Play-Money Prediction Platforms

What settles in play money, and what that removes

markets.settlement: play_money is not "the same product without the money". Removing the money removes a whole layer of the thing: there is no deposit, no withdrawal, no commission, no custody and no regulator. It also removes the gatekeeping, which is why these platforms carry questions no exchange would list, and why every card in this collection is open globally, open to US persons and asks for no identity documents.

The catalogue makes half of that structural rather than editorial. A card that claims play-money settlement and a non-zero taker or maker fee fails validation, because there is nothing to take a cut of. Manifold says the same thing from inside its own source: the taker fee constant and the flat trade fee are both zero.

The currency is the part readers get wrong most often. Manifold's mana is bought at a flat 100 mana per US dollar and, by its own terms, has no cash or redeemable value and cannot be cashed out, sold, traded, transferred, gifted or exchanged — there is no charity route out either. So a mana pack is a purchase, not a deposit, and the sweepstakes layer that briefly looked like an exception is gone: Manifold announced the wind-down on 13 February 2025, stopped issuing sweepcash immediately, gave holders until 28 March to redeem, and converted the remainder at 100 mana per sweepcash.

Money does reach forecasters here, but always from the other direction, as a prize for a result rather than a redemption of a balance. Hypermind runs contests with a fixed reward in euro or dollars, split among the forecasters who finish with positive performance and paid as Amazon gift certificates, while the trading itself is in a virtual currency with no value outside the game. Prediki calls it an Incentive: a sponsor pays a sum in advance and it is apportioned at settlement by how early and how precisely you forecast — which also means that the money arrives only when a sponsor does. Metaculus divides a fixed tournament pool in proportion to the squared sum of your Peer scores, and a negative sum pays nothing.

Half of these are not markets at all

This is the split to notice before picking one, and the flag that carries it is economics.liquidity_model.

Three of these cards make a price. Manifold is an automated market maker with limit orders resting on top of it, Hypermind runs an order book, Prediki moves a price as you invest Credits. A crowd is taking sides and the number you read is the result.

The rest make no price at all. Metaculus, Good Judgment Open, Fatebook, Confido, Quantified Intuitions and the Social Science Prediction Platform ask you to state a probability and score it against a proper scoring rule when the question resolves. Nothing is bought, nothing is held, and there is no position to close early. They are in this collection because the settlement field is the honest answer for a platform where nothing settles in money — not because they are cheap versions of a venue.

That difference decides which one you want. A crowd price aggregates what many people believe, weighted by how much of a consumable they were willing to put behind it. A proper score measures you, and on Metaculus the design is explicit: everything is built on the log score, the Baseline score compares you to a fixed benchmark, and Peer scores sum to zero by construction across everyone on the same question.

Who this is actually for

• Learning calibration. Quantified Intuitions' Pastcasting hands you questions that already resolved and scores you against the crowd of the day, so the feedback arrives in ninety seconds instead of six months. Fatebook is the same instinct pointed at your own work: write the prediction down in Slack or a browser, mostly privately, and collect a Brier score and a calibration chart over time.
• Running it inside an organisation. Confido is AGPL-3.0 and self-hosted, with rooms, tournaments and calibration curves, for a team whose questions cannot go on somebody else's site.
• Testing a strategy, or code, before it costs anything. Manifold has an order book, an API and automation, and being wrong costs mana. The Manifold API is a realistic rehearsal surface for the shape of code you would later point at a cash venue — and the rehearsal is missing the two things that will actually break it, which are fees and slippage against people who paid to be there.
• Being somewhere the cash venues are not. No deposit means no licence, no identity check and no geoblock. For a reader the venues will not accept, this collection is the part of the catalogue that is open — with the understanding that what is open is the question and the score, not a payout.
• Research. The Social Science Prediction Platform asks what an unpublished study will find and ranks forecasters by error against the real estimate, which is a different exercise from everything above.

The trap, stated precisely

A price on a thin cash market is unreliable because there is not enough size on it: the incentive to correct a mispricing is real, but nobody large enough has shown up yet. A play-money price fails differently. The incentive itself is missing. Mana goes in at a fixed rate and never comes out, so moving a price costs whatever the currency cost and correcting one pays a number on a profile. There is no pool of capital waiting for a mistake, and the question that is wrong can stay wrong for as long as nobody finds it interesting.

Three consequences worth carrying into any number you read here:

• Profit is not accuracy. Someone who farms many small mispricings in thin questions and someone who is well calibrated on hard ones end up in very different places on a mana leaderboard, and the leaderboard cannot tell you which you are looking at.
• The rules of the question are looser. On Manifold anyone can open a question and resolve it; the recourse when a creator gets it wrong is a report to a moderator rather than a rulebook. On a cash venue, resolution is where the money is, and it is governed accordingly.
• Incentive schemes bend the price towards the scheme. Prediki apportions its sponsor money by how early and how precisely you forecast, which rewards moving first. Hypermind pays only forecasters who finish positive. Neither is hidden, and both are things the number is partly measuring.

None of which makes the prices worthless. Manifold's own platform-wide Brier score is a real result and it is published rather than claimed. But it is a statement about thousands of resolved questions in aggregate, and the question in front of you may have had four participants and a creator who wrote the resolution criteria in one line.

Nothing here has been traded by this site with a funded account, because on these platforms there is no such thing. Currency rules, prize mechanics and scoring formulas above are read from each platform's own documentation and terms, and dated on the cards.