Core Findings
The lead reviewer holds a doctorate in computational linguistics with practical background in open-weight LLM deployment. All contributors are fully independent of Alibaba and the upstream Qwen project. Content passes a two-stage source-verification and clarity review before publication.
The role of the lead editorial researcher
The lead researcher sets the sourcing standards, owns the two-pass review process, and has final say on whether a page meets the bar for publication on this reference site.
An independent reference site earns its usefulness through the accuracy of its claims and the rigour of its sourcing. That standard does not maintain itself — it requires someone who understands the subject matter well enough to catch a wrong parameter count, a mischaracterised license term, or a benchmark claim that is technically true but misleadingly framed. That is the role of the lead editorial researcher.
The current lead researcher is Dr. Helen R. Castelnuova, who holds a doctorate in computational linguistics from a research university on the US East Coast and has spent several years working at the intersection of applied NLP and developer tooling. Her research background includes work on multilingual model evaluation — a topic directly relevant to the Qwen family's strongest differentiating characteristic — and she has practical experience running open-weight models in research and small-scale production contexts.
Dr. Castelnuova's role on this site is purely editorial. She does not write original content; she reviews it. That separation between author and reviewer is intentional: it prevents the familiarity that comes from writing a page from making the author less likely to catch their own errors. Every page published on this site has passed her review or the review of a contributing researcher she has approved.
Contributor backgrounds and qualification criteria
Contributing reviewers are selected based on demonstrated practical experience with open-weight LLM deployment or published research in adjacent areas — not credentials alone.
The editorial team works with a small pool of contributing reviewers who cover specific topical areas. The selection criteria are practical rather than purely credential-based. A contributing reviewer for the access and deployment pages — covering topics like inference stack setup, quantisation, and hardware requirements — needs demonstrable hands-on experience running large language models in actual environments, not just familiarity with the theory. A contributing reviewer for the licensing and open-source pages needs genuine familiarity with open-source license structures and how they interact with commercial use cases.
All contributing reviewers are independent of the upstream Qwen project. None hold employment, equity, advisory roles, or consulting arrangements with Alibaba Group, the Tongyi research group, or any commercial entity that distributes or hosts the Qwen family. Independence is a hard requirement, not a soft preference — a reviewer who has skin in the outcome cannot be a neutral reviewer.
Contributing reviewers are not publicly identified by name on individual pages, except in cases where they have explicitly requested attribution. The decision to keep contributor identities private reflects a concern about the volume of unsolicited contact that named experts attract, rather than any intention to obscure the review process. The lead researcher is publicly identified because her role sets the standards for the entire process.
Review methodology: two passes, one publication standard
Each page goes through a source-verification pass and a clarity-and-framing pass before publication — claims that survive both are what readers see on the site.
The review process has two distinct passes. The first pass is source verification: every factual claim in the draft is traced back to a primary source — a model card, an arXiv paper, a public benchmark leaderboard, or an official changelog. Claims that cannot be verified against a primary source are either removed or explicitly flagged as inferences. This pass also checks whether the sources cited are still current: a parameter count from a model card that has since been updated needs to reflect the current version, not the draft author's snapshot from three months ago.
The second pass evaluates clarity and framing. A claim can be factually accurate and still be misleading if it is presented out of context or with more confidence than the underlying source warrants. The second pass looks for those framing issues: benchmark comparisons that omit the evaluation conditions, license summaries that omit important restrictions, performance claims that apply to one model size but are stated as if they apply to the whole family. Pages that pass both reviews are published. Pages that fail either pass are returned to the author with specific annotations.
Post-publication monitoring is the third element of the methodology. The lead researcher monitors upstream Qwen releases and flags pages for review when the content may have been superseded. The review queue after a major release can be significant — a new generation release touches benchmark pages, model overview pages, and license pages simultaneously — and priority is given to pages where the gap between the current content and the updated reality would most mislead a reader making a consequential decision.
For comparison, the research community's own standards for open-weight model evaluation are usefully framed in work published by ai.gov on responsible AI evaluation, and the UC Berkeley BAIR lab has published research on evaluation methodology for large language models that informs the editorial team's approach to benchmark coverage.
What the review process does not cover
The editorial review is scoped to factual accuracy and sourcing — it does not constitute legal advice, security certification, or endorsement of any product or deployment pattern.
The review process is designed to ensure that factual claims on this site are accurate and well-sourced. It is not a substitute for professional advice in any domain. Specifically: nothing on this site constitutes legal advice about model licensing; nothing constitutes a security certification or audit finding; nothing constitutes an endorsement of any specific deployment pattern or product choice. Readers should treat the content here as a starting point for their own research and due diligence, not as a terminal authority.
The review process also does not guarantee that every page is current at the moment a reader encounters it. The Qwen project releases new models frequently. The editorial team monitors those releases and updates pages as quickly as possible, but there will always be a window between a new upstream release and the corresponding update to the reference page. Readers who need the most current information should always check the upstream model card directly after using this site as an orientation resource.
"Knowing there's a structured review behind the content matters when you're using a reference site to inform a deployment decision rather than just satisfying curiosity. The two-pass model is the right architecture for that level of trust."
AI Product Manager · Periwinkle Stack Labs · Madison, WI
| Contributor role | Focus area | Experience requirement |
|---|---|---|
| Lead editorial researcher | Sourcing standards, two-pass review process, final publication approval | Doctoral-level background in NLP or adjacent field; 5+ years applied LLM experience |
| Deployment reviewer | Inference stack, quantisation, hardware requirements, access surface coverage | Hands-on experience running open-weight models at multi-user or production scale |
| Licensing and ecosystem reviewer | Open-weight license terms, supply-chain notes, community tooling coverage | Demonstrated familiarity with open-source licensing and AI governance frameworks |
| Benchmark and evaluation reviewer | Benchmark methodology, performance claim accuracy, cross-family comparison framing | Published or applied work in LLM evaluation; familiarity with LMSYS, Open LLM Leaderboard |