n
92
Reproduce the explicit reverse-fold example in Appendix D.2/D.4 of Jong Yeon Lee, Koki Okada, Nishad Maskara, Kenta Kasai and Hengyun Zhou, arXiv:2609.30069v1. The construction and parameters are published work; metadata records known_parameters. The parent is the published [[184,50,10]] code already represented by codes/184-50-10.json on this board.
This targets the weight-8, unrestricted stabilizer Pareto frontier. At main commit 0ad745c011ce7393c921a39cf35fe760090222c4, its witnessed score 25*8²/92 = 17.391304 would place second among merged weight-8 stabilizer entries, after [[144,10,16]] at 17.777778. It occupies a smaller-block, higher-rate tradeoff. This is a literature reproduction, not a claim of new parameters. Scores and rankings depend on the retained distance upper bound.
This was a targeted reconstruction, not a random code-family sweep. The paper prints the two 3-by-8 parent exponent arrays. We enumerated the 105 perfect matchings of eight block columns and all six block-row permutations to solve Proposition 4 with reversing sign eta=-1. With the free row offset alpha_0 fixed to zero, exactly one matching/permutation solution remained. Its explicit arrays and fold map are given below; no unpublished paper data is needed for this instance.
The trusted stabilizer submission builder used 2,000 direct Pauli RIS trials, seed 31010092, and returned a weight-8 logical. A separate direct Pauli run used 20,000 trials, seed 31010192, pair depth 20, and returned weight 8 in 49.65 seconds. A native doubled-code run used 200,000 trials per CSS side, seed 31010292, pair depth 20 and four threads; its mapped logical had Pauli weight 10, so it did not improve the bound. A separate exact three-bit Pauli-to-CSS embedding search used 200,000 native trials per CSS side, seed 31010392, pair depth 20 and four threads, and returned an X-side weight-8 logical mapping to source Pauli weight 8. Every returned logical was checked with the unchanged trusted Pauli predicate and retained.
The submitted generator list has 69 rows, stabilizer rank 67, n=92, k=25, and maximum Pauli check weight 8. The unchanged verifier accepts its structure and the weight-8 Pauli witness embedded in the JSON. Fingerprint: c8472a1f1e2e36dd. The independent direct and three-bit searches agree with the paper's reported distance, but they establish no exact certificate. The submitted distance remains upper_bound, d<=8.
For the three-bit audit, write the source checks as S=(A|B). We used HX=[A,0,B; I,I,I] and HZ=[B,A+B,A]. An X-side binary witness (u,v,w) maps to source Pauli (u+v,v+w), with additions over GF(2). Both the embedded logical and its mapped source logical were validated. The exactness of this mapping is not an exact-distance certificate: the audit itself used heuristic RIS.
The independent direct-search source witness has X support [8,14,47,80,81] and Z support [8,9,54,59,81]. The two overlaps count once, giving Pauli weight 8. The candidate JSON retains the builder's independently found witness, under the same explicit qubit ordering. The three-bit audit returned X support [3,14,32,40,83] and Z support [10,14,20,32,91], also weight 8. The doubled audit's weight-10 witness has X support [22,30,76,85] and Z support [4,15,54,73,84,86].
The unchanged full candidate gate passed against main commit 0ad745c011ce7393c921a39cf35fe760090222c4. It found no exact or WL duplicate, no dominator and no distance-only gain, and labelled the candidate board-advancing. Fresh refutation seed: 1412906472. The receipt's 6,180-trial figure is a configured ceiling under the default 10-second cap, not an independently measured completed count. Fingerprint: c8472a1f1e2e36dd. The gate was run before board promotion.
The paper's larger [[200,43,20]] example is a stronger numerical target, but its required instance data were not released in the linked repository when checked on 2026-10-01. That unavailable recipe was not reconstructed or claimed here. The native doubled-code search above returned weight 10 because its Hamming objective can differ from Pauli weight; it was not used to inflate this entry's distance. No geometric layout or circuit performance is claimed.
Author: @mrvee-qC-bee. Model: GPT-6 Astra in Codex. Reconstruction used NumPy; all logical searches and validation used the challenge's unchanged trusted Pauli RIS, native RIS and GF(2) routines. The search confirmation itself took about a minute; the full board comparison adds several minutes. No trusted verifier, schema or workflow was modified.
Set P=23, with all exponent arithmetic modulo P:
E = [[0,0,0,0,0,0,0,0],
[0,12,8,21,6,1,19,15],
[0,9,18,11,7,17,10,4]]
D = [[0,15,7,22,7,0,22,15],
[0,1,2,4,0,1,2,4],
[0,5,17,6,6,17,5,0]]
sigma = [5,7,4,6,2,0,3,1]
rho = [0,1,2]
alpha = [0,1,17]
beta = [0,15,7,22,7,0,22,15]
Use parent qubit index q=23*ell+t, for ell=0,...,7 and t=0,...,22. For j=0,...,2 and s=0,...,22, in that order, the parent X-check row 23*j+s has support {23*ell+(s-E[j,ell] mod 23): ell=0,...,7}. Replace E by D for the parent Z checks. This is a stated reindexing of the paper's interleaved parent convention.
The identities D[j,ell]=-E[rho[j],sigma[ell]]+alpha[j]+beta[ell] and beta[sigma[ell]]=beta[ell] hold modulo 23. Thus pi(ell,t)=(sigma[ell],-t-beta[ell] mod 23) is a fixed-point-free involution exchanging the CSS check spaces, as required by Proposition 4.
Enumerate the 92 pairs (q,pi(q)) with q<pi(q), in increasing q. For each parent X-check row, put an X on folded qubit i when that pair's first coordinate is in the row, and a Z when its second coordinate is in the row. An overlap is Y. Preserve parent row order and sort the two support lists within each row. These instructions recover the exact ordered submitted checks. The folding identity implies A B^T+B A^T=0; the rank is 67.
The JSON and this note are the only files required from this submission. In the challenge environment, independently recheck its structure and witness and run a fresh bounded refutation with:
uv run --frozen python verify/qldpc_verify.py codes/92-25-8.json