Ocean Data Farming is Launching

Original Source Here

Ocean Data Farming is Launching

216M $OCEAN program to unlock an Ocean of data, with up to 125% APY

1. Introduction

Ocean Data Farming (DF) is launching. DF incentivizes for growth of data consume volume in the Ocean ecosystem. To start, it rewards OCEAN to pool liquidity providers as a function of consume volume and liquidity. It’s like DeFi liquidity mining, but tuned for data consumption. The aim is to achieve a minimum supply of data for network effects to kick in, and once the network flywheel is spinning, to increase growth rate.

DF has these phases:

  • DF Alpha. Counting starts Thu June 16. 10K OCEAN are rewarded at the end of every week, for the activity of the previous week. It runs 4 weeks. The aim is to test technology, learn, and onboard data publishers.
  • DF Beta. Counting starts Thu July 14. Rewards are up to 100K OCEAN per week. It runs up to 20 weeks. The aim is to test the effect of larger incentives, learn, and refine the technology. Expected APY is 125%.
  • DF Main. Immediately follows DF Beta. Rewards are up to 718K OCEAN per week. It runs for decades; at least 215.7M OCEAN total is committed. Expected APY is 125% over many months (once fully ramped), staying generous over the long term.

The rest of this post is organized as follows. Section 2 describes the reward function. Section 3 tells how to earn. Section 4 elaborates on DF phases, and future evolution. Section 5 does a walk-through of OCEAN emitted and expected APYs. Section 6 concludes. The appendices have further details on the reward function, and more.

It’s time to farm.🧑‍🌾🚜

2. Reward Function Info

2.1 Who Receives Rewards

Pool liquidity providers (LPs) receive rewards based on the reward function.

2.2 Reward Function

The reward going to an LP for a given pool depends on LP’s liquidity and how much that dataset is being consumed. Here is the reward function:

RFij= Sij * Cj


  • RFij = relative reward going to LP i in pool j
  • Sij= LP i’s relative stake in pool j = (LP’s $-value stake in pool j) / (total $-value stake in pool j). We express in $ terms, to allow for for pools with OCEAN or H2O as basetoken.
  • Cj = consume volume of data asset in pool j, spread among pools with that asset = (price of datatoken j in $ terms) * (# of consumes during the week) / (# pools with that data asset)

Actors receive rewards pro-rata according to RFij. There is an upper bound of OCEAN rewarded each week according to the max released per phase, and such that APY ≤ 125%. The appendix contains details.

There are rewards only on pools where the data has consume volume. The higher the consume volume, the more the rewards. The more you stake, the more the rewards. If you want to max out rewards, create or find pools that people find useful and consume.

2.3 Pools that Qualify

It’s data pools that receive DF rewards. A data pool contains Ocean datatokens for given data service. That data service may be of any type — dataset (for static URIs) or algorithm for Compute-to-Data.

To qualify for DF, a data pool must follow these criteria:

  • Asset & pool must have been created by Ocean smart contracts, deployed by OPF to production networks
  • Asset & pool must be visible on Ocean Market
  • Pool must use OCEAN or H2O as basetoken
  • Asset can’t be in purgatory

3. How to Earn in DF

To earn, you need to (1) plant the seeds, ie provide liquidity, and (2) harvest, ie claim rewards. This section describes elaborates.

Left: seeding. Right: harvesting

3.1 How to LP

To provide liquidity, towards earning DF rewards:

1a. Go to Ocean Market webapp

1b. Find data assets that have consume volume (or that you expect will get consume volume)

1c. Add liquidity

Where to add liquidity from Ocean Market webapp.

3.2 How to Claim Rewards

At the end of the weekly cycle (Thursday), the DF core team computes rewards for LPs using this code. Rewards are sent to the claims contract, deployed to each Ocean-supported network (Eth, Polygon, etc).

As an LP, here’s how to claim rewards:

2a. Go to DF webapp Claim Portal

2b. Connect your wallet

2c. For each network: select network, click “Claim”, sign the tx, get rewards:)

Rewards will accumulate over weeks, so you can claim rewards at your leisure. If you claim weekly, you can re-stake your rewards for compound gains.

Claim Portal of DF Webapp. It’s linked directly from oceanprotocol.com.

4. DF Phases & Evolution

This section elaborates on DF phases, and how DF will evolve.

4.1 Phase: DF Alpha

This phase distributes modest rewards. It’s low because the primary aim is for everyone to learn:

  • For the DF Core team to identify issues in the DF process, and resolve them before the stakes are high.
  • For data publishers to publish datasets (and pools), and finding consumers of their data.
  • For Data Farmers — anyone staking in an eligible pool — to learn more about Data Farming process, and to observe how rewards are received.

4.2 Phase: DF Beta

DF Beta aims to test the effect of larger incentives, learn, and refine the technology.

It will run at least 10 weeks, and up to 20 weeks. Rewards are at least 10K OCEAN per week, up to 100K OCEAN per week.

The DF Core Team will decide the amount each week, based on learnings so far. The aim is allocate OCEAN efficiently to activities that drive data consume volume in a beneficial way, rather than to mal-intentioned actors gaming the system. We hope to ramp to 100K OCEAN quickly and to complete in minimal time; but let’s see.

4.3 Phase: DF Main

DF Main is designed to give an APY of 125% until the amount staked gets super-high, with a cap according to the Bitcoin-like curve described below. An annual percent yield (APY) of 125%, means a weekly percent yield (WPY) of 1.57171%. Therefore total DF rewards distributed per week = min(curve_amount, total_staked * WPY).

Here’s where the funds come from. OCEAN total supply is 1.41 billion tokens. Ever since Ocean was launched, 51% of total OCEAN supply (719M OCEAN) has been earmarked for the Ocean community; 30% of this will go directly to DF Main (=15.3% of total supply, or 215.7M OCEAN).

The baseline emissions schedule is like Bitcoin, including a half-life of 4 years. Unlike Bitcoin, there is a burn-in period:

  • The curve initially gets a multiplier of 10% for 0–12 months
  • Then, it transitions to multiplier of 25% for 0–6 months
  • Then, a multiplier of 50% for 0–6 months
  • Finally, a multiplier of 100%.

Each transition is made under control of a multisig by OPF.

The burn-in period ratchets up value-at-risk over time. It gives us time to learn and for future scope to get built, before full-blown emissions.

4.4. DF Evolution

With each weekly cycle, the DF core team may tune the Reward Function or make other changes based on learnings.

After DF Main is shipped, the DF core team will expand scope. Current plans are:

  • DF Partners Program — While other DF rewards come from the baseline OCEAN budget, these are additional token rewards coming from Ocean’s collaborators: H2O stable asset, storage networks, L1 chains, and more. APY will sum across these; it may go much higher than the target 125% from baseline OCEAN.
  • Add rewards for fixed-price data.
  • Add rewards for free/open data.
  • DF Crunch ProgramKaggle-style data science competitions. Initially, there will be a weekly reward to the best predictor of OCEAN.
  • Over time, DF changes will slow. Then, we will move to automate DF further towards full decentralization.
Image: #McConaghyFarms

5. Walk-Through Numbers

This section walks through example numbers. The first subsection describes OCEAN disbursement with the most aggressive possible schedule; the subsection after describes a conservative schedule. The likely scenario is somewhere in between. A third subsection describes possible APYs.

5.1 OCEAN Schedule with Aggressive Ramp

Here’s a scenario that goes through DF the most aggressively: DF Beta takes 10 weeks and emits 100K OCEAN per week, and each multiplier in DF Main time takes 0 time (goes to 100% multiplier immediately). All plots are computed from this Google Sheet.

The image below shows the first half-year for the aggressive-ramp scenario. The y-axis is OCEAN released each week. It’s log-scaled to easily see the differences. The x-axis is time, measured in weeks. We can see the distinct phases for DF Alpha, DF Beta, then DF Main.

OCEAN released to DF per week — first 0.5 years, aggressive ramp

The image below is like the previous, but now shows for the first five years. DF Main starts at week 14 with full-blown OCEAN emissions. Recall that DF Main emission follows a Bitcoin-style curve where the rewards decrease according to an exponential with a 4-year half-life. With each DM main multiplier taking 0 time, it means that half of all the DF Main rewards are distributed in its first 4 years.

OCEAN released to DF per week — first 5 years, aggressive ramp

The image below shows the total OCEAN released by DF for aggressive-ramp scenario. The y-axis is log-scaled to capture both the small initial rewards and exponentially larger values later on. The x-axis is also log-scaled so that we can more readily see how the curve converges over time.

Total OCEAN released to DF — long term, aggressive ramp

5.2 OCEAN Schedule with Conservative Ramp

Here the opposite-extreme scenario, having a highly conservative ramp of DF rewards. DF Alpha runs at 10K OCEAN per week, over four weeks. DF Beta runs at 10K OCEAN per week, over 20 weeks. DF Main starts at the 10% multiplier, transitions to 25% at 12 months, to 50% at 18 months, and to 100% at 24 months. The image below shows the first five years.

OCEAN released to DF per week — first 5 years, conservative ramp

The image below shows the total OCEAN emitted for DF. The y-axis is log-scaled to capture both the small initial rewards and exponentially larger values later on. The x-axis is also log-scaled so that we can more readily see how the curve converges over time.

DF Beta starts at week 8, where the curve gets steeper. DF Main starts at week 20, where the curve is again more gradual. The curve gets steeper with each multiplier ratchet at weeks 48, 72, and 98, finally arriving at a shape just like the Bitcoin exponential curve.

Total OCEAN released to DF — long term, conservative ramp

5.3 Example APYs

The table below explores possible APYs, calculated from this Google Sheet. We assume 2.2M OCEAN staked in week 0, OCEAN staking growth initially 20% per week, growth rate reduces by 5% relative per week, and aggressive ramp.

Here are the highlights. In the first week that DF gets generous, APY is 125%. It stays pegged at 125% from week 8 until week 27. After that the amount of OCEAN staked has become sufficiently high that OCEAN released is the limiting factor to APY. APY gently reduces over time, to 66% one year in, and is still at 21.68% five years in. This APY does not account for other positive-APY factors such as OCEAN price rising; or negative-APY factors such as impermanent loss.

Expected APYs (aggressive ramp)

6. Conclusion

Ocean Data Farming incentivizes for data consume volume in the Ocean ecosystem. It gives expected APYs of up to 125%, and is designed to last for many decades.

Seed, harvest, then fill your bins. Image: #McConaghyFarms

7. Appendix: Who are Data Farmers

A Data Farmer is anyone who adds liquidity (stake) to a qualified data pool, to receive rewards.

Various actors in the Ocean ecosystem are therefore Data Farmers, each with their own unique advantages.

  • Data Publishers can be Data Farmers. You’ve published a dataset, added liquidity, and got the dataset to be qualified for DF. Now, your liquidity will count for DF rewards. Your advantage is your understanding of how much a dataset might get consumed, as you published the dataset.
  • Data Curators / Stakers can be Data Farmers. You don’t have to publish data or consume data to get DF rewards. Just simply stake on the datasets eligible for DF rewards that you believe will offer you the best returns, ie datasets with good consume volume and where you can get a decent proportion of liquidity.
  • Data Consumers can be Data Farmers. If you only bought and consumed a dataset without adding liquidity, you’re not eligible for DF rewards. However as by consuming the dataset, you know its usefulness better than most, and are therefore well positioned to predict its consume volume. You can capitalize on this knowledge by staking on the most promising and eligible datasets, making you eligible for DF rewards.
  • Data Traders can be Data Farmers. If you’re just buying and selling data assets without staking, you’re not eligible for rewards. However if you’ve worked to understand the relative value of datasets enough, you can capitalize on this knowledge by staking on the most promising eligible datasets, making you eligible for DF rewards.

Basically, we can all be Data Farmers. 👩‍🌾

You: Data Farmer. Image: #McConaghyFarms

8. Appendix: Reward Function Details

8.1 Candidate Reward Function

This section describes the reward function — a measurable formula that each actor can individually optimize for.

The initial reward function is given in the next subsection. Here’s the thought process that led to it.

What if the reward function was simply for the number of datasets added? Here’s the problem: people would create large numbers of random or garbage datasets. This doesn’t add value, of course.

What if it was the trading volume of a dataset, measured in OCEAN or $? This has two problems. First, most volume is speculative. This isn’t real value creation from a data perspective, it’s just money sloshing around in a (nearly) closed chamber. Furthermore, high trading volume of a token may not be a great proxy for value creation of that token, as we’ve seen with several scam-coins in crypto.

Where is the value creation for data? We’ve arrived at: it’s when people are actually consuming a data asset. Here’s an example. Researchers may have data for thousands of people; for each person there is genetic information, age, and whether the person has Parkinson’s. From this, the researchers build a model mapping genetic information and age to the chance of getting Parkinson’s. Then general practitioners use this model for early-stage screening. The model can report the chance of someone getting Parkinson’s; if it’s high they can take measures to slow the onset. This is an example of value creation, via consuming data. There are countless examples across many verticals.

We’ve established that consuming data is useful to aim for. What should be the specific measure? A first approach is the number of consumes of a data asset in a time period. But that doesn’t capture that some assets may be far more expensive than others. For example, compare a $1 dataset is bought and sold 1000 times, versus a $1000 dataset bought and sold 1000 times. Under this measure, all else being equal, both datasets would lead to the same reward. This is not desirable.

8.2 Towards a Chosen Reward Function

Better yet is consume volume, expressed in $ or OCEAN. Ocean Market uses OCEAN so we will focus on that. Then: consume volume = (# times a data asset was consumed in a time interval) * (price of that data asset, in OCEAN). This accounts for some assets being more valuable than others, yet avoids speculation volume.

Thus, consume volume (in OCEAN) is the key measure to incentivize for.

We can refine the incentive further. Ocean V3 introduced datatoken pools. These enabled automatic discovery of the price of data, and helped curation of data (higher liquidity in a pool is a useful signal). More liquidity in a datatoken pool means lower slippage when buying datatokens. In short, more liquidity is good.

Therefore liquidity of datatoken pools (in OCEAN) is the secondary measure to incentivize for.

8.3 Chosen Reward Function

We combine consume volume and liquidity into an overall DF Reward Function.

RFij= Sij * Cj

Precise definitions of each term are given in the “Reward Function” section earlier.

  • Sij is LP’s stake. It reflects the actor’s belief in the relevance, quality, or potential usage of the datatoken. This incentivizes Providers to publish data assets that they believe will be used, and stakers to try to identify useful assets early.
  • Cj is data consume volume. It measures the usage of the data asset, and is a proxy for its relevance or quality.

This RFij can be summarized as a binding of predicted popularity * actual popularity (stake * volume).

We considered log() of each term, to incentivize for a variety of assets. Alas, these causes Sybil attack problems.

9. Appendix: Wash Consume Attack

Wash trading is where a malicious actor trades with themselves to fake trading volume, in cases that they are rewarded for high trading volume.

The equivalent in Ocean is wash consuming, where people buy and consume datasets for themselves, to get rewarded for high data consume volume. There are many possible ways to combat this; most require manual intervention.

Our solution is: knowing that there’s a maximum x% APY, then set a rule: for a data asset to be eligible for DF, it must have a % consume fee of y%. Our current calculations put this value around 3%; therefore this is the value of Ocean Community Order Fee. This fee goes to the Ocean community every time a data asset is consumed.

10. Appendix: Why Now?

Q: DF was first announced right after Ocean V3 launch, more than a year ago. Why is it only shipping now?

Here’s the answer. A pre-requisite for DF was for V3 to be hardened / stabilized further. Then, rug pulls happened in V3. Staking wasn’t safe enough to incentivize on with DF.

OceanONDA V4 One-Sided Staking solved this. It went into production yesterday (June 8, 2022). Now, we’re ready for Data Farming.

Let’s go 🚜🚜🚜! Image: #McConaghyFarms


Trending AI/ML Article Identified & Digested via Granola by Ramsey Elbasheer; a Machine-Driven RSS Bot

%d bloggers like this: