Researcher bridging LLM post-training and agent systems engineering; two 2026 arXiv papers, compact four-person team
2
papers here
2
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
—
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 yet: the snapshot contains no verifiable publication counts, venue record, or output-volume metrics beyond the two cataloged arXiv preprints. This axis would measure total and per-year research output once a fuller bibliographic record is linked.
Impact—
Not scored yet: citation counts, h-index-style influence, and downstream adoption are unavailable in this snapshot, and both cataloged papers are recent 2026 preprints. This axis would measure scholarly uptake and real-world influence of the work.
Focus0/100
Focus scores 0 because the field distribution is an even 50/50 split between 'machine learning — LLM post-training and reinforcement learning' and 'AI systems engineering (autonomous agent orchestration)', one paper each in the two-paper catalog. This reads as a deliberate dual track rather than scattered work: one line builds certified post-training for coding agents, the other re-engineers agent pipeline execution. More papers would show whether the split consolidates.
Collaboration32/100
Collaboration scores 32 on the 0-100 scale, based on 4 distinct coauthors across the two cataloged papers: Cheng Li, Jiexiong Liu, Chi Hong, and Ziheng Huang. That is a small, stable working group, consistent with the middle-author positions on both papers. The score indicates modest collaboration breadth rather than absence of teamwork; broader external networks would raise it.
Activity—
Not scored yet: publication cadence cannot be established from a snapshot holding only two 2026 arXiv preprints. This axis would measure publishing rate, consistency, and project velocity across years once a complete dated record is available.
Reproducibility20/100
Reproducibility scores 20, the mean readiness across 2 assessed papers, both rated low. 'Incremental Consistency Execution' reports n=1 case studies with no code release and misaligned citations; 'Train What You Deploy' grounds its +3.0 headline gain in single runs below the noise floor. The pattern is conceptually careful but empirically thin — released artifacts and multi-seed results would move this score quickly.
Research summary
Yixuan Chen is a researcher whose cataloged work spans two complementary fronts of modern AI engineering. The first is machine learning — LLM post-training and reinforcement learning: the 2026 preprint 'Train What You Deploy:Token-Faithful Post-Training of a Production Coding…' proposes C-DPPO, a certification-theoretic trust-region method built on total variation divergence, designed to keep training token-faithful to the deployed coding agent; Evolve's review credits the theory as sound while flagging that the headline +3.0 gain rests on single runs below the noise floor. The second front is AI systems engineering for autonomous agent orchestration: 'Incremental Consistency Execution for Autonomous Intelligent Systems' introduces perturbation-built invariant domains, task fact contracts, and field-level dependency masks to enable incremental re-execution of agent pipelines; the review calls the design promising but notes n=1 case studies, no code release, and misaligned citations. Across both papers Chen appears in a middle, non-corresponding author position, working with a stable four-person co-author group (Cheng Li, Jiexiong Liu, Chi Hong, Ziheng Huang). The radar shows an even 50/50 field split (focus score 0), 4 distinct coauthors (collaboration score 32), and a reproducibility mean of 20 across 2 assessed papers, both rated low. This is an early-stage profile: no institution, ORCID, or citation record is available, and output, impact, and activity remain unscored.
Collaboration pattern
The cataloged work is produced with a compact recurring team: Cheng Li, Jiexiong Liu, Chi Hong, and Ziheng Huang — the same 4 distinct coauthors behind both 2026 papers. All cataloged positions are middle-author and non-corresponding, suggesting a contributor role inside a stable group rather than an independently-led line. No cross-institution or long-range collaborations are visible in this snapshot.
Research keywords
Highlights
Certified trust-region post-training (C-DPPO) for production coding agents
Perturbation-built invariant domains for incremental agent re-execution
A shared guarantee-first, evidence-thin pattern across both lines
Both cataloged papers lead with formal structure — trust-region certification on one side, perturbation-built invariant domains on the other — while empirical validation stays minimal: single runs below the noise floor and n=1 case studies, with neither paper releasing code (mean readiness 20 across 2 papers).
Timeline
- 2026arXiv preprint 'Incremental Consistency Execution for Autonomous Intelligent Systems' — middle author, non-corresponding; autonomous agent orchestration work reviewed by Evolve with reproducibility rated low.
- 2026arXiv preprint 'Train What You Deploy:Token-Faithful Post-Training of a Production Coding…' — middle author, non-corresponding; C-DPPO certification theory for coding-agent post-training, reproducibility rated low.
- 2026Both cataloged outputs land in the same year, splitting the field distribution 50/50 between LLM post-training/RL and agent systems engineering, with 4 distinct coauthors in total.
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.
- Radar scores are computed from public bibliometric data limited to the two cataloged 2026 arXiv preprints; no wider publication-database record is reflected in this profile.
- The name 'Yixuan Chen' is common; this profile is name-resolved from public bibliographic data and lacks ORCID and institution verification.
- Output, impact, and activity axes are unscored because citation and publication-count data are unavailable in this snapshot.
- Both cataloged papers are middle-author, non-corresponding arXiv preprints; peer-reviewed venue status is unknown.