Open-question packager

Every review ends with things it could not settle. They usually survive as a paragraph of caveats near the back, written in the last hour before the deadline. Six months later nobody can tell which of them mattered. A packager turns each one into a piece of work with edges.

The unit of output is a one-page brief per open question. It states the question, what the review already established around it, why an answer would change something, a method that would produce one, and roughly what that costs. Somebody who was not in the round should be able to read it and start.

This is a quality step, not a speed one. It adds a day to the end of a round everybody wants to finish. What it buys is that the review's honest gaps stay legible after the team that found them has moved on.

The briefs are not findings, and they should not read as a critique of the work. An open question means the round reached its edge on schedule. Reviews reporting no open questions are usually the ones that stopped looking.

The raw material already exists in four places. Caveats written into the findings. Cells left as unknown during scoring. Claims two reviewers argued over and settled by agreeing to note the split. And assumptions the conclusion rests on that no source in the pool actually tested.

A model can pull candidates from those places, and it should quote the passage each one came from. A question with no passage behind it is an invention. That one rule keeps the harvest tied to the round rather than to the topic in general.

Where confidence was recorded as a range rather than a verdict, the harvest is easier. A register holding decisions as distributions already marks which claims were held loosely, and those are the first candidates.

Then cut hard. A long review can generate forty candidate questions and most are not worth a page. Keep the ones where an answer would change an action somebody is planning to take. The rest go in a plain list at the back with no brief attached.

If the round produced no written caveats and no unknown cells, the packager returns nothing. It is not a generator of fresh research agendas, and it cannot audit work it was never given.

The question has to be answerable. Weak ones read like themes: long-term effects, the role of context, implementation quality. A usable question names a population, an outcome, and a comparison. Turning a theme into a question is most of the work in this step.

The reason has to point at a decision. Not that the literature is thin, but that a named body is choosing between two things and this gap sits under the choice. Questions that tie to no pending decision are honest curiosity and belong in the plain list.

The method is a sketch, not a protocol. A design, a data source that plausibly exists, and a rough scale. Three sentences is enough. The point is to show that an answer is reachable, and to expose the questions where it is not.

The stopping point separates a brief from a wish. It says what result would close the question, and what result would leave it open. A brief without one becomes an open-ended workstream, which is how two weeks of work turns into a year.

The wording should match the round it came from. Where scope produced a hypothesis tree and quantifiable screening criteria, the brief reuses those terms. A question written in new vocabulary reads as a different project.

Everything that follows, counts included, is a constructed example. A justice ministry's research unit has finished a twelve-week review of education programs in adult prisons and their effect on reoffending. The conclusion supports continued funding for vocational courses and is cautious about everything else.

The harvest returns 31 candidate uncertainties. Nineteen come from caveats in the findings, seven from unknown cells in the scoring table, and five from disagreements the panel recorded rather than resolved. Four survive the cut.

The first asks whether course completion or course enrollment carries the effect. Almost every study in the pool measured enrollment, and the ministry is about to set a target on one of the two. The sketch is an analysis of linked records the prison service already holds. It closes if completion and enrollment show the same association across two years of releases.

The second covers women's prisons, where the pool held three studies against forty. The third asks whether the vocational effect survives in labor markets with no local demand for the trade taught. The fourth is a cost question the review could not answer, because only two studies reported cost at all.

Each brief fits on a page and names the passage in the review it came from. The other 27 uncertainties sit in an appendix, one line each, with no method attached. Two of the four briefs are commissioned within the year. That beats the caveat paragraph the unit used to write.

The usual failure is volume. A model asked to extract open questions from a long review will find dozens, and each will look reasonable on its own. A pack of forty briefs reads as a research program nobody asked for, and the recipient shelves all of it.

Generic questions are next. Every review can end with a call for longer follow-up, larger samples, and better cost data. Those lines are true of almost any evidence base, so they say nothing about this one. A brief that would fit another review unchanged is not specific enough to keep.

Some gaps are not research questions. A gap that exists because a data holder will not share a file is a permissions problem. A gap that exists because nobody funded the work is a budget problem. Packaging those as studies sends people to run analyses that evidence was never blocking.

Briefs also go stale. Date every one, and check the live ones whenever the review is revisited, because a question answered elsewhere can otherwise be commissioned twice.

The quiet failure is separation. Briefs get filed away from the review, so a reader of the findings never learns which parts rest on a gap. Caveats that can be hidden stop working, which is why the versions written for each audience carry the same open questions.

A brief is a proposal. Somebody has to say which two of four are worth doing, and that is usually the research lead who owned the round. The ranking is a judgment about the organization's plans, not about the size of the gap.

Funding sits with the recipient. A ministry, a committee, or a program director decides whether a question is worth commissioning. They weigh things the review never saw: budget cycles, political timing, and work already running elsewhere. A pack that assumes its own briefs will be funded reads badly.

When a question comes back answered, somebody has to fold it in. Structured check-ins with the recipient are the usual route, and a closed question should be marked closed in the original review rather than left standing.

This step sits near the ranking of which inputs would most reduce decision uncertainty, and the two should not be confused. That ranking runs while a decision is still open and orders work by influence on the choice. Packaging runs after the round closed and records what stayed unknown.

No brief is a finding. It says the round reached the edge of what it could answer, and wrote that edge down where somebody else can see it.