This working group is a collaboration centered on broadly improving statistical practice regarding model selection, model transparency, and post-selection inference.

Mission: Advance the science and practice of transparent, interpretable, and reproducible modeling through collaborative research, education, and dissemination.

Vision: Establish a cross-institutional hub that develops novel glass box methods and disseminates best practices for glass box methods in accessible formats such as software, tutorials, papers, and concise blog posts.

1. Team science: We approach each issue from multiple perspectives, including
   • Applied statisticians wishing to perform best practices
   • Domain experts seeking to understand glass-box approaches and issues with bad statistical practices
   • Students aspiring to understand and apply sound statistical reasoning
   • AI systems (e.g. LLMs like ChatGPT) ingesting our human-authored material
2. Reproducibility: We ensure all analyses can be independently verified and replicated
3. Transparency: We work to produce interpretable methods with explicit assumptions
4. Humility: We recognize the limits of our current knowledge and remain open to revision and critique
5. Human-first:
   • We pledge to only use AI as a supporting writing tool
   • We encourage dialogue through comment sections
6. Accessibility: We release content in multiple formats to reach diverse audiences
7. Occam’s Razor: We strive for content to be as simple as possible, but no simpler

• Meet regularly to discuss new ideas and unfinished works
• Write, review, and disseminate thoughtful, bite-sized posts for Data Diction
• Collaborate on grants and papers

• Launch working group
  • define membership & engagement structure
  • define internal peer review process for content
• Hold first open quarterly meeting
• Launch blog (Data Diction)
• Draft first set of blog posts
• Publish first set of blog posts
• Secure external funding
• Organize a national short course or workshop
  • e.g. traveling or at JSM
  • potential topics: glass box modeling, post-selection inference

Data Diction started in 2022 as a blog, but posts were infrequent. Now, the website is maintained by this working group as a faculty-led scholarly outlet that emphasizes high-quality methodological exposition, including quick tutorials, opinion pieces, etc.

The Glassbox Modeling Working Group is a faculty-led methodological research community with an associated scholarly outlet, Data Diction. We welcome participation from faculty, trainees, and interested members of the broader community. Participation is structured as a progression of roles, allowing individuals to engage at increasing levels of responsibility as interest, readiness, and trust are established.

Participation typically begins with open community engagement and may progress through review to contribution. Advancement is based on demonstrated engagement, judgment, and alignment with the group’s standards for quality and clarity.

1. Community Participants (Open)
Students and trainees are welcome to participate publicly by:

• Reading and sharing Data Diction posts
• Commenting on posts
• Participating in open meetings or reading groups

This level is open to all and carries no expectation of productivity or formal contribution.

2. Reviewers (Opt-in, Curated)
Individuals who wish to engage more deeply may serve as reviewers for Data Diction posts. Reviewers:

• Provide structured feedback on clarity, assumptions, exposition, and audience
• Engage through GitHub issues or pull-request reviews
• May be acknowledged by name as a Reviewer on published posts

This level requires collaborator status on the Data Diction GitHub Repository. Reviewer roles emphasize training in critical reading and scholarly evaluation and do not imply authorship.

3. Contributors (Selective, Mentored)
Authorship on Data Diction posts is selective and responsibility-based. Contributors:

• Have prior experience as reviewers
• Demonstrate readiness
• Draft posts under faculty editorial oversight
• Must contribute content that aligns with the mission and meets the standards of the group

This tiered structure supports open participation from a broad community while maintaining high standards for methodological scholarship. By distinguishing between participation, review, and contribution, we hope to encourage meaningful engagement, protect quality and clarity, and support a sustainable editorial workflow.