# Community Proposal for the Design of the Community Integrity Pool (CIP)

Great proposal! I fully support it. The CIP rewards mechanism is fair and incentivizes high-quality data. It’s a well-thought-out plan that benefits the PythNetwork community.

1 Like

this is a great proposal and i am in FULL support!

it’s an idea that proposes data staking and slashing for data providers with rewards for delegated stakers (individuals like you and me)!! which is why it’s something that benefits the whole pyth community <3

please take your time reading through this. these proposals effect the entire future of the pyth network and it’s important that they don’t get written off. i’m looking forward to a future where the pyth network is able to expand and grow and allow all of us to prosper together

3 Likes

Thanks for the proposal @CMS!

We agree with the overall direction and the formulas are quite nice and well formed!

One downside we see is that there is no mechanism to account for the quality of a publisher’s price feeds and their uptime. It also doesn’t take into account the value of a data feed and its publishers which is fine for now if rewards are purely from the DAO’s treasury, but this might be contentious if protocol fees begin flowing to stakers & publishers.

4 Likes

Glad people are having this discussion.

Despite the fact that this increases the costs and may result in a higher barrier to entry for new publishers, we are generally supportive as this direction appears to benefit the community and token in general.

Would be curious to see the next steps and how we can support.

2 Likes

this proposal very interesting. The majority of Pyth stakers do this just hoping to receive an airdrop. This proposal gives another use to PYTH staking. We must also set a staking limit for each publisher so that there is no imbalance and this will allow each publisher to be more motivated to provide quality and reliable data.

1 Like

this proposal makes sense - as the network grows it is important to increase the accountability of the data publishers and having staking and slashing would achieve that.

but interested to know how would stakers discover/decide on how to stake with given the extensive (and growing) network of data publishers today :thinking:

1 Like

@CMS Thanks for the proposal. Really excited about staking! Can you share the next steps and implementation details?

2 Likes

Hi everyone,

First, I think it’s a great idea to try to find ways to include the $PYTH owned by the community in the equation. Finding ways to attract people is key for the future.

I’ve seen this idea in multiple messages, but in my opinion, this is not a good design. Most $PYTH holders don’t have any idea of the quality of data provider price feeds. So why would we give such power to people who don’t have the necessary information to judge that and will be attracted by marketing and low fees?

I also think:

  1. Helping data publisher to reach a cap and earning rewards for that can be a good idea. But I disagree with the fact to slash the community for bad price feed they have no control on.

  2. I would prefer a staking design based on a pourcentage of the protocol revenues, because:

  • It’s more intuitive for the community (people are learning about Pyth but don’t want to study all the data providers)
  • The community wants to participate or to bet on the $PYTH token because they believe in the evolution of the entire Pyth ecosystem and not on a specific data provider.
  • It’s a win-win

This topic is very interesting but also very complex. I would be delighted to continue discussing it with all of you!
Have a great day. :sunny:

4 Likes

I agree we can end things here. I’ll be answering to your responses first though.

  1. The Publishers won’t rely on staked tokens to maintain price feeds, this is wrong the staked tokens will help incentive them. In the bear market even if people take tokens that shouldn’t be much of an impact as it would be uniform for every publisher. If the rewards are disturbed in a weighted matter this would have ZERO net impact. Publishers can have their minimal stake and they will also appreciate additional stake from the community. Not sure how point of these points being concurrent is confusing you but that’s fine.

  2. Publishers will have their own tokens AT RISK, but they will also have social pressure as there will be delegated tokens for which they are getting fees in their pool too! People/Companies always want more opportunities to get more rewards therefore I disagree here, Publishers with their own tokens on risk + extra delegation fees from stakers rewards will definitely be EVEN more motivated than if it was just their own. It’s an additive therefore this would only improve the security and add to make this system better. Community stake will get publishers more attention and therefore create indirect pressure to produce high quality data so they don’t get negative press etc. The decisions of the PYTH DAO have nothing to do with this so don’t really understand that.

  3. Psychology is the study of human behaviour…I would assume we’re all humans so it is extremely applicable LOL. I think my point on Social Identify Theory was misunderstood here sadly. The delegate staking model combining both Publishers and Stakers is much better than 2 split models. I think unification and working as a collective is way more powerful than a split system, which we clearly disagree on.

I guess we won’t know who’s right either your split model or this collective model is implemented. Time will tell who was right :raised_hands:

Great discussion Carrot it’s been insightful

Hi folks,

Austin Muñoz from Amber Group here.

We are supportive of this proposal - it seems like a good way to get $PYTH holders more directly involved in the success of the protocol by helping to validate the quality of publishers and earn rewards for doing so.

One comment on this section of the proposal:

“Total rewards are bounded by the number of symbols and the target stake per symbol. Adding new symbols requires either an increase in total rewards or a decrease in individual rewards.”

This may discourage the publishing of new symbol prices, which presumably is not an ideal incentive structure for Pyth as a whole. Perhaps rewards could actually be weighted towards new symbols, as long as there are above a certain number of publishers for the given symbol. Said differently, if a publisher begins a price feed on a new symbol, and enough publishers join in, then they will receive additional rewards above what they would normally receive for publishing a long-standing price feed.

The formula could even be based on the number of publishers and subscribers for a given feed (e.g. more subscribers and fewer publishers would lead to higher rewards).

Just my two cents!

3 Likes

Let me just respectfully disagree here as it seems like we both are repeating the same arguments in different models and my brain got a bit tired of saying and hearing prettt much the same things.
Cannot say I agree with your perspectives but I gotta respect it as we both want to see this project growz despite it being in 2 different ways.

I guess it will be up to the community to decide what model they prefer

2 Likes

Rewarding publisher for better data is good.
As a staker myself, I’m not fond of this model.
It is not up to me to decide if a publisher is doing good or bad. That is up to the data.
My 2 cents…

2 Likes

As one of the main data publishing providers, we are generally in agreement with the proposed staking mechanism for Pyth. This approach follows the proven models of delegated proof of stake, enhancing the alignment of interests and security within the network. By requiring publishers to stake tokens and allowing non-publishing stakers to delegate their tokens, the proposal effectively incentivizes high-quality data contributions and ensures accountability.

Furthermore, this mechanism marks a significant evolution for Pyth, steering it towards a DAO-like structure with sophisticated tokenomics. This shift is likely to sustain long-term engagement from data providers, and cultivate a more robust and secure ecosystem. In our view, these changes will substantially bolster the credibility and reliability of Pyth’s oracle services, yielding benefits across the broader DeFi landscape.

4 Likes

Hi everyone, very interesting topic here that caught my attention.
Ivo Galic from Trebuchet.Network here.

Objective

Pyth aims to provide highly accurate and frequent price updates. When publishers are aligned and accurate, the system works smoothly. However, discrepancies in reported prices can occur, and this is where an automated slashing and rewarding mechanism becomes crucial.

Example Scenario

Suppose we have three publishers—A, B, and C—who report prices as follows:

  • Publisher A reports $1.00
  • Publisher B reports $1.50
  • Publisher C reports $0.50

In this scenario, Publisher A’s price is the median. This raises the question: Should Publishers B and C be penalized for deviating from the median, and should A be rewarded?

Proposal for an Automated System

To address discrepancies and ensure fairness, I propose an automated system based on predictive market capabilities. This system would extend the publishers’ responsibilities to include price predictions for a future time (T+N), enhancing accountability.

Predictive Horizons
  • Defining Conditions: Establish conditions for future price validation. For example, if a publisher claims a price will be $1.50 in two hours, this prediction will be checked against actual market data at that time.
Time-Bound Vaults
  • Implementation of Vaults: Create multiple vaults where rewards and penalties are stored based on different time intervals, such as 5 minutes, 2 hours, and 1 day.
    • Example: A 5-minute vault checks short-term predictions, while a 1-day vault assesses longer-term accuracy. Whatever makes sense from our business perspective.
Automated Slashing and Rewarding
  • Automation: Employ smart contracts to automatically manage the slashing and rewarding process based on the accuracy of the predictions relative to the actual prices.
  • Directives: Publishers who consistently predict within a tight margin of error are rewarded, while those whose predictions deviate significantly are penalized.

Benefits of the Automated System

  • Efficiency: Automation reduces the burden on the DAO by handling routine decisions and enforcing rules.
  • Clarity: Clear, predefined rules about predictions and outcomes make the consequences of publishing data more predictable.
  • Fairness: The system treats all publishers equally, basing rewards and penalties solely on measurable accuracy.

Challenges and Considerations

  • Complexity of Implementation: Developing and testing the automated system will require substantial technical resources.
  • Risk of Manipulation: There must be safeguards against manipulation of predictions or gaming of the system.
  • Community Engagement: Even with automation, it’s crucial to maintain community engagement and oversight to ensure the system evolves with the market.

Conclusion

By adopting a predictive, automated approach to manage discrepancies in published prices, Pyth can enhance its reliability and integrity. This proposal aims to make the process more transparent and less dependent on manual oversight, ultimately supporting Pyth’s goal of providing accurate and timely market data.

3 Likes

Thats very interesting proposal that could actually make pyth even more accurate over time.
I believe defining a certain percentage of “allowed” deviation from the eventually published price (depending on the volatility of an asset) might be really good for the porposal so that if a certain publisher “missed” the price by too much for a certain period of time it will be punished.
Also if they were accurate over that defined time period they will be rewarded accordingly

1 Like

Let’s use the scenario of three publishers with their respective stakes and predictions:

Publishers’ Predictions and Stakes

  • Publisher A: Predicted $1.00
  • Publisher B: Predicted $1.50
  • Publisher C: Predicted $0.50

All publishers have 100 PYTH staked in their respective vaults. After 5 minutes, the actual average price is determined to be $1.30.

Calculating Deviations

First, we calculate the absolute deviation of each publisher’s prediction from the actual price:

  • Publisher A: |$1.30 - $1.00| = $0.30 (SHORT)
  • Publisher B: |$1.30 - $1.50| = $0.20 (LONG)
  • Publisher C: |$1.30 - $0.50| = $0.80 (SHORT)

Sorting Publishers by Deviation

  • Least deviation: Publisher B ($0.20)
  • Middle deviation: Publisher A ($0.30)
  • Most deviation: Publisher C ($0.80)

Calculating Slashing and Rewarding

We will now calculate the percentage deviation relative to the actual price to determine the slashing amount. Then, we’ll distribute a part of the slashed PYTH from the most deviating publisher to the least deviating one.

  • Percentage Deviation:
    • Publisher A: $0.30 / $1.30 ≈ 23.08%
    • Publisher B: $0.20 / $1.30 ≈ 15.38%
    • Publisher C: $0.80 / $1.30 ≈ 61.54%
  • Slashing Amounts:
    • Publisher A: 100 PYTH * 23.08% ≈ 23.08 PYTH
    • Publisher B: 100 PYTH * 15.38% ≈ 15.38 PYTH
    • Publisher C: 100 PYTH * 61.54% ≈ 61.54 PYTH

Reward Distribution

Let’s assume we redistribute 50% of the slashing from the publisher with the highest deviation (Publisher C) to the publisher with the least deviation (Publisher B):

  • Redistribution Amount: 61.54 PYTH * 50% = 30.77 PYTH

Net Results After Redistribution

  • Publisher A: 100 PYTH - 23.08 PYTH = 76.92 PYTH
  • Publisher B: 100 PYTH - 15.38 PYTH + 30.77 PYTH = 115.39 PYTH
  • Publisher C: 100 PYTH - 61.54 PYTH = 38.46 PYTH

Conclusion

  • Publisher A ends with 76.92 PYTH, having a moderate deviation but no rewards.
  • Publisher B gains by ending with 115.39 PYTH, benefiting from the least deviation and receiving redistributed PYTH.
  • Publisher C is heavily penalized for the greatest deviation, reducing their stake to 38.46 PYTH.

This system incentivizes accuracy by penalizing those furthest from the actual value and rewarding those closest, promoting a competitive environment where precision is financially rewarded.

2 Likes

Thats for clarifying but I still have a few questions.

What will happen with the other 50% slashed?
What do you think of distributing the rewards between all the publishers who were below a certain deviation percentage (depending on the asset volatility) and punish the others?

1 Like

The slashing and rewarding variables to be adjusted for sure. I think it makes sense to distribute between all the publishers who were below a certain deviation percentage

Reward Distribution Based on Deviation Percentage

The idea to distribute rewards to publishers who are below a certain deviation threshold is excellent. This would not only punish inaccuracy but also encourage consistently accurate reporting. The threshold could be dynamically adjusted based on the asset’s historical volatility and market conditions to ensure fairness and adaptability. For example:

  • Low Volatility Assets: A tighter deviation threshold, as smaller price movements are expected.
  • High Volatility Assets: A looser threshold, recognizing the inherent unpredictability.

Prediction Windows and Stakeholder Incentives

Each stakeholder in the web3 oracle space has different desires regarding the behavior of the oracle:

  • Liquidators might prefer quick, drastic price changes to trigger more liquidations.
  • Market Makers benefit from high volume and volatility, as it provides more opportunities for profit.
  • Market Gamblers (retail traders) generally desire stability to reduce the risk of sudden losses.

Impact on Lending and Borrowing Protocols

Lending and borrowing protocols require highly accurate and stable price feeds to ensure the security of collateral and the proper functioning of liquidations. Erratic or manipulated price data can lead to unfair liquidations or insecure platforms.

Potential for Collusion and Attacks

The concern about “publisher cartels” forming to manipulate prediction outcomes, especially in shorter windows like 5 minutes, is significant. This risk underscores the need for a system that discourages collusion and makes it unrewarding or too risky.

Developing a Reasonable Reward Distribution Model

Given the competitive nature of publishers, a balanced reward system could look like this:

  • Top Quartile Performers: Receive a larger share of the slashed funds, rewarding the most accurate predictions.
  • Second Quartile: Receive a smaller reward, still recognizing their accuracy.
  • Third and Fourth Quartiles: No reward and potential penalties for the least accurate.

Proposal for a New Distribution System

  1. Dynamic Reward Pooling: Based on the deviation percentage, create dynamic pools where publishers compete within tiers. This reduces the likelihood of manipulation as competing within a smaller peer group makes large-scale collusion more difficult.
  2. Staggered Slashing Scale: Instead of a flat rate of 50%, use a staggered slashing scale based on deviation severity. For example:
  • 0-10% deviation: No slash
  • 10-20% deviation: 10% slash
  • 20-30% deviation: 20% slash
  • Over 30% deviation: 30% slash
  1. Reinvestment in Data Integrity Measures: Use part of the slashed funds to invest in tools and technologies that enhance the data verification processes and secure the infrastructure against attacks.

Conclusion

By implementing a tiered, dynamic reward and penalty system, we can foster a competitive yet fair environment that discourages collusion and promotes the overall reliability and credibility of the web3 oracle space. Each stakeholder’s incentives are considered to ensure the oracle remains robust against gaming and manipulation, thereby supporting the broader ecosystem’s health and growth.

2 Likes

Thats actually really good and well thought.
The idea behind this thing is very impressive though I guess the numbers and percentages are still a subject to change but overall theyre in the right area.

Would be great to see it becoming an official proposal in this forum.
I believe that this could also work extremely well with the proposal I made for stakers and voters rewarding/slashing system and grow this project and make it progress even more

1 Like

There have been a number of ways, discussed or implemented, that data providers (“publishers”) can be held accountable:

There’s already tooling available for tracking the timeliness, accuracy (calibration and predictiveness of the aggregate price series), and uptime of data providers: Publisher Metrics.

You can see these metrics for any publisher by clicking on any of the “price components” for any product in the price feeds catalog e.g. “BTC/USD”

In fact, one of the older Pyth Network whitepapers also discussed two ways of measuring or thinking about a publisher’s data quality (and therefore, how to calculate its rewards or penalties).

A quality score measuring how well the publisher’s price series predicts future price changes.

This score would measure how well a publisher’s price series predicts future changes in the aggregate price by training an online regression model that predicts the future price from several features of the publisher’s price series.

The calibration of the publisher’s confidence intervals.

This score would measure whether the publisher’s confidence interval accurately represents their uncertainty, and interprets the confidence interval as the standard error of a Laplace distribution within which the publisher expects to find the aggregate price.

In other words, one could optimize for rewarding publishers with more predictive price series and better-calibrated confidence intervals. And perhaps negative scores of quality, calibration, etc. could be used to determine…negative rewards? (i.e. penalties).