Question twelve awaits peer review

A regional pub quiz league could, by 2030, subscribe to a tier of questions whose answers have not yet been confirmed by science. The questions would be drawn from freshly posted preprints, too new for any assistant to have read, and the answers would be marked correct pending replication. Teams would be told the outcome within eighteen months.

AI assistants built into earbuds and smart glasses can now answer most general-knowledge questions within seconds, and pub quiz organisers have responded with phone bans, sealed pouches and questions that reward local knowledge or things only visible in the room. Language models know the world only up to a training cutoff, while preprint servers publish research findings months before peer-reviewed journals do. Question writers increasingly favour recent or hard-to-search material to stay ahead of search tools. Academic publishers and repositories have been experimenting with paid, tiered services.

Customer: My team lost the tiebreak on question 12 and the answer on the sheet has now changed. Agent: I can see that. Question 12 was drawn from the Preprint tier, so its answer remains provisional until a second laboratory confirms the result. That exchange, from a support transcript a quiz league would publish as good practice, describes the situation. Under the scheme, a league running quizzes in some forty pubs would pay a monthly fee to a preprint repository for questions whose answers exist nowhere else. The tiebreak would be settled in the pub. The answer would be settled wherever the second laboratory happened to be.

The logic would be straightforward. Assistants in earbuds and glasses can answer any question that has ever been written down, and the league's phone ban would have stopped working the week the glasses became indistinguishable from glasses. The only questions a model could not answer would be the ones nobody had answered yet. The repository's Quiz tier would therefore supply findings posted in the previous seventy-two hours, formatted as questions, with the authors' own conclusion printed as the answer. A footnote would add that the authors' conclusion had been reached by the authors. Teams would be advised to treat the footnote as part of the question.

On a Tuesday in a pub near a ring road, the science round would run as follows. Question nine: at what temperature would a particular lithium compound have stopped behaving as expected in a laboratory the previous week. Question ten: how many of the mice. A team called Norfolk and Chance, four people who had never before discussed superconductivity, would write “four kelvin” on the grounds that it sounded cold. “We got a point for it,” Dawn Pryce, 52, a dental receptionist, would say. “Then we got an email in March saying we hadn't. I'd never been retracted before. It's a strange feeling, but you get used to it.”

Points on the Preprint tier would be held in escrow. The league table would carry two columns, one for points awarded and one for points confirmed, and the season would end when the second column caught up with the first, which it would not. A team could be promoted in May and relegated in September on the strength of a failed replication in another hemisphere. The rules committee would introduce a third column for points that had been confirmed and then withdrawn again. It would be called the Contested column. It would be the longest column.

Some teams would stop waiting. A team of retired engineers in a market town would replicate question 14, concerning the load-bearing behaviour of a new foam, in a shed, using a fridge and a bathroom scale. Their result would disagree with the preprint. They would email the authors, who would email the repository, which would email the league, which would adjust the score. “We were credited as co-authors in the end,” Terry Oakes, 71, would say. “The paper's been cited twice. Once by another quiz league.” The foam would, on balance, hold.

The repository's director of product would describe the Quiz tier as “the first citizen-science programme funded entirely by crisps.” Revenue from the tier would overtake institutional subscriptions within two years, and the repository would begin formatting all new submissions as questions by default, with a field for plausible wrong answers that authors would be encouraged to fill in themselves. Universities would ask whether their own researchers could enter as teams; the answer would be yes, at the institutional rate. Authors would find the wrong-answer field the most honest section of the form. Peer reviewers would find it the most useful.

The support desk would remain the busiest part of the operation. A team that had answered correctly with a result later disproved would keep the point on the grounds that it had been correct at the time, which the league's terms would describe as the historical standard. A team that had answered wrongly with a result that later turned out to be right would receive nothing, because the result had not been available. Dawn Pryce would ask whether this meant the league was marking teams on their ability to be wrong in the right year. The agent would confirm that it did. The chat would be closed as resolved.