AL #71: How Polymarket's Architecture Supports Its Mainstream Success
The order book prediction market that speaks to more than just crypto natives.
Whenever a crypto app achieves mainstream success, the first order question is usually their go-to market strategy.
Due to the positioning and regulatory challenges associated with crypto, any app that is able to grow past its crypto native user base has done something interesting.
Polymarket, the leading prediction market, definitely fits the bill.
Nate Silver, one of the most recognizable authorities on prediction, explained why Polymarket’s product is so special:
And while distribution is part of driving success, I'm also always interested in what architecture and dependencies successful crypto apps use behind the scenes.
Since mainstream audiences are so demanding when it comes to user experience, these architectures, while sometimes making concessions on trustlessness, can be indicative of future trends in smart contracts
Last year, we did a deep dive on Friend.tech and some of the technology behind it:
It’s time to look at Polymarket the same way.
But what exactly is the prediction market use case and why does it matter?
Why Prediction Markets are Cool
Prediction markets allow “event trading” or speculation on the clearly defined outcomes of certain events.
They are a form of derivative – an asset that does not have a physical representation but merely relies on contracts where some outcomes designate one party as the winner and the other as the loser (to various degrees).
Fun fact: Prediction markets were the first use case of smart contracts that I was exposed to.
Matt Clifford told me about a platform called Augur that was built on Ethereum. I didn't know about either that time.
Years later I helped audit a fork of Augur called Veil and since then have been interested in prediction market design.
The thing that makes prediction markets interesting is that they create a public good which is a high-quality real-time aggregated probability of an event occurring.
Prediction markets:
Can react to events faster than the media;
Incentivize participants with new information to participate;
Often outperform other forecasting methods like Delphi.
Limitations of Prediction Markets
While it’s exciting that prediction markets can underpin new crypto-native news services, the implied probabilities of a market can differ from the prices of the outcomes.
Limitations related to numeraire
Dan Robinson has an excellent thread on this. It started very innocently:
Limitations related to hedging
The relationship between prediction markets and futures is explained well in this post by Filippo Armani:
By the way, if you would like to read more about futures, we covered them earlier:
Limitations related to reflexivity
Trump likes to refer to his Polymarket results on Truth Social.
Presumably he does that because his team believes that a high win probability helps increase the odds of success.
If that is true then there must necessarily be some positive value to market manipulation.
Whether the ROI is worth it at current market prices I’ll leave for another blog but the possibility for reflexivity is very real especially for lower liquidity markets.
Polymarket's Architecture
There are 3 architectural choices that have enabled Polymarket to scale:
Sophisticated use of “arcade wallets”
Central Limit Order Book (CLOB) with off chain operators
Reliance on well-understood primitives (Uma Optimistic Oracle, Gnosis Conditional Token Framework)
1. Sophisticated use of “arcade wallets”
We’re still in a stage of crypto adoption where services (just like friend.tech and NBA Top Shot did in the past) are more successful when they use arcade wallets.
Arcade wallets create wallets on behalf of users so they don't have to rely on self-custody and can use web2 auth.
Thread below:
2. Central Limit Order Book (CLOB) with off chain operators
While Polymarket settles to Polygon, it stores its order book off chain via operators.
While Polymarket’s initial AMM model was great for bootstrapping markets, CLOBs scale better for liquid markets, in particular they support much larger trades at viable slippage levels.
While the CLOB model is less decentralized, Polymarket aims to mitigate these effects by exposing an API.
One interesting aspect of prediction markets is that they illustrate the coupling of issuance and trading.
Different tokens necessitate different ways of trading.
For example stablecoins are best traded with very low slippages around the 1:1 peg price.
Prediction markets have to support matching and splitting in trading – owning 1 share of a Yes and No outcome should yield the full amount of collateral.
As Uniswap V4 emerges, it may support its own native prediction markets as hooks (that could still leverage off-chain order books).
3. Reliance on well-understood primitives
Polymarket is commendable by building on known primitives:
Gnosis Conditional Token Framework
UMA Optimistic Oracle
Each of them may have their issues (Gnosis Conditional Tokens were built for their own prediction market and the UMA Optimistic Oracle’s parameter choices like challenge windows may impact how many markets can be actively operated).
But they are components that the Polymarket team doesn't have to build until they need to make radical improvements.
Instead, Polymarket can focus on the user onboarding and the performance of the order book to ensure the best possible UX.
While none of these elements is rocket science, their combination has resulted in a web2-like experience for Polymarket’s users.
This experience in turn is reinforcing Polymarket’s market leadership through partnerships with non-crypto platforms like Substack:
There is a lot more to say on how Polymarket is accruing and maintaining market power but we will leave that for another post.
Comment below if you are interested in how that plays out.