n
46
codes/50-19-3.json is @vprusso's construction; this is a reduction of it, following the precedent @mathysrennela set on this board of submitting a reduction of someone else's entry as a contribution in its own right. 4 qubits come out under the general form of the move below, and the accepted grafts were .
Let S be any element of the row space of H_X — an element of the stabilizer group, not necessarily one of the published generators — with support T, and let a ∈ T.
1. Replace one generator of the subset summing to S by S itself. The span is unchanged: that generator equals S plus the rest of the subset. 2. Add S into every other X row meeting a, so a survives only in S. 3. Apply the CNOT fan-out from a to T \ {a}, which acts on the check matrices as H_X[:, b] ^= H_X[:, a] and H_Z[:, a] ^= H_Z[:, b].
After step 2 only S meets a, so step 3 turns S into the weight-1 stabilizer X_a and leaves every other X row alone; and column a of H_Z becomes the XOR of its entries over T, which commutation forces to be 0. Qubit a is disentangled and is deleted: n → n-1, k unchanged, and the Z rows only lose an index, so their weights and support diameters cannot rise.
The weight of S selects the regime, entirely through the clearing step R ^= S:
|S| = 1 — nothing is added anywhere, so weight, radius and distance are allpreserved exactly and no distance search is needed. This is the row-space form of the weight-1 stabilizer cleanup of Liang, Eberhardt and Chen (arXiv:2504.08887 Sec. III D step 4); research/local2d/boundary_engine.py fires only on a *literal* weight-1 generator ROW, and row-space membership is strictly weaker.
|S| = 2 — the row loses a and toggles one other index, so its weight changes by 0or −2 and can never rise.
|S| ≥ 3 — the row can grow, so every touched row is checked against the code's weightclass and its locality radius, measured in the source's own layout.
Low-weight row-space elements are enumerated as sums of at most three checks with a connected overlap pattern.
Fresh-seed RIS ladder: 3 at 20k → 200k → 1M → 5M → 20M trials per rung, independent seeds, both sides searched jointly. Nothing lighter than 3 appeared. Each graft was also screened and confirmed twice in the loop with independent seeds; those rungs are a filter, not the evidence. The distance is a witness-backed upper bound — on this project a [[682,10,38]] candidate survived three fresh seeds to five million trials and then fell at twenty million, so a flat ladder is not a proof.
verify/validate_candidate.py passes: board_advancing: true, dominated_by: [], no duplicate and no WL-equivalent entry, cell weight-8 × unrestricted.
Max check weight 8 throughout. Interaction radius unchanged at 0, so the reduced code is no less local than the source; layers unchanged. Qubit positions are the source's — every surviving qubit keeps the coordinate it has in codes/50-19-3.json, and the reduction only deletes. kd²/n 3.420 → 3.717.
It dominates 50-19-3 on (n, k, d, w).