UCLA researcher building evaluation benchmarks and meta-level schedulers for recursively self-improving LLM agents.
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
8
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-10-16) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not scored: the underlying output data — venue-resolved publication records sufficient to normalize across venues and years — is not yet available for this profile. This axis would measure total research output volume and its distribution over time.
Impact—
Not scored: beyond a preliminary h-index of 8 and i10-index of 8, citation uptake data is insufficient to benchmark influence within the field. This axis would measure citations, venue visibility, and downstream reuse of the work.
Focus0/100
The focus score of 0 reflects a highly dispersed record. The field distribution splits evenly — 0.5 for software engineering (LLM agent evaluation) and 0.5 for machine learning (recursive self-improvement of LLM systems) — and OpenAlex topics span five areas: privacy-preserving technologies in data (24), domain adaptation and few-shot learning (23), topic modeling (21), adversarial robustness in machine learning (16), and natural language processing techniques (15).
Collaboration74/100
A collaboration score of 74 rests on 40 distinct coauthors across 74 works in the last three years. Recurring names include Yangxu Liao, Jiajun Wu, Leixin Sun, Zihan Tan, Yitao Liu, Shuo Li, Jiaru Qian, Shanghaoran Quan, Chuangxin Zhao, Yang Liu, and Bin Chong. Both reviewed 2026 papers place Wan last among the authors, though not as corresponding author.
Activity100/100
Activity is scored at the maximum of 100. OpenAlex counts by year show 1 work in 2023, 11 in 2024, 31 in 2025, and 32 in 2026 — 74 works over the most recent three years. The trajectory indicates a fast-accelerating publication cadence concentrated in the last two years of the record.
Reproducibility55/100
The reproducibility mean is 55 across 2 assessed papers. Both catalog entries carry a medium reproducibility readiness: SWE-Prometheus applies rigorous controls (no-op, template, and gate ablation) but its single-rollout LLM-judged scoring caps precision, while MetaRSI's validation stays entirely within the closed-format slice it critiques — careful engineering with bounded evaluation coverage.
Research summary
Guancheng Wan is a researcher at the University of California, Los Angeles whose profile is name-resolved from OpenAlex bibliographic data. The record shows a steep output ramp: 1 work in 2023, 11 in 2024, 31 in 2025, and 32 in 2026, totaling 74 works over the last three years and earning a maximum activity score of 100. Collaboration is broad — 40 distinct coauthors in that window — with a recurring group including Yangxu Liao, Jiajun Wu, Leixin Sun, Zihan Tan, Yitao Liu, and Shuo Li. Wan's field distribution splits evenly (0.5/0.5) between software engineering, centered on LLM agent evaluation, and machine learning, centered on recursive self-improvement of LLM systems. Evolve has reviewed two 2026 arXiv papers, both listing Wan in the last author position: SWE-Prometheus, a benchmark measuring whether coding agents improve engineering governance in real repositories, and MetaRSI/RSI2, a meta-level scheduler that orchestrates data, harness, and weight self-improvement under one budget. Both carry a medium reproducibility readiness, and the reproducibility mean is 55 across two assessed papers. Broader topic signals — privacy-preserving technologies, domain adaptation and few-shot learning, topic modeling, adversarial robustness, and NLP — suggest a wider bibliographic footprint than the agent-focused catalog alone, with an h-index of 8.
Collaboration pattern
Wan's collaboration score of 74 reflects 40 distinct coauthors across 74 works in the last three years. A recurring core spans Yangxu Liao, Jiajun Wu, Leixin Sun, Zihan Tan, Yitao Liu, Shuo Li, Jiaru Qian, Shanghaoran Quan, Chuangxin Zhao, Yang Liu, and Bin Chong. On both reviewed 2026 papers Wan occupies the last author position without being corresponding author, consistent with a senior-slot role inside a large, fast-moving team.
Research keywords
Highlights
SWE-Prometheus: a benchmark for whether LLM agents improve real-repo governance
MetaRSI: a meta-level scheduler for recursive self-improving systems
An evaluation-first methodology shared across both catalog papers
Timeline
- 20231 work recorded in OpenAlex — the earliest activity in the available record.
- 2024Annual output rises to 11 works as the publication record begins to build.
- 2025Output accelerates to 31 works in a single year, with the OpenAlex topic footprint spanning privacy, domain adaptation, topic modeling, robustness, and NLP.
- 202632 works recorded; both reviewed papers appear on arXiv — SWE-Prometheus and MetaRSI/RSI2 — with Wan listed as last author on each.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- All bibliometrics are name-resolved from OpenAlex; the topic profile spans areas (privacy-preserving technologies, topic modeling, NLP) beyond the two reviewed LLM-agent papers, so some coverage noise is possible.
- The output and impact dimensions are unscored because the underlying data is insufficient.
- The reproducibility score of 55 is based on only 2 assessed papers, both 2026 arXiv preprints.
- Per-paper citation counts, venue acceptance records, and funding information are not part of the provided data.