Companion to the d = 5 dense-packed submission [[101,5,5]] (see notes/101-5-5.md). Same construction family (arXiv:2511.06758) at distance 7. The paper evaluates dense packing at d = 5, 7, 9, 11 and finds the hook-avoiding dense patch matches or beats standalone at larger distance and lower physical error rate. This is the d = 7 member: five logical qubits, weight-4 checks, single-layer honest layout, nearest-neighbour tilted lattice (r = √2).
Reconstruction identical to notes/101-5-5.md, with distance = 7 in research/build_dense_surface.py (the five_dense_num mask scales with d). n = 197 data qubits, k = 5, weight 4.
GF(2) sanity: CSS commutation exact, rank(H_X) = rank(H_Z) = 96, so k = 197 − 96 − 96 = 5.
d ≤ 7 (X and Z).
genuine [[197,5,7]] witness-backed upper bound.
d = 7 claim would need the repo's MILP certificate (verify/certify.py) or a deeper RIS refutation. Confidence: upper_bound.
local-2d-single / weight-4 geometric efficiency: g ≈ 1.24, matching thed = 5 member and above the previous cell best ([[672,85,3]], g ≈ 1.14).
Pareto-incomparable (d 7 vs 5, n 197 vs 101), so both sit on the frontier.
Same k-counting / ancilla trap as the d = 5 instance (see notes/101-5-5.md); resolved by the verifier's data-qubit-only n rule. No density collapse at larger d: the shared-region structure persists and the n-savings vs standalone (5 × 49 = 245 data qubits) hold.
research/build_dense_surface.py (this repo) with argument 7; numpy GF(2) rref; ./qldpc submit (RIS witness search, schemebook, verifier, locality).
python research/build_dense_surface.py 7 # writes /tmp/dense_7.npz (hx, hz, coords) ./qldpc submit /tmp/dense_7.npz --coords /tmp/dense_7.npz --layers 1 \ --authors @mathysrennela --family topological