Core network terms
- Reppo Protocol: the incentive layer that coordinates AI training data markets through staking, publishing, and voting.
- Datanet: an owner-defined market for sourcing, curating, and monetizing AI training data. See Datanets.
- Pod: an atomic data unit published into a datanet, such as a tweet, image, video, or annotation. Each pod is minted as an NFT on Base. See Pods.
- Data Exchange: the marketplace where Reppo-produced data is packaged, discovered, and monetized downstream. See Data Exchange.
- RL environment: the task, scoring function, and feedback loop used to improve a model. On Reppo, datanets act as market-based RL environments.
- RLHF: reinforcement learning from human feedback. Reppo supports RLHF-style workflows through market-based publishing and curation.
- Stake-Assured Human Feedback: Reppo’s mechanism for turning economic conviction into usable training signal. See Votes & Curation.
Roles
- Publisher: a participant who submits raw data, task output, or model output into a datanet.
- Voter: a participant who locks REPPO for voting power and uses stake-backed judgment to support or oppose what they think is useful.
- Datanet owner: the participant who creates a datanet and defines its rules, fees, incentives, and access controls.
- Data contributor: another name for a publisher.
- Domain expert: a voter whose judgment is valuable in a specific market or task area.
Token and market mechanics
- REPPO: the network utility token used for locking, incentives, and network-level fees. See Token Utility.
- veREPPO: voting power received by locking REPPO. It is allocated across datanets and epochs.
- Locking: the network-level action of locking REPPO to receive veREPPO.
- Staking: datanet-level reward participation. It is separate from locking and separate from publishing.
- Mining: publishing data or task output into a datanet.
- Curation: the process of using stake-backed judgment to rank, support, filter, or oppose submissions and markets.
- Epoch: the 48-hour market window where voting and curation happen.
- Prediction market: the stake-backed market structure Reppo uses to price quality, surface signal, and coordinate incentives.
Rewards, quality, and evaluation
- Performance Pool: the pool that funds datanet staking rewards. It is funded by a share of datanet spin-up fees, publishing fees, and data access fees.
- EVOF: one of the on-chain metrics used to score datanets for reward distribution.
- Publishing fee: the fee paid to submit into a datanet.
- Access fee: the fee paid to consume or access monetized datanet output.
- Learning signal: the useful preference, ranking, or evaluation data produced through market activity.
- Evals: structured evaluation workflows used to test model behavior or model quality.
- Benchmarking: repeated measurement of model performance against a known task or standard.
Verification and provenance
- Reppo Agent: the planned protocol-native agent layer for verification, provenance checks, and validation workflows. See Verification, Provenance, and Reputation.
- Provenance: the record of where data came from, how it was submitted, and how it was curated.
- Reputation: the historical quality and performance record of contributors and datanets.
Whitepaper and legacy terms
- Public and private datanets: V1 language for datanet access models. In V2, datanet creation is permissionless and each owner sets access rules directly.
- Miners: an older or informal synonym for publishers.
- Validators: an older or informal synonym for voters.