Qianqian Xie

Wuhan University

Unclaimed · AI-structured profileOpenAlex

Researcher building diagnostic benchmarks for LLM time-series reasoning and coding-agent game-logic evaluation.

papers here

reviewed by Evolve

first-author

corresponding

—

works · OpenAlex

—

h-index · OpenAlex

Bibliometric figures come from OpenAlex (2026-10-13) 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—

Output data is not yet available in this profile, so no score is given. This axis would measure publication volume and cadence — total papers, preprints per year, and authorship roles — none of which can be estimated from the current records.

Impact—

Impact metrics are unavailable, so this axis is unscored. It would measure citations, downstream adoption of the two benchmarks, and influence on subsequent LLM-evaluation work, none of which is present in the provided data.

Focus0/100

The focus score of 0 reflects an exact 50/50 split in the field distribution between "natural language processing / LLM evaluation for time-series reasoning" and "software engineering (LLM coding-agent evaluation)". Measured strictly as field concentration, this is maximal dispersion. Practically, however, both halves share one methodological identity — building diagnostic, reproducible benchmarks for LLM evaluation — so the low score captures breadth across domains more than a lack of coherent direction.

Collaboration62/100

With a collaboration score of 62, the bibliometric record lists 21 distinct coauthors for this researcher. The Evolve catalog's two papers are both middle-authorship and non-corresponding, consistent with team-based work rather than solo-led projects. Recurring coauthor names include Jie Gong, Maowei Jiang, Zhiwei Liu, Yankai Chen, Guojun Xiong, Xue Liu, Min Peng, Sophia Ananiadou, Xinyu Che, Yunfei Ge, and Shihao Li, indicating sustained collaboration across both evaluation lines.

Activity—

Activity data is insufficient, leaving this axis unscored. It would measure recent publication velocity and engagement — outputs per recent year and continuity of preprint releases — which cannot be inferred from the current snapshot.

Reproducibility78/100

Reproducibility is the strongest recorded axis, scoring a mean of 78 across the two reviewed papers. GameLogicBench rates high: it is deterministic, tick-level, and rigorously validated, though single-run, Godot-only, and scaffold-dependent. TimeLitmus rates medium, constrained by rule-based labels and a small number of pairs. Both papers ship as arXiv releases with explicit readiness ratings, supporting a broadly reproducibility-first evaluation style.

Research summary

Qianqian Xie is a researcher at Wuhan University whose catalogued work centers on rigorous evaluation of large language models across two distinct application domains. The Evolve catalog contains two 2026 arXiv papers, both with middle-position authorship. The first, TimeLitmus, is a diagnostic benchmark for cross-modal understanding and explanation faithfulness in event-conditioned time-series prediction; Evolve's review judges it a solid diagnostic benchmark showing that LLMs' event-conditioned accuracy overstates genuine cross-modal skill, with caveats around rule-based labels and a small number of pairs. The second, GameLogicBench, evaluates coding agents on runtime game logic using tick-level state assertions in Godot; it is described as rigorously validated and deterministic, with credible results, though limited to a single run, the Godot engine, and a particular scaffold. Public bibliometric data record 21 distinct coauthors (collaboration score 62), an even 50/50 split between LLM evaluation for time-series reasoning and coding-agent evaluation (focus score 0), and a mean reproducibility score of 78 across the two reviewed papers. Recurring coauthor names include Jie Gong, Maowei Jiang, Zhiwei Liu, Xue Liu, Min Peng, and Sophia Ananiadou. Output, impact, and activity metrics are not yet available and are left unscored.

Collaboration pattern

The record shows 21 distinct coauthors (collaboration score 62). Both catalogued papers are middle-authorship and non-corresponding, pointing to contributor roles within larger teams rather than sole leadership. Recurring names include Jie Gong, Maowei Jiang, Zhiwei Liu, Xue Liu, Min Peng, Sophia Ananiadou, Yankai Chen, Guojun Xiong, Xinyu Che, Yunfei Ge, and Shihao Li, spanning both the NLP/time-series and coding-agent evaluation lines.

Research keywords

event-conditioned time-series predictionLLM evaluationdiagnostic benchmarkcounterfactual evaluationexplanation faithfulnessshortcut learningcross-modal understandingcoding agentsgame developmentGodottick-level assertionsdeterministic evaluation

Highlights

TimeLitmus: diagnostic benchmark for event-conditioned time-series prediction
Evolve's review calls TimeLitmus a solid diagnostic benchmark showing that LLMs' event-conditioned time-series accuracy overstates genuine cross-modal skill; it probes explanation faithfulness and shortcut learning, with caveats around rule-based labels and a small number of test pairs.
GameLogicBench: tick-level, deterministic evaluation of coding agents
The benchmark evaluates coding agents on runtime game logic in Godot via tick-level state assertions; Evolve judges it rigorously validated and deterministic with credible results, though limited to single runs, the Godot engine, and a specific agent scaffold.
A reproducibility-first benchmarking portfolio
Across the two reviewed papers, the mean reproducibility score is 78 on a 0-100 scale: GameLogicBench rates high and TimeLitmus medium, suggesting a consistent emphasis on deterministic, verifiable evaluation design over headline model results.

Timeline

2026TimeLitmus released on arXiv: a diagnostic benchmark for cross-modal understanding and explanation faithfulness in event-conditioned time-series prediction.
2026GameLogicBench released on arXiv: tick-level state assertions for evaluating coding agents on runtime game logic in Godot.
2026Evolve reviews both papers, recording reproducibility readiness of medium (TimeLitmus) and high (GameLogicBench), with a mean reproducibility score of 78 across the two.

Papers in the Evolve catalog

Frequent co-authors

Jie Gong×1 Maowei Jiang×1 Zhiwei Liu×1 Yankai Chen×1 Guojun Xiong×1 Xue Liu×1 Min Peng×1 Sophia Ananiadou×1 Zhiwei Liu×1 Xinyu Che×1 Yunfei Ge×1 Shihao Li×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.

Bibliometric dimensions (collaboration, focus, reproducibility) are computed from public OpenAlex-derived data; citation and venue-level metrics are not included.
Output, impact, and activity axes are unscored because publication counts, citations, and recent-velocity data are missing from the input snapshot.
"Qianqian Xie" is a common Chinese name; this profile is name-resolved from public bibliographic records and may not reflect a single individual.
Both reviewed papers are 2026 arXiv preprints; peer-review status and final venues are not reflected in the data.