Xunkai Li

Output—

Output data is insufficient: this axis measures publication volume, venue throughput and productivity trends over time. Only three 2026 arXiv preprints are cataloged for this researcher, so no defensible score can be computed and none is estimated.

Impact—

Impact data is insufficient: this axis would measure citation counts, citation velocity and influence on downstream work. The provided bibliometric snapshot contains no citation metrics for this researcher, so the axis is left unscored rather than guessed.

Focus0/100

The focus score of 0 reflects maximal dispersion in the raw field distribution: three fields carry equal weight (0.333 each) — LLM-based multi-agent systems, LLM-based multi-agent systems evaluation, and graph representation learning — and each cataloged 2026 paper sits in a different one. The breadth spans method-building (MACE, BRAIN) and benchmark design (OpenMAS-GCom), which reads as deliberate cross-area work rather than a single deep vein, though a longer record is needed to confirm the pattern.

Collaboration60/100

The collaboration score of 60 rests on 19 distinct co-authors in the raw record. Recurring names — Zekai Chen, Rong-Hua Li and Kairui Yang appear more than once in the frequent-coauthor list — point to a stable core group at Beijing Institute of Technology. On all three 2026 papers this researcher holds a middle author position and is never the corresponding author, consistent with contributing inside a team. A second 'Xunkai Li' entry flags a possible same-name collaborator.

Activity—

Activity data is insufficient: this axis would measure publication cadence, recent-year output and research continuity. All three cataloged papers fall in 2026, which is too short a window to establish a trajectory, so no score is assigned.

Reproducibility55/100

Reproducibility scores 55, the mean readiness across 3 reviewed papers, each rated medium. Evolve's verdicts flag reproducibility-adjacent risks: MACE uses single-run evaluations with sub-0.1pp margins and uncited baselines; OpenMAS-GCom's intervention evidence covers only five of 29 datasets; BRAIN shows narrow margins and untested cross-domain transfer. The uniform medium rating suggests artifacts and benchmarks are released, but reported gains are thin enough that independent replication remains an open question.