Cell: unrestricted x any weight. The verifier derives this from the matrices rather than from a self-declared label: there is no layout, so the locality class is unrestricted, and the max check weight is 10, which is past weight-8, so the only cell this entry joins is unrestricted x any weight.
The opening was structural rather than statistical. Against the 1440 entries the board held when this was staged, this is the only one with n <= 170, k >= 32 and d >= 14 at the same time. The next entries reaching that corner are codes/200-40-14.json (n=200, k=40, d=14, w=15) and codes/254-42-14.json (n=254, k=42, d=14, w=10). At n=170 the board already held codes/170-52-3.json (k=52, d=3), codes/170-34-8.json (k=34, d=8), codes/170-16-10.json (k=16, d=10), codes/170-8-20.json (k=8, d=20) and codes/170-2-13.json (k=2, d=13) — every point on the k-d plane at n=170 except the k >= 32, d >= 14 corner.
That corner could be filled by reconstruction instead of search: arXiv:2608.08996v1 publishes a free-action lifted product over Z_85 with exactly [[170,32,14]], and its Supplementary Information gives both supports.
No parameter sweep — this is a reconstruction, not a discovery. The two supports over Z_85 were taken from the paper, the matrices rebuilt from them, structurally verified, and then put through the same board screen a searched candidate faces: an entry is only worth submitting if it is non-dominated on (n, k, d, w) in a cell it joins. This one came back non-dominated.
The counting is consistent throughout: |A| = |B| = 5, so every check has weight 10, n = 2 * 85 = 170, and the ranks of H_X and H_Z leave k = 32.
verify/qldpc_verify.py recomputes n=170 and k=32 from the ranksof the two matrices (85 X checks and 85 Z checks), max check weight 10, CSS commutation, and the locality and weight classes the track grid uses.
verify/validate_candidate.py returned passed: true on this candidate:structural checks ok, refutation refuted: false (no lighter logical in 8000 RIS trials, seed 1161346672), exact_duplicate_of: null, wl_equivalent_of: null, board_advancing: true, dominated_by: [].
This is a witness-backed upper bound, not a certificate. The paper reports d = 14, which agrees with the witnesses; the submission's own provenance records 20,000 pure-Python RIS trials per side at seed 20260828 finding nothing lighter.
joins. It dominates no entry and is dominated by none. kd^2/n = 36.894 makes it the highest-scoring entry on the board at n <= 170, ahead of codes/136-34-12.json at 36.000; by that same score it ranks 162 of 1440 overall, against a cell headline bar of 1456.855 held by codes/674-170-76-b.json.
entries trade an axis instead of losing one — codes/200-40-14.json has more k and more n, codes/254-42-14.json has more k at more n, codes/170-34-8.json has more k but only d=8 at w=9. The contribution is a new trade-off point in an empty corner, not a takeover.
staged candidates are reported with those entries.
d >= 14 together, so no entry could be dominated by this one. The frontier gains a point rather than losing one, which is what a reconstruction of a published code can offer.
Python 3, the repository's lifted-product construction helpers, the ./qldpc submission builder, and the trusted verification stack (verify/qldpc_verify.py, verify/validate_candidate.py). provenance.model is human: the matrices are a reconstruction of published data and no search produced them. Compute: seconds, no solver time.
Free-action 2BGA (lifted product) over Z_85 with A = {0, 3, 4, 10, 67} and B = {0, 5, 29, 31, 37}, from the Supplementary Information of arXiv:2608.08996v1.
Let C_S be the 85x85 circulant over GF(2) whose first row is the indicator of S, with indices read mod 85. The submitted matrices are
H_X = [ C_{-A} | C_{-B} ] H_Z = [ C_{B} | C_{A} ]
which is what the checked-in file contains: the first 85 columns of H_X carry support (0, 18, 75, 81, 82) = -A, the last 85 carry (0, 48, 54, 56, 80) = -B, and H_Z carries B then A in that order. Each row of checks.X and checks.Z is the 10-element support of one check. Rebuild the two matrices from those supports, then re-run verify/validate_candidate.py before trusting the distance.