How a Take Gets Written (And Why Some Never Make It)

A prediction market is only as good as its question. You can build a flawless pricing engine and a flawless resolution pipeline, and a sloppily worded take will still produce an argument nobody can win.

So authoring is not a content task on this platform. It's a quality-control process with a rejection rate. Here's what it involves.

What a good question has

Four properties, and a candidate has to have all four.

A single, unambiguous event. "Will the team have a good season" is not a question. "Will the team qualify for the knockout stage" is. If two careful readers can disagree about whether the event occurred, the question is not finished.

A checkable source. Every take is written against a class of source that will actually publish the answer: a scorecard, an official announcement, a certified reporting body, a governing body's result. If no source will produce a definitive answer, the question cannot be settled fairly, no matter how interesting it is.

A window that reality can meet. The resolution deadline has to be after the event and after the source will have published. These are different times, and conflating them is one of the most common authoring errors — a market on a box-office threshold that resolves the morning after release is a market that will be checking a number nobody has certified yet.

Genuine uncertainty. A question whose answer is already known is not a market. Neither, in practice, is one at 99% — nobody learns anything and nobody has a reason to participate.

The stale-event check

The hardest bug we've had to fix in authoring was not a wording problem. It was a time problem.

A market can be generated about an event that has already happened. The mechanism is subtle: a question gets drafted, its deadline is validated as being in the future, and the validation "helpfully" moves a past date forward to satisfy that rule — turning a question about last Tuesday into a question about next Tuesday. The result is a market about a settled fact, priced by people who may or may not already know the answer.

That is unfair in a way that's difficult to detect from the outside, so authoring now runs a two-sided check: the event window has to be in the future and the deadline has to be inside a sane horizon. A deadline that only became valid because something adjusted it is treated as a rejection, not a fix.

Why takes have a horizon cap

Every take has to resolve within a bounded window from when it opens, and we prioritise the short ones.

This is a product decision with a real trade-off. Long-horizon questions are often the most interesting ones — season outcomes, career milestones, year-end results. But a market that resolves in eleven months ties up Stars for eleven months, gives no feedback, and teaches a new player nothing about whether their judgement is any good. A platform full of them feels dead.

Short horizons produce a feedback loop: make a call, find out, adjust. That loop is how anyone actually improves. So the library skews short, with longer questions as a deliberate minority rather than the default.

Coverage, and the boring problem of what to write about

There are more possible questions than anyone can author, and they aren't equally worth writing. Two forces decide what gets made.

Demand — what people are actually engaging with. A subcategory where players are active deserves more markets.

Saturation — how much already exists there. Fifteen markets about the same match is not coverage, it's clutter, and it splits liquidity across questions that all resolve on the same event.

Balancing those two is what stops the library from collapsing into whatever is loudest this week. It also means a quiet category with real interest gets markets it wouldn't get from pure popularity ranking.

The questions we reject

A rejected candidate is usually rejected for one of these:

No definitive source. Interesting, unanswerable. The most common rejection by far.

Subjective criteria. "Will the film be well received" — by whom, measured how? If the answer requires a judgement call, the resolution requires a judgement call.

Ambiguous wording. Anything where the sneaky words are doing unexamined work: "officially", "consecutive", "any", "during", "announced". Either the criteria pin these down or the question goes back.

Impossible windows. The event can't happen inside the window, or the source can't report inside it.

Already decided. Caught by the stale-event check.

Nothing at stake. Questions where the answer is effectively known.

Human review, and where the line is

Not every take is auto-approved. Candidates that pass the automated checks but score below a confidence bar go to a person, with the reason for the hold attached in plain language — not an error code, but "the deadline is tight against a slow source" or "the criteria don't say what happens if the match is abandoned."

The point of surfacing the reason rather than just the flag is that it makes the fix obvious. A reviewer who knows why something was held can repair the wording in a minute. A reviewer looking at an opaque rejection re-derives the problem from scratch every time.

The rule we don't break

Resolution criteria are never edited after a take opens.

If we discover a question was written badly, we don't quietly redefine it — we void it and return everyone's Stars. That costs us a market and it occasionally costs someone a call they would have won under the clarified wording. It's still the only defensible policy, because the alternative is a platform where the terms of your prediction can change after you've made it.

Which is also why so much care goes in before the take opens. Once it's live, the wording is the contract.

What this means for you

Two takeaways.

The criteria are the question. Not the headline. When a take's title and its criteria seem to differ in scope, the criteria are what gets answered.

A take that looks obviously mispriced deserves a second read before it deserves Stars. The authoring process is good, not perfect. Sometimes an odd-looking price means the crowd is wrong. Rather more often it means everyone else read a word in the criteria that you skimmed.