Training ARC RL Agents
Last updated
Last updated
There are two primary stakeholders groups in the training of ARC's RL Agents: Sponsors and Players.
Sponsors can create and deploy untrained RL Agents by staking a material amount of $NRN tokens. This staking requirement ensures that RL Agents remain scarce and valuable, encouraging strategic investment and early adoption. In return, Sponsors share in the profits generated by their agents, earning 10% of the rewards from game competitions and campaigns. The limited supply of RL Agents, paired with a structured issuance mechanism, enhances their perceived value and drives demand over time.
Players can contribute their gameplay data to train RL Agents of their choice. Players must first stake or lock $NRN tokens, which generates Data Capsules. These act as containers, allowing players to submit gameplay data to train specific RL Agents. However, creating a Data Capsule doesn’t require immediate data contribution, giving players flexibility in how they engage. Here’s how it works:
By staking $NRN, players create Data Capsule on a 10:1 basis (i.e. 10 $NRN for 1 Data Capsule). These capsules serve as containers that index gameplay data, track contribution metrics, and determine campaign rewards. Players have two options with these Data Capsules:
Passive Staking: If players choose not to contribute data, they still earn a passive reward from a base-level pool that's collected by the ARC Platform per campaign. This pool offers modest rewards for simply staking $NRN, acknowledging the time value of staking and providing a minimum incentive for participants.
Active Data Contribution: When gameplay data is submitted via a Data Capsule, it becomes eligible for campaign-specific rewards, which are tracked and indexed based on the data’s quality and contribution impact.
When players submit gameplay data, it is evaluated by an attribution algorithm. This system ensures that:
High-quality contributions yield higher rewards, encouraging players to submit valuable data that's high in quality and uniqueness. It is important to note that, high quality data does not mean that the player has to be supremely skilled. Poorly skilled players can also contribute valuable and quality data, because "bad" data helps the RL agent learn what NOT to do.
Sybil resistance is maintained, as redundant or undifferentiated data streams are penalized, ensuring the ecosystem benefits from diverse, meaningful contributions.
Each Data Capsule is linked to specific campaigns and tracks its associated returns.
Players can view their contributions and performance in real-time through a personalized dashboard, which displays their data’s relative share, performance points, and accrued rewards. This transparency enables players to monitor the impact and value of their gameplay data.
To claim rewards, players must use the burn | redeem mechanism:
End-of-Campaign Redemption: When a campaign wraps up and rewards are settled, players can burn their Data Capsules to unlock $NRN tokens and claim any accrued rewards.
Early Redemption: Players have the option to burn their Data Capsules at any time to unlock their $NRN, but if they redeem before rewards are issued, they forfeit any potential campaign rewards, which are then reallocated to the $NRN community treasury. This system provides liquidity for players who wish to access their staked tokens sooner while maintaining the integrity of the reward structure.
Rewards from competitions and campaigns are distributed to incentivize all contributors, sustaining the ARC RL ecosystem:
70% to Players (Data Contributors): Players receive the majority share, distributed based on their data quality and performance impact.
10% to Sponsors: Sponsors receive a profit share for their RL Agents, as they provide branding, marketing, and distribution to attract top players.
20% to $NRN Community Treasury: The remaining rewards flow to the community treasury, supporting ARC RL's growth and future development.
10% of the Community share is passed through to $NRN stakers in ARC RL.
This balanced reward structure and burn/redeem mechanism creates a flexible, sustainable system that allows players to participate at varying levels and aligns incentives across the ecosystem.
