Polaris

Every consequential decision rests on a belief about the future. Markets, governments, institutions, and individuals all act on forecasts - and yet forecasts are systematically wrong, routinely dishonest, and structurally biased by the incentives of those who make them.

Experts hedge. Analysts anchor to consensus. Institutions protect incumbents. The people with the most accurate private information have the least reason to share it.

This is the oracle problem: how do you extract distributed, private knowledge and compress it into honest, public predictions? How do you make truth the dominant strategy?

In 2002, Robin Hanson proved something elegant: a market can be a scoring rule.

When participants trade against a continuously-available automated market maker, they are compensated for accuracy and charged for error. The total cost of eliciting probability estimates from any number of participants, over any number of interactions, depends only on the initial and final states - not on the number of trades, not on the path. Arbitrarily many minds can contribute at no additional cost.

The market becomes an information aggregator. Its prices are the crowd's honest probabilities. The mechanism does not ask anyone to be honest - it makes honesty the only rational strategy.

The deep mathematics behind why this works is the theory of proper scoring rules.

A scoring rule is proper when the unique strategy that maximizes expected reward is to report your true beliefs - not an exaggeration, not a hedge, not what you think others want to hear. Proper rules derive entirely from convex functions. Gneiting and Raftery showed that propriety, entropy, and Bregman divergence are three faces of the same structure.

The market does not need to trust anyone. It constructs an environment where the mathematics of incentives and the mathematics of truth are the same thing. Honesty is not a virtue here. It is an equilibrium.

Traditional prediction markets ask binary questions. Will event X happen? Price it between zero and one. But the world is not binary, and forcing it into binary format destroys information.

When will a major AI model ship? Not a yes or no. Not a set of buckets chosen in advance by someone who had to guess the distribution before seeing any data. The honest answer is a probability distribution over all possible dates - a curve, not a number. The question is not which bucket but what shape.

Dave White's Distribution Markets showed how to build automated market makers over these infinite outcome spaces. The invariant uses L² norms in function space - an elegant generalization of Uniswap's xy=k to probability distributions. The market's reserves directly encode the crowd's collective belief as a full probability density. You can trade on a Gaussian centered at a specific date with a specific variance. The AMM's current state is the market's answer, continuously updated, readable at any moment.

This is what prediction markets become when the mathematics catches up to the problem.

Polaris markets never close. There is no opening bell, no settlement date that freezes the price, no window in which information can only accumulate but not be priced.

Every development, every signal, every trade reshapes the distribution instantly. The market is a living probability estimate - the aggregated intelligence of everyone who has taken a position, continuously revised as the world changes. Abernethy, Chen, and Vaughan showed that each price update is formally equivalent to a step of mirror descent: the market is running an online learning algorithm, minimizing regret against all possible outcomes simultaneously.

This is not a poll conducted monthly and reported with a margin of error. It is a machine that processes information in real time and emits a probability. You can query it at any moment and receive the crowd's current best answer to any question it covers.

Multiverse Finance completes the picture. When outcomes partition the world into verses - conditional universes - tokens denominated in the same verse share a critical property: if the verse fails to resolve, both the collateral and the debt vanish simultaneously.

This eliminates liquidation risk from conditional positions. A trader who holds conditional ETH and borrows conditional dollars, both conditional on the same event, faces no liquidation cascade if the event resolves adversarially. Both sides of the position are extinguished at once. The traditional oracle problem that breaks lending protocols - the gap between when collateral becomes worthless and when debt must be repaid - does not exist here.

Prediction markets become not just information infrastructure but financial infrastructure. Positions in the future carry the same engineering rigor as positions in the present.

The deepest unsolved problem in prediction markets is the incentive to discover. Why do the work of identifying talent, spotting trends, or finding signals - if acting on the discovery means revealing it to competitors who did none of the work?

A scout who bets heavily that an unsigned artist will succeed broadcasts the discovery to everyone watching the order book. The market extracts the information but does not compensate the discoverer for the cost of generating it.

Opportunity Markets address this directly. Private, sponsor-controlled prediction markets where only the first N participants can trade, where fills are delayed, where the order book is not public. The record label creates a market on whether it will sign a specific artist. Ten scouts participate independently, each rewarded in proportion to accuracy, none able to observe the others. The value of discovery stays with the discoverer long enough to be captured.

Information markets, finally, that reward the creation of information - not just its transmission.

Polaris is building continuous prediction markets as the information layer of the world. The infrastructure through which organizations, markets, and institutions price the future honestly - not as an aspiration but as a mathematical consequence of how the mechanism is constructed.

Our markets are always on - there is no closing time, no freeze. They are proper - incentive-compatible by construction, not by policy. They are complete - trading on full probability distributions over continuous outcome spaces, not on binary bets or preset buckets. They are conditional - with a financial engineering layer native to the structure of contingent reality. And they are private where privacy is the condition for discovery.

The lineage runs from Hanson's logarithmic scoring rules through convex duality and online learning theory to on-chain AMMs operating in function space. Two decades of mathematics, converging on a single machine for turning distributed human knowledge into honest public prices.