Chainlink comes to mind when people think about oracles. The most popular and widely utilized oracle today, Chainlink, a gem of the 2018–19 bear market, secured almost $12 billion across over 200 protocols, with the majority of TVS (Total Value Secured) coming from the Ethereum lending protocols Aave and Compound. Almost half of Chainlink's TVS is made up of just these two lending protocols. Pyth Oracle is a new player who began last year with a concentration on Solana and is now trying new things.

With Chainlink so deeply embedded in DeFi, it's reasonable to wonder why we need another oracle. Pyth, in addition to meeting the oracle requirement on Solana, takes a completely distinct and novel approach to oracles, obtaining information exclusively from first-party sources and avoiding the use of data aggregators in the intermediary. While some oracles, such as Chainlink, have some first-party run nodes, they also use aggregators (such as CMC and CoinGecko) as part of their operation. Pyth also introduces the idea of confidence intervals, a tool that provides protocols with more information regarding not just price but also prices uncertainty, especially during periods of high volatility/low liquidity.

Pyth’s comparative advantages

Only a limited number of users have timely and relevant market information. Trading businesses, market makers, and centralized exchanges are examples of these. Pyth has these participants post prices directly to Pyth, eliminating data aggregation expenses and considerably lowering latency from off-chain price determination to on-chain reporting, resulting in more timely and higher-quality prices. Many of Pyth's publishers are well-known, ranging from crypto corporations such as Amber and Wintermute to traditional financial institutions such as Jane Street and the Cboe.

Pyth collects prices from many publishers and combines them. Instead of a single price, they offer a price within a confidence interval (CI). There exists no such thing as a true price for assets that trade on several venues with varying participants, information, deposit/withdrawal limitations, liquidity profiles, and so on. Using confidence intervals across a single price might represent a more realistic market state and current liquidity levels, especially during periods of extreme volatility.

While this method necessitates additional considerations from DeFi protocols, it does provide certain benefits. Assume the aggregate price of ETH is $1,500 +/- $5. Multiply the CI by 3 to produce a price with a 99.7% chance of being right (assuming proper normal distribution estimations). This implies that the genuine value of ETH is most likely between $1,485 and $1,515. Protocols can determine their own CIs to determine whatever amount of conservativeness/aggressiveness they are comfortable with. A cautious borrow/lend protocol, for example, may value collateral at the low end of the range ($1,485), limiting the number of money customers may borrow while taking out a loan. A more ambitious lending protocol, on the other hand, can assess the collateral at the higher end ($1,515) to avoid needless liquidations.

The development progress

Pyth had some early problems that harmed its image. We'll discuss what happened and why, and then go into the specifics of how the Pyth engine works and how to avoid such problems in the future.

The first serious problem occurred on September 20th, 2021, around the BTC/USD pricing during a 2-minute timeframe. The aggregated price ended up being far off BTC's genuine price of $42,000, citing values as low as $5,402 with confidence intervals of +/- $21,623. This was due to two out of eleven publishers making a decimal point error and thus the aggregation logic overweighting their contributions. This inaccuracy resulted in worthless (and maybe harmful) price information. For protocols that did not include the confidence interval and instead relied solely on the median price of $5.4k, this had negative consequences, such as erroneous liquidations.

This incident would provide credibility to oracles that use aggregators. It's easy to argue that data aggregators would not have reported such pricing because it's their responsibility to filter anomalies like these. This is a tradeoff of not utilizing aggregators; by eliminating middlemen and the associated fees, you may get faster and perhaps higher-quality updates. Unfortunately, if the algorithm or stake weighting fails to account for outliers like the one described above, major problems arise. This intermediate filtering/aggregating step is a check that Pyth lacks, instead depending on crypto-economics and algorithms. So, what actions did Pyth take to correct the situation?

Created a testing protocol for fresh publishers with improved onboarding.

The aggregation mechanism has been tweaked to correctly weigh prices that cover such a wide range.

Improved developer documentation on how/why/when to utilize the pricing intervals. As previously said, this is a more complicated procedure for protocols, and many people are unfamiliar with the notion of CIs, instead accepting the median as they would with some other oracles.

Added more publishers: BTC/USD currently has 30 publishers, up from 11 when the issue occurred.

In the future, crypto-economic security will be achieved by stake weighing and slashing with the PYTH token.

Although protocols must comprehend how to use Pyth's data and confidence intervals, it is evident that an oracle cannot provide inaccurate pricing. Furthermore, having median prices and CIs that are so far from reality leads to a liveness failure at a best and financial loss at worst.

With all that being said, Pyth has been a lot more trustworthy since these early concerns, especially as their coverage (number of publications) has grown. It's also critical to discern between what is a true oracle failure, such as the BTC issue mentioned above, and what is market manipulation, such as the recent Mango or Solend assaults (keep in mind that Pyth did not cause either regardless). Oracles should be examined on whether their prices represent the genuine pricing on exchanges. Illiquid assets with small markets are vulnerable to manipulation, and standards for underwriting and determining risk boundaries for them must be strict.

Since the Mango attack, we've seen other protocols, such as Compound, delist some of these illiquid assets as a precautionary step, and this is another case where broadening confidence intervals or utilizing a TWAP, SMA, or EMA might help limit undesirable effects. Pyth has also begun to focus more on the notion of Liquidity Oracles, bringing an asset's liquidity into the underwriting process and thereby incorporating the market impact of selling the collateral asset into the limit formula. The purpose here is to keep users from opening positions that are too huge to close.

Developing a publisher-first oracle from the ground up is difficult owing to the early lack of market penetration, and publishers may be hesitant to join due to technical or reputational issues before the network is established. Pyth has begun to recover from its shaky start in the last year and continues to attract credible publishers. In addition to an improved onboarding/integration procedure and documentation, they've made significant modifications to their price aggregation method and will be adding additional measures with crypto-economic security around their planned token PYTH in the future. They've also recently released Pythnet, a clone of the Solana codebase that may be used as a standalone chain for non-Solana networks. We'll now go more specific about how this all works.

The mechanism of Pyth

Before we begin, it is important to remember that several aspects of this scheme (such as delegator staking and data fees) will not be operational until the PYTH token is released.

Pyth has three primary stakeholders: users, publishers, and delegators.

Users: These are those who utilize price feeds, primarily smart contracts, and DeFi protocols. Price feeds are provided for free in Pyth v1 on Solana. They pay minor "on-demand" fees to acquire pricing from Pythnet (v2) via Wormhole for protocols outside of Solana. Once the PYTH token is launched, users will be able to pay insurance to protect themselves against Oracle errors.
Publishers: Publish prices. They receive PYTH token incentives as well as a 20% discount on data expenses (subject to change). They will be required to stake PYTH tokens for the things they price, and their stake will be reduced if they make an error.
Delegators: PYTH token holders. They use pricing feeds to establish publisher weights and collect insurance costs from customers.

Delegators play a crucial role in establishing the aggregate pricing reported by Pyth. It is their responsibility to choose whether publications should have more or less impact on a feed, and they receive 80% of the fees paid by customers. Different markets will have varying amounts of risk and consequently, fee income/yield, and delegators must determine which markets to stake on. For example, a feed such as MINA/USD may give a 5% return to delegators compared to 1% for BTC/USD, representing the failure risk in each feed. These yields should grow more efficient at pricing genuine failure risk over time. The procedure of demonstrating a flawed oracle price will be carried out using the HUMAN protocol. We won't go into detail here, but the idea is that the on-chain oracle price released will be compared to off-chain "reference exchanges." If the published price differs from the reference price (within a +/- 3 CI band), the claim is successful. This procedure is divided into two phases. The first step is to claim a file. While anybody may register a claim, they must bind PYTH tokens in order to avoid claim spam. Following that, filed claims are approved (voted on) by PYTH token holders.

It's also critical to separate the danger of slashing between delegators and publishers. In the event of a failure, all delegators who stake on that particular feed will be slashed, regardless of whose publisher caused the negative pricing, even if they did not stake on them. Publishers will only be penalized if they are directly responsible for the inaccuracy. Delegators will need to balance several considerations when staking, such as overweighting high-reputation publishers while also ensuring that no single publisher has too much power, avoiding a small number of publisher failures from causing an aggregate failure.

As previously stated, this stake weighting will not be active until the PYTH token is released. Currently, each publisher is given equal weight. Pyth has revamped the aggregation method to be more tolerant of outliers, in addition to the upcoming stake weighting. Each publisher receives three votes from the algorithm: one at the median, one at the lower end of the spectrum of their interval, and one at the high spectrum. These votes are then combined, and the aggregate price is the median of all the votes, with the 25th - 75th percentiles centered on the aggregate confidence interval. This approach has the following advantages:

Strong against manipulation: Publishers that post prices that aren't in line with the truth will be disregarded.
Weight based on accuracy: Publishers with tighter ranges (more liquid exchanges) will have a disproportionate impact on the median aggregate price. A price of $101 +/- $1 published by one exchange will have a greater impact on the whole price than a price of $110 +/- $10.
Reflects variation: No asset trades at precisely the same price on every market venue, thus the stated price should take variance into account.

Publishers will be rewarded based not just on their stake weight but also on their prediction ability and honesty in disclosed confidence ranges once stake weights are in effect, which will use all of these criteria to decide the price.

Pyth on Solana

The fact that Pyth is involved in several of Solana's transactions should be the first thing that stands out. Pyth is in charge of about 30% of Solana's transactions after taking consensus votes out of the equation. Pyth's frequent updates function because it is on a chain with fixed transaction fees and minimal latency. Publishers are adjusting pricing on-chain every 600 ms in a Solana slot, which would be too expensive and impractical in other ecosystems.

Even with roughly 10 million transactions each day, becoming a Pyth publisher isn't prohibitively expensive. If we look at the publisher with the most pricing feeds (remember that publishers are anonymized), they paid $7.6k in the last month to update 80 feeds. Publishers will receive PYTH tokens, data fees, and will (probably) earn a token allocation at launch. In addition, Pythnet (Pyth's solution for non-Solana chains) offers on-demand "pull" price updates as opposed to Solana prices being "pushed" on-chain every slot. This will be detailed further below, however, the on-demand pricing from Pythnet will create extra fees if pulled on-chain.

solana-price-feed-fee

Source: Dune analytics

The second point to mention is how Pyth is utilized. While borrow/lend protocols often safeguard the most value (Aave and Compound for Chainlink, Solend for Pyth), they do not typically require the most price updates. This is due to derivatives protocols, which must take advantage of low-latency pricing, as opposed to a borrow/lend protocol, which does not need to update every slot. Pyth's oracles have enabled several unique products:

Mango: An exchange that utilizes Pyth to price its perp market's funding rate and for its borrow/lend book. For its borrow/lend book, Mango was one of the first projects to include Pyth's confidence intervals. It should be noted that the recent tragic market manipulation of Mango has nothing to do with Pyth.
Zeta: As mentioned above, a derivatives exchange that employs Pyth's confidence intervals to avoid settlement manipulation and to price their options and futures. Pyth is also used in their latest product, FLEX, which is a permissionless and customizable choices protocol.
01: A derivatives exchange that provides both standard perps and power perpetual, comparable to Opyn's Squeeth. These are conventional perpetual boosted to a power level, providing option-like exposure. Pyth is used to calculate funding rates.

Finally, we can look at which prices have been both posted and requested on the blockchain. This is arguably more beneficial for publications to know which feeds are not worth paying (ALGO, MATIC, LTC, BCH) and which are possibly under-covered, such as GMT. Being on Solana, it should come as no surprise that SOL is the most sought pricing feed. Every significant asset currently has more than 15 publishers.

While Pyth had a bumpy start on Solana, they've experienced considerable performance gains in the last year and currently cover virtually the entire ecosystem (90%+ of Solana's TVS). But Pyth has ambitions beyond Solana, which is why they established Pythnet in August.

Further adoption

Pythnet is a fork of Solana's codebase that serves as a separate chain for Pyth publishers to input pricing. They incorporated BSC on October 4th, Aptos on October 19th, then Ethereum and Optimism on November 1st. While Pythnet operates similarly to Pyth on Solana for publishers, the approach is different for users. This is because, unlike Solana, price changes are not pushed on-chain, instead relying on Wormhole to carry cross-chain communications. The biggest distinctions between here and Solana are as follows:

Pull vs push model: The prices are "pulled" from Pythnet and adjusted on-chain when an app requests them, as opposed to being "pushed" on Solana automatically.
Increased latency: The time between the CEX pricing and on-chain is 3 seconds. Two of these are from Pythnet to pricing service and one is from CEX to Pythnet (introduced by Wormhole messages).
Fees for price updates: Non-Solana protocols must pay to fetch price updates on-chain from Pythnet whereas Solana protocols can utilize Pyth for free. They begin inexpensively at the smallest denomination on-chain (for instance, 1 wei), but governance can subsequently alter them. You may view Pyth's fee collector on Aurora for a real-time illustration.
Gas efficiency: Only when consumers actually require prices are gas expenses paid.
Reliability: Pythnet is a trustworthy independent chain. So Pythnet continues to function even if the Solana network is unavailable.

pyth on other chains

Source: Pyth network

While the latency is not optimal when compared to the sub-second Solana edition, mitigations can be implemented, such as:

To avoid potential front-running, orders are settled asynchronously. The user transmits the market order; after some time (e.g., 2 seconds), the protocol asks for the most current Pyth price off-chain; the price is pulled on-chain, and the transaction is conducted.

Staleness thresholds: rejecting price changes that are older than a certain interval. Bind a price update request to a user's transaction, only processing it if it uses the most recent price.

Also take notice that because of gas inefficiency, many other oracles that push pricing on-chain do not update prices every block. The Pythnet approach enables an on-chain stream that can be fetched anytime necessary in real-time off-chain. Like everything else, once it has been applied in reality on numerous chains and more data is available, we will learn more.

With the adoption of SIP-285, Pyth has begun to take steps outside of Solana, with Ethereum protocols such as Euler and Ribbon, and most recently as the oracle provider for Synthetix v2's perps product. Chainlink has a long history with Synthetix, thus this is a crucial connection for Pyth, especially because Chainlink recently announced their new low-latency oracle, which should be available early next year. This will undoubtedly be the most crucial non-Solana integration to watch to see if Synthetix remains with Pyth after the Chainlink solution is online.

In terms of the token, while we know a big portion of the token is reserved for the team and investors, we don't sure where the initial circulating supply will come from. Publishers are likely to receive allocations, as are protocols and (perhaps) users of these protocols. With a 7-year vesting schedule, the unlock period is relatively substantial, ensuring a steady unlock over time. Because initial liquidity is anticipated to be modest, especially given publishers would need to stake tokens, this has the prospective to be a low-circulating/high-FDV token (effectively rendering them off the open market). We should learn more about the first distribution plans shortly, and they are critical since a well-distributed token is critical with the stake-weighting.

pyth tokenomics

Source: Pyth whitepaper

Conclusion

Pyth approaches the Oracle industry from an unusual perspective. This is not a cheap fork or a knockoff of another model, but a truly unique and daring one. They went through some growing pains but have been a lot more efficient subsequently. While they span almost the whole Solana ecosystem, it remains to be seen how effective they will be outside of it and if they will be able to make significant inroads with Pythnet. The Synthetix integration will be an excellent initial test, and Pyth should put it to use. We are witnessing Chainlink’s rival in the making.

Disclaimer:
This article is intended for and only to be used for reference purposes only. No such information provided constitutes advice or a recommendation that any investment or trading strategy is suitable for any specific name.

Pyth Oracle: A Worthy Opponent Of Chainlink?