Questions before papers
Start with the decisions and uncertainties that matter. Papers, evaluations, data, and models then become evidence attached to those questions.
1. The proposal
A small, persistent map—not another large database
Each group could maintain a limited set of questions and research clusters that seem especially likely to inform consequential decisions. Individual members could maintain their own prioritization; the group view could summarize the discussion without implying agreement where there isn't any.
Keep disagreement visible
The list needn't be a consensus ranking. Members can give separate judgments, propose alternatives, and explain why they would prioritize something differently.
Maintain, prune, and revisit
Each active item could have a human owner and a review date. Deprioritized items could be folded or filtered rather than deleted, with a short explanation.
How this could fit with field specialist work
For field specialist groups and meetings, this could happen alongside discussion of actual papers, prioritization assessments, final voting, evaluation management, and discussions about our process. It could integrate with some of these activities by providing a wider view of themes, possible directions, and connections to impact.
2. A possible structure
One compact table, with several views
The same underlying material could support a group shortlist, individual members' suggestions, prompts imported from existing prioritization work, and a record of retired items.
| Question or cluster | Who might use the answer? | Why it may matter | Current disagreement | Promising next work | Owner and review |
|---|---|---|---|---|---|
| Short, decision-relevant formulation | Named types of funders, policymakers, researchers, or advocates | A concise path from better evidence to better decisions | What informed people disagree about, including minority views | Evaluation, synthesis, replication, modeling, new data, or experiment | Human owner Review every 6–12 months |
Suggested views
- Group-curated: no more than five to seven prioritized questions or clusters.
- Member suggestions: separate nominations and judgments, not forced into a common ranking.
- Prompts to consider: items drawn from our databases, funder agendas, crux mapping, or AI-assisted research.
- Deprioritized or retired: preserved with reasons and a date.
Possible lightweight judgments
- How much could an answer change real decisions?
- How uncertain or disputed is the answer?
- Is useful progress tractable in the next few years?
- Is the area receiving attention roughly proportionate to its importance?
- What evidence would most change the member's view?
3. Illustrative starting points
Examples to react to—not proposed group conclusions
These are concrete prompts drawn from papers already discussed or prioritized by The Unjournal and from current Pivotal Questions work. They are deliberately incomplete, and the groups may reject or substantially rewrite them.
Animal welfare, markets, and attitudes
- What share of plant-based meat purchases displaces chicken, beef, or no animal product—and how does this vary with price, availability, and consumer type?
- Which inputs—growth-factor prices, cell density, bioreactor scale, capital costs, or energy use—dominate uncertainty in cultivated-meat cost projections?
- How should evaluators compare interventions that affect different species, numbers of animals, lengths of time, and severities of suffering?
Global health and development
- Which lead-exposure interventions in lower-income countries—paint regulation, cookware replacement, remediation, or nutrition—are most cost-effective, and how uncertain are current burden estimates?
- How durable are gains from group psychotherapy or cognitive behavioral therapy in lower-income settings, and when does therapy outperform or complement cash transfers?
- Do agricultural information services raise smallholder productivity at scale, and are language access, credit, or insurance the more important constraints?
AI governance and catastrophic risks
- How rapidly do returns to reinforcement-learning compute diminish relative to pretraining or inference-time compute, and what would this imply for capability timelines?
- Do current frontier-model dangerous-capability evaluations predict the capabilities that appear after deployment, tool use, or stronger elicitation?
- Will the fixed costs of compute and training concentrate frontier AI in a few firms—and would concentration make effective oversight easier or harder?
Psychology, behavioral science, and attitudes
- Which misinformation interventions produce durable changes in sharing or real-world behavior, rather than only short-lived corrections to stated beliefs?
- Which messages or experiences expand moral concern toward farmed animals, future generations, or artificial beings—and which produce backlash or polarization?
- When people use AI advice, what predicts appropriate reliance rather than over-trust, and how does AI use affect confidence, polarization, or willingness to defer to experts?
4. Human judgment and AI
AI can supply prompts, but it shouldn't become the group's view
AI research prioritization, literature mapping, and crux mapping may be useful inputs. The group should retain authorship, judgment, and control.
Independent input first
Members could record their own candidate questions before seeing AI-generated suggestions, reducing the risk that the prepared material anchors the whole discussion.
Clear provenance
AI-generated and database-imported suggestions should appear in a separate, labeled view, with the source and date. They are prompts to criticize, combine, rewrite, or discard.
Human responsibility
A named person should rewrite and take responsibility for anything promoted to the curated list. AI should never overwrite a member's judgment or silently revise the record.
5. A flexible meeting format
One example agenda—not a standard template
The schedule below is only an illustration. The mix should vary with what is timely: if several strong papers need decisions, paper discussion may take most of the meeting; at other meetings, the group might spend longer on the broader priority map.
0–5 min
Updates and agenda choices
Agree on where the meeting's attention is most useful today.
5–25 min
Actual research papers
Discuss promising papers, prioritization assessments, final votes, and possible evaluation managers.
25–35 min
Process questions
Clarify how research moves through the pipeline and identify avoidable frictions or gaps.
35–50 min
Pivotal questions and directions
Add, challenge, refine, or retire items in the group's longer-term priority map.
50–60 min
Decisions and ownership
Record next steps, owners, unresolved disagreements, and what should return at a later meeting.
Optional +20
Members' research and research use
An informal space for people who can stay longer to discuss their own work, how research is being used, possible collaborations, or useful feedback.
This is one possible allocation, not a standard agenda. The group could shorten, extend, reorder, or omit sections. Some meetings might be almost entirely about papers; an occasional meeting might focus mainly on the broader priority map.
6. Questions for Monday
What should we keep, change, or drop?
About the underlying idea
- Would a maintained question-and-cluster map actually help your group?
- Should there be a short group-curated list, individual lists, or both?
- Who might use the result—and what would make it genuinely useful to them?
- How often should it be revisited?
About the meeting and workflow
- How much meeting time, if any, should this receive?
- Should independent member suggestions come before prepared database and AI prompts?
- What is the minimum information needed for an item to enter the curated list?
- Would anyone be willing to help pilot or maintain this for one group?
A possible small pilot
Try this with two or three active, decision-linked groups. Ask members for independent suggestions, use existing Unjournal prioritization material as prompts, and leave each group with no more than five to seven prioritized items. Revisit the experiment after one or two meetings before building a more elaborate Coda system.
Why start with a hosted page?
This lets us discuss the concept without adding another permanent Coda structure first. If the framework seems useful, the parts worth maintaining can later move into a compact Coda table or connect to the existing research-prioritization database.