Reputers
Overview
Reputers verify worker predictions by sourcing ground truth data and calculating loss functions. They stake tokens to participate in topics and earn rewards based on how closely their loss calculations align with network consensus.
Reputers serve as the quality assurance layer, ensuring workers are rewarded accurately for prediction quality.
Core Functions
1. Source Ground Truth
Reputers retrieve ground truth data as specified in topic metadata. For example, for an ETH price prediction topic, reputers fetch the actual ETH price at the specified timestamp from authoritative data sources.
Ground truth must be:
- Retrieved from reliable sources (exchanges, oracles, official APIs)
- Captured at precisely the time specified by the topic
- Formatted according to network data standards
2. Calculate Loss
Reputers apply the topic's loss function to each worker's inference against ground truth. For instance, if a topic uses L1-norm loss, reputers calculate the absolute difference between each worker's predicted ETH price and the actual price.
They submit a ValueBundle of losses (opens in a new tab) containing calculated losses for all worker inferences and forecast-implied inferences.
3. Stake on Topics
Reputers stake tokens on topics to participate. Higher stakes provide greater weight in consensus calculations. Reputers can also receive delegated stake from other network participants, increasing their influence on topic consensus.
Stake serves two purposes:
- Provides economic security for the topic
- Determines influence weight in consensus calculations
4. Earn Rewards
Reputers receive rewards proportional to how close their reported losses are to the stake-weighted consensus. The network calculates a stake-weighted average of all reported losses per topic per epoch. Reputers whose values are closer to this consensus earn higher rewards.
This incentivizes accurate loss calculations that align with other reputers' independent evaluations.
Reputer Workflow
- Monitor active topics requiring evaluation
- Retrieve ground truth data from authoritative sources
- Apply loss functions to worker inferences
- Submit loss calculations to the network
- Receive rewards based on consensus proximity
Technical Considerations
Data Sources
Access to reliable, low-latency ground truth data is critical. For financial topics, use:
- Exchange APIs for price data
- Oracle networks for cross-validated data
- Official sources (central banks, government agencies) for economic indicators
Loss Calculation
Implement loss functions with high numerical precision. Handle edge cases like:
- Missing worker inferences
- Invalid or out-of-range predictions
- Data format mismatches
- Timing precision requirements
Stake Management
Balance stake allocation across topics based on:
- Available ground truth data reliability
- Topic activity and worker participation
- Competition from other reputers
- Your stake capacity and risk tolerance
Deployment
Deploy a reputer node to begin operations, then set and adjust stake on your chosen topics. Query reputer data to monitor performance and rewards.
See the coin prediction reputer example for a practical implementation reference.