Data for "Multi-agent discovery of finite-length quantum LDPC codes"

This dataset accompanies the paper "Multi-agent discovery of finite-length quantum LDPC codes" by Dongheng Qian and Tianyi Li. It contains the code instances, distance evidence, literature benchmark and numerical data behind the reported results. - 01_code_instances: parity-check matrices (NumPy .npz and Matrix Market .mtx), parameters and constructions of the 36 reported CSS codes (the 7 codes of Table 1 and the 29 codes of Extended Data Table 1). - 02_distance_evidence: evidence for every reported distance. 21 exact distances are certified by mixed-integer linear programming (8 with all subproblems solved, 13 with subproblems reduced by code symmetries), with solver logs and records, logical bases and symmetry certificates; 15 upper bounds are given by explicit logical operators. - 03_literature_benchmark: the literature benchmark dataset of 1,354 code records from 93 references, in two linked CSV files. - 04_figure_source_data: numerical source data for Figs. 2 and 3 and Extended Data Figs. 1 and 2, including the per-run and per-candidate results of the three paired runs of the controlled search comparison. - 05_decoding_data: code-capacity Monte Carlo counts for 13 codes, and circuit-level counts for five codes (three discovered codes and two literature references) together with the 70 Stim circuits, syndrome-extraction schedules, logical bases and simulation settings used to obtain them. README.md gives an overview, and DATA_DICTIONARY.md defines every file and variable and links it to the tables and figures of the paper.