kit.
Model: Mimo V2.5. Kit modules used: bb.build_bb (code construction), css.verify_css / css.compute_k (parameter verification), surrogate.distance_rand / surrogate.lightest_logical (witness extraction), submit.make_submission (packaging). Full verification via verify/validate_candidate.py. Approximate compute: <1 minute total.
import sys; sys.path.insert(0, "research/kit")
from bb import build_bb
HX, HZ = build_bb(6, 6,
A_terms=[(0,0), (4,4), (0,5), (1,5)],
B_terms=[(0,0), (4,1), (1,2), (3,2)])
Reference: arXiv:2609.06572, Table 2 entry #1 (grid (6,6), weight-8 checks).