Haonan Huang

—

Unclaimed · AI-structured profileOpenAlex

AI researcher building evaluation-first benchmarks — LLM coding agents tested in games, video object removal judged by human perception

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-09-21) 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—

No output score is computed yet: the record lacks publication counts and venue distribution, covering only the two cataloged 2026 arXiv preprints. This axis would measure how much and how often this researcher publishes across venues.

Impact—

Impact cannot be scored yet: no citation counts, h-index, or follow-up influence data are available, and both cataloged papers are 2026 preprints too recent to accumulate citations. This axis would measure scholarly uptake and downstream influence.

Focus100/100

This axis measures how concentrated a researcher's output is across fields, and here it scores a perfect 100: the raw field distribution contains exactly one field — artificial intelligence (LLM agents, game-playing evaluation) — holding the full field mass. Both 2026 preprints are evaluation-centric, a game-playing agent study and a video object removal benchmark, reinforcing a tightly focused benchmark-and-evaluation agenda.

Collaboration54/100

With 14 distinct coauthors recorded and a collaboration score of 54 (mid-range), Huang works in multi-author teams rather than solo or in very large consortia. Twelve recurring collaborators appear in the data, including Joey Xiao, Tianrui Qiu, Xianghao Zang, Yinan Du, Chi Zhang, and Hao Sun. The observed pattern pairs first-author benchmark building (VOR-Bench) with last-author seniorship on the multi-author agent study (Compiled Agency).

Activity—

Activity is not scored yet: only two 2026 preprints are known, with no multi-year publication cadence on record, so recent-output trends cannot be computed. This axis would measure publishing pace and continuity of output over time.

Reproducibility55/100

Reproducibility scores 55, the mean over just one rated paper: Compiled Agency, whose artifact readiness Evolve rates medium. The Evolve verdict flags that the StarCraft II evidence is strong but session counts are tiny and the Civilization examinations were incomplete at the paper's freeze, limiting replication scope — consistent with a moderate rather than high score.

Research summary

Haonan Huang is an early-stage artificial intelligence researcher whose cataloged work centers on evaluation and benchmarking, defined by two 2026 arXiv preprints. As last author of 'Compiled Agency', Huang contributed to a study claiming that frontier general-purpose coding agents can compile winning game players from bare interaction across Flappy Bird, StarCraft II, and Civilization; Evolve's review finds the StarCraft II evidence strong but notes tiny session counts and incomplete Civilization examinations at the paper's freeze, rating reproducibility readiness medium. As first author, Huang leads VOR-Bench, a human perception-driven benchmark for video object removal, which has not yet received an Evolve verdict. The bibliometric fingerprint is unusually concentrated: the raw field distribution holds a single field — artificial intelligence (LLM agents, game-playing evaluation) — yielding a perfect focus score of 100. Collaboration sits mid-range at 54, with 14 distinct coauthors and twelve recurring collaborators, and Huang alternates between first-author benchmark construction and last-author seniorship. Reproducibility averages 55 across the one rated paper. Citation, output-volume, and multi-year activity data are absent, so those dimensions stay unscored; with no confirmed institution, country, or ORCID, this profile is name-resolved from public bibliographic data.

Collaboration pattern

Huang's recorded network spans 14 distinct coauthors, with twelve recurring collaborators — including Joey Xiao, Tianrui Qiu, Xianghao Zang, Yinan Du, Zhixiang He, Chi Zhang, Hao Sun, and Kongming Liang — yielding a mid-range collaboration score of 54. The observed pattern is team-based: first-author benchmark construction on VOR-Bench, and last-author seniorship steering the multi-author Compiled Agency study. No solo-authored work appears in the catalog, and no institutional affiliation is on record to contextualize the network.

Research keywords

coding agentsLLM agentsgame playingbenchmark evaluationprogrammatic policy synthesisStarCraft IIvideo object removalhuman perceptionvideo editing evaluationagent evaluation

Highlights

Coding agents that compile winning game players from bare interaction
As last author, Huang's team reports that frontier general-purpose coding agents can build winning players for Flappy Bird, StarCraft II, and Civilization from a single bare session. Evolve's verdict: the StarCraft II evidence is strong, but session counts are tiny and the Civilization exams were incomplete at freeze; reproducibility readiness is rated medium.
VOR-Bench: benchmarking video object removal through human perception
As first author, Huang introduces a benchmark that evaluates video object removal against human perception judgments. The paper is cataloged from arXiv (2026) but has not yet been reviewed by Evolve — no verdict, keywords, or reproducibility rating exist for it yet.

Timeline

2026Serves as last author on 'Compiled Agency', which claims frontier general-purpose coding agents build winning game players from bare interaction across Flappy Bird, StarCraft II, and Civilization; Evolve rates its reproducibility readiness medium.
2026Publishes VOR-Bench as first author on arXiv — a human perception-driven benchmark for video object removal, not yet reviewed by Evolve.
2026Both cataloged works land in the same year, establishing an evaluation-first agenda that spans LLM agent benchmarks and video editing benchmarks.

Papers in the Evolve catalog

Frequent co-authors

Joey Xiao×1 Tianrui Qiu×1 Xianghao Zang×1 Yinan Du×1 Zhixiang He×1 Chi Zhang×1 Hao Sun×1 Zhongjiang He×1 Tianwei Cao×1 Xuchong Zhang×1 Hongbin Sun×1 Kongming Liang×1

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.

Bibliometrics are drawn from public bibliographic data only; no institution, country, or ORCID is on record, and 'Haonan Huang' is a common name, so this profile is name-resolved and may merge or split namesakes.
The evidence base is only two 2026 arXiv preprints; citation counts, venue history, and h-index are unavailable, so output and impact remain unscored.
VOR-Bench has not been reviewed by Evolve: no verdict, keywords, or reproducibility rating exist for it yet.
The reproducibility score (55) averages a single rated paper (reproPapers = 1) and should be read as provisional.