Paper: arXiv:2610.06623, "Large Language Model-Guided Discovery of Weight-Five Bivariate Bicycle Codes" (component of presentation PL-78c3e7bf).
Target: the general-stabilizer board's unrestricted x weight-6 (and looser) cells. Same PBB construction family as the companion [[144,12,12]], [[216,4,10]], [[12,2,3]], [[144,4,8]], [[180,2,11]], [[360,4,12]], [[360,12,20]], and [[360,4,24]] submissions in this batch (block-1 z-part [C | D] added to a CSS bivariate-bicycle stabilizer matrix: H = (A B | C D; 0 0 | B^T A^T)).
Source: a weight-5-targeted LLM-guided evolutionary search (OpenEvolve, "pbb-large" campaign) over 4-tuples (A, B, C, D) at (ell=12, m=12), screened by k, isotropy, the combined non-CSS gate (evaluation/noncss_gate.py), and a BP-OSD distance estimate. The surviving presentation PL-78c3e7bf (n=288, k=24) is a direct sum of 6 isomorphic components (evaluation/connectivity.py:stabilizer_components + components_isomorphic, confirmed homogeneous and BLISS-isomorphic); each component is [[48,4,4]], with the parent's exact MILP distance (results/weight5_publication_catalogue.jsonl, `distance_class.status_code: "E"`) transferring to each component.
Unlike the companion [[144,4,8]] and [[180,2,11]] submissions, no standalone explicit presentation of this component already existed in the source repository as its own clean (ell', m') pair. This submission builds one directly: the parent presentation's symplectic matrix was constructed via evaluation/pbb_code.py:build_pbb_code(12, 12, ...), then restricted to one component's qubit support via evaluation/connectivity.py:_induced_presentation_subcode (the generator-presentation-preserving induced restriction, not the row-echelon one) — performed in this submission session, using only the source repository's own, already-published tools.
Parent: (ell, m) = (12, 12), A = 1 + x^3y^6, B = 1 + x^2y^6 + x^4, perturbation C = 0, D = 1 + x^4 (0-indexed exponent pairs: A=[[0,0],[3,6]], B=[[0,0],[2,6],[4,0]], C=[], D=[[0,0],[4,0]]), n=288, k=24.
(distance_class.status: "exact", status_code: "E"), transferred to each of the 6 isomorphic [[48,4,4]] components by the standard component-isomorphism argument.
matrix (built this session, see above) was run through this repo's own cli/qldpc.py submit --dry-run (RIS search, 20,000 trials, gf2_fast accelerator at 1,000,000 trials). Independently found d<=4, matching the source's exact claim.
Final claim here: witness-backed upper_bound; the source repository's own evidence tier is exact.
Not applicable — this submission mined an existing, already-vetted and now-published campaign result (plus a direct component extraction performed this session) rather than running a new search.
Model: jointly attributed to Claude Opus 5 and GPT-5.6 Sol (both aliases recovered against the parent's lineage log in the source repository's historical_attribution.jsonl), matches provenance.model. The component extraction and packaging in *this* repository was done separately, using only this repository's own and the source repository's published tooling.
from evaluation.pbb_code import build_pbb_code
from evaluation.connectivity import stabilizer_components, _induced_presentation_subcode
parent = build_pbb_code(ell=12, m=12,
A_terms=[(0,0),(3,6)],
B_terms=[(0,0),(2,6),(4,0)],
C_terms=[],
D_terms=[(0,0),(4,0)])
comp = stabilizer_components(parent)[0]
sub = _induced_presentation_subcode(parent, comp)
# sub.matrix is the (48, 96) symplectic stabilizer matrix [X-part | Z-part]
Source: github.com/qiskit-community/qcode-discovery @ d6abdbd, evaluation/pbb_code.py, evaluation/connectivity.py, results/weight5_publication_catalogue.jsonl.