Announcing Ocean Data Farming
A program to incentivize a supply of relevant and high-quality data assets #datafarming
Overview
Ocean Data Farming is a program to incentivize a supply of relevant and high-quality data assets into Ocean Protocol.
Ocean data farmers can earn OCEAN for providing liquidity to OCEAN-datatoken pools; the amount earned will be multiplied by usage of the datasets they stake on, and more.
We are announcing Ocean Data Farming now in order to give a thorough picture of OCEAN token design leading up to the public Ocean V3 release. We will provide a launch date after Ocean V3.0 has been publicly released and hardened further.
Ocean Data Farming aims to maximize the data supply reward function (RF). The RF is a function of liquidity added to datatoken pools, dataset usage, and potentially more. We will start with an initial RF and a program budget of OCEAN over a time interval (on the order of months). Each week at a pre-set time, the program managers will:
- Off-chain, calculate contribution of each actor to the RF for that week (actor = Ethereum address).
- Manually airdrop OCEAN to each actor proportional to their contribution, the total being that week’s budgeted OCEAN.
- Declare the RF for use in the following week, to fine-tune towards the most relevant and high-quality datasets, and away from “gaming” that is against the spirit of the program.
During the week, the program managers will:
- Gather feedback from the community on possible ways to improve the RF, using soft signaling in Discord or similar [1]. Later generations may use OCEAN token voting.
This design allows the program managers to start simple, then to quickly learn and improve the RF every week. This design draws inspiration from the successful Balancer Liquidity Mining Program [2]. If this initial pilot is successful, it may be repeated (potentially for a much longer time scale).
For the pilot, OCEAN will be drawn from the treasury of Ocean Protocol Foundation (OPF). Future versions may draw from OPF treasury as well as OceanDAO funds.
Reward Function
This section describes a tentative initial RF. It may change before the program initiates.
RFij is the (non-normalized) contribution for actor i on datatoken j:
where:
- Sij = actor i’s OCEAN stake in data asset j = (actor’s # BPTs in datatoken j’s pool / total # BPTs in pool) * (# OCEAN in pool) [2][3][4]
- Dj = # times data asset j has been consumed in the last week (= # consume transfers to the datatoken j’s Provider)
The first term is log10(Sij). It reflects the actor’s belief in the relevance, quality, or potential usage of the datatoken. This incentivizes Providers to publish relevant data assets. We use log10 to level the playing field with respect to OCEAN whales; and so that OCEAN whales are incentivized to make a greater number of data assets available. This has theoretical roots in Kelly Betting: applying the log is the optimal strategy for an individual to maximize their utility in betting. The +1 is to avoid negative values for Sij=0.
The second term, log10(Dj), is a direct measure for the usage of the data asset, and a proxy for its relevance or quality. We use log10 to incentivize actors to focus on usage of a greater number of assets. The +1 is to ensure there’s a reward even if the dataset is used once in a given week, and to avoid negative values for Dj=0.
This RFij can be summarized as a binding of predicted popularity * actual popularity in terms of their orders-of-magnitude. As discussed, this is the initial RF. We anticipate feedback from the community to further improve it.
To help long-term thinking and drive value to the protocol, we are considering lockup periods to ensure that rewards are not immediately sold by traders [ref]. It could even be “longer lockup, longer reward” like in Filecoin’s ICO or xDai Chain’s staking incentives. There may be other time-related criteria.
Next Steps
#datafarming is coming, stay tuned:)
Appendix
Astute readers may recognize this design from the 2018 and 2019 versions of the Ocean whitepaper, to maximize the supply of relevant data. That’s correct. There are two main differences. We’ve changed it from smart contracts in the core of the system, into a program to catalyze growth run manually and off-chain for rapid learning and iterations. Also, the rewards formula has a simpler implementation, due to using staking inside datatoken pools and datatoken transfer calls.
Further Reading
- Here’s the tweetstorm summarizing this post.
- “Ocean Protocol V3 Posts: Links to all V3-Related Stories” [link]
Acknowledgements
Thanks very much to the following people for reviews: Sarah Vallon, Monica Botez, Manan Patel, and Bruce Pon. Thanks also to Fernando Martinelli, Julien Thevenard, and Simon de la Rouviere for the discussions that greatly influenced this approach.
Notes
[1] E.g. giving a thumbs-up or thumbs-down to questions posted in Discord.
[2] Balancer Labs is a longtime collaborator with Ocean Protocol.
[3] BPT = Balancer Pool Token. People receive BPTs proportional to the OCEAN or datatoken liquidity they add to a pool. Conversely, when they remove liquidity, they are essentially selling BPTs for OCEAN and datatokens.
[4] In the first iteration, rewards will only go to liquidity in Ocean-customized Balancer pools. Over time, we anticipate this to expand to other exchanges. Ocean-customized Balancer pools are optimized for low gas cost, but because they do not come from Balancer’s official factory, they aren’t currently eligible for Balancer Liquidity Mining. Balancer V2 will optimize gas costs, at which point we expect Ocean libraries to use Balancer official factory (and therefore be eligible for BAL rewards in addition to OCEAN data farming).
