# White Paper

The goal of prediction markets is to seek ground truth through the pricing of events which are probabilistic. Suppose then, you want to:<br>

* Incentivize a group of agents (traders) to price the events with statistical signal
* Hone the signal by allocating a reward to those who can most likely inform the market<br>

The prediction market could respect the amount of "skin in the game" the agent has (trade size), which signifies the scale of conviction. On the other hand, they could also choose to filter skill by the cumulative return (ROI) an agent has when taking positions over time.  In the real world both notions are usually considered in order to demarcate agents who can consistantly [adversely select](#user-content-fn-1)[^1] in prediction markets.

To incentivize the provision of consistent, valuable signals, we propose a tournament structure on [**Bittensor**](https://bittensor.com/) where miners compete by providing predictive signals, and reward those whose information demonstrates positive edge. First, [**qualified** **volume**](#user-content-fn-2)[^2]**,** a miner’s past flow or historical volume which aligned with the true outcome, shows how much usable weight they bring into the system. Second, **return on investment (ROI),** their historical profitability, demonstrates whether that flow is actually informed and positively selective.\
\
We optimize miner output for superior information via a [convex program](#user-content-fn-3)[^3] that balances these aspects: it **maximizes routed qualified volume** across eligible miners while constraining the reward budget (alpha token emissions) to minimize the cost of the volume, then uses ROI to shape payout costs to be provably efficient. In this way, both volume and ROI matter: volume drives the scale of contribution, while ROI determines whether a miner qualifies and how expensive it is to fund them further.

For a prediction market like [Polymarket](https://polymarket.com), the tournament structure increases the likelihood of concentrating rewards to miners whose signals align with truth and avoids those who provide spurious, short term noise. Through careful incentivization, Almanac manufactures effective *signal concentration* by culling weak statistical signals that do not adapt to everchanging market conditions.

[^1]: Adverse Selection:  A market situation in economics where one party in a transaction has more or better information than the other, leading to inefficient market outcomes.

[^2]: The volume bet into a market which resolves as a winning prediction.  Losing Volume is omitted.  This is what is meant by "qualified".

[^3]: Convex optimization is a subfield of mathematical optimization that studies the problem of *minimizing convex functions over convex sets*.