predictionmkts

i have been working on prediction markets for the past few months.
( launching beta this week )

prediction markets are a novel defi primitive but the following things in current exchange such as Polymarket, Kalshi kept me bugging : 

- you can not create a prediction market on your own
- the markets gets settled by a cen. third party

i think i have finally worked it out, here's the design of PNP Exchange,
a truly permissionless exchange.

2/n

most of the centralised dependencies in current prediction markets arise from the fact that all the market trades happen offchain via an orderbook model.

so, the first step would be to bring trading outcome choices completely onchain.
we do that by using a Pythagorean Bonding Curve, acting as a counterparty to all trades.

dev, 22, researching permissionless prediction markets, @iitdelhi, prev intern @biconomy

4/n
We use intelligent LLM as a reasoning agent and task them to settle the markets.
An LLM model that has access to real-time factual information from around the world. the timing can not be more better.
Perplexity Sonar API has been found to correctly settle the test markets we have created.
So with the third aspect of a market i.e. market settlement being solved,
we finally get a complete permissionless flow.

5/n
- Anyone can create a prediction market.
- User puts question(string) onchain.
- YES and NO tokens are available for trading.
- Trading happens against bonding curve, onchain :)
- Market settlement done by an AVS network of AI models.

This is pnp exchange

If anybody is interested in trying this out do let me know,
I am launching this beta this coming week.
Would love to hear your feedback

3/n

Market creation -> onchain
Market trading -> onchain
Market Settlement ?

here is another centralised vector in the markets run by Polymarket, Kalshi : they way they settle.

these exchanges use Optimistic Settlement which basically means there's a guy in the end, fetching "relevant data" from "truth source" of that particular market, making a reasoning around it ( human bias? )
and pushing the answer onchain.

this is so 1800s.

we solve this problem by using Perplexity Sonar API.