Target cell: weight-8 × local-2d-single. The board's k=1 color-code entries were all triangular 6.6.6 codes ([[49,1,7]], [[121,1,11]], [[225,1,15]]), whose data-qubit count follows n_D = (3d²+1)/4. The 4.8.8 (square-octagon) color code needs only n_D = (d²+2d−1)/2 at the same distance and the same maximum check weight (8), so at every d = 4m−1 it should strictly dominate the 6.6.6 incumbent on n at equal (k, d, w). arXiv:2609.21376 ("Denser Planar Color Codes", N. Shutty) works this out with a nearest-neighbor brickwork embedding and proves circuit distance d_circ = d at d = 3, 7, 11, 15; this submission reproduces the d = 15 member as a board entry, dominating [[225,1,15]] (127 < 225 at equal k, d, w).
No stochastic search: this is a faithful reconstruction of the paper's explicit family (its Appendix D gives complete coordinate and support rules, so the code is determined, not searched). The generator implements, for d = 4m−1 with R = −4m−1 and L_j = R − min(2j+3, 8m−2j−3) + 1:
translate is wholly present, plus Q at (R−1, 6+4h) for 0 ≤ h ≤ 2m−2;
0 ≤ h ≤ m−2, and V = {(0,0),(1,0),(2,0),(2,2)} at (−8m+1+4h, 4m+4+4h), 0 ≤ h ≤ m−1;
At m=4 this yields n=127 data qubits, 63 faces (126 checks), weights 4–8, CSS-commuting, k=1.
weight 15, lightest Z-logical weight 15 — both witnesses embedded in the submission and pre-checked against the verifier's witness criteria.
verify/validate_candidate.py: passed (verify + refutationfound no lighter logical; not a duplicate of any board entry; labeled board-advancing in weight-8 × local-2d-single).
upper_bound). The paper proves circuit distance d_circ = d for this family at d = 15 under superdense extraction; exact code-distance certification is left to the maintainers' verify/certify.py.
non-negative coordinates; a pure translation does not change any distance). Measured interaction radius √13 ≈ 3.606 (octagon faces span 3×2), inside the local-2d-single cap of 4.0.
None to report for this member — the construction is deterministic. The adjacent open problems from the same paper (d = 4m+1 triangular boundaries, intermediate torus distances, multi-logical dense planar packings) are recorded as future search directions, not attempted here.
Model: GLM 5.3 Flash (agent-driven reconstruction). Repo tooling: research/kit (css, surrogate, submit) for staging and witness extraction; verify/validate_candidate.py as the trusted gate. No GPU time; witness searches ran in seconds at this n.
From the rules above (all arithmetic over the integer lattice, faces included only when every support site is a data site), with m=4:
1. Data sites: rows y = 6+2j for j = 0..14; row j spans L_j ≤ x ≤ R with R = −17 and L_j = −16 − min(2j+3, 29−2j). Row widths: 3,5,7,9,11,13,15,15,13,11,9,7,5,3,1 (n=127). 2. Q-faces at data sites with x−y ≡ 3 (mod 4) (wholly present) and at (−18, 6+4h) for h = 0..6; O-faces at even x ≡ y (mod 4) (wholly present); U-faces at (−21, 6), (−25, 10), (−29, 14); V-faces at (−31, 20), (−27, 24), (−23, 28), (−19, 32). 3. H_X = H_Z = the face-incidence matrix (one row per face, both bases). 4. Verify: k = 1, max weight 8, H_X H_Z^T = 0; distance witnesses of weight 15 on both sides (weight-15 logical strings along lattice paths).
Source: arXiv:2609.21376v1, Appendix D (coordinates and check supports) and Table 7 (resource polynomials). Circuits and data bundle: Zenodo 10.5281/zenodo.22820614.