# Introduction ### We're building market infrastructure for AI training data Reppo is a decentralized network for sourcing, curating, and monetizing AI training data. It supports labeling, annotation, RLHF workflows, evals, and market-based RL environments. The Reppo network is organized into domain-specific prediction markets called Datanets. Each Datanet coordinates two core roles: * **Publishers** submit raw data, task output, or model output. * **Voters** lock REPPO for voting power (veREPPO) and use stake-backed judgment to curate quality. Datanets can set their own access rules, publishing fees, incentives, and monetization model. Need a quick reference for network and whitepaper terms? See [Glossary](/reppo-labs/resources/glossary.md). #### What is an RL environment? An RL environment defines the task, scoring function, and feedback loop used to improve a model. It is the setting where a model acts, gets evaluated, and learns. On Reppo, datanets act as market-based environments for generating and curating that learning signal.
> ## **“Scale AI coordinates training data pipelines with contracts. Reppo coordinates them with markets, access controls, and on-chain incentives.”** Publishers pay to submit into a datanet. That makes publishing an economic decision, not a free action. Voters lock REPPO to receive veREPPO, then allocate that voting power across datanets and epochs. Their votes help surface what is useful and suppress what is weak. By putting stake behind both contribution and curation, Reppo is designed to address the AI training data trilemma.
**AI Training Data Trilemma**
#### Reppo's approach [**Data Exchange**](/reppo-labs/roadmap/data-exchange.md) is the marketplace for discovering, packaging, and monetizing data produced through Reppo. Datanet owners can keep outputs private, share them selectively, or monetize them through downstream distribution. The network is built on Reppo Protocol, an incentive layer for crowdsourcing [**Stake-Assured Human Feedback**](/reppo-labs/protocol-mechanics/stake-assured-human-feedback.md). The go-to-market focus is AI training data on demand. The same mechanism can also support evals, benchmarking, and other markets where the scarce input is human preference. Each datanet acts like an on-demand data business. It lets data owners, contributors, curators, and buyers coordinate around the same market. For crypto-native readers, Reppo can be understood as a market network for AI training data. The mechanism, token design, and role structure are specific to Reppo. #### What Reppo is not Reppo does not assume capital is the same thing as expertise. Stake prices conviction and accountability. Expertise still comes from who a datanet allows in, what task it defines, and how the owner sets quality standards. That is why each datanet can set its own access rules. Open markets fit some tasks. Restricted, expert-only markets fit others. Reppo is also not a claim that every annotation workflow should become a public token market. It is a coordination layer for markets where human judgment is scarce, valuable, and worth pricing continuously. For high-stakes or specialized workflows, datanet owners can limit participation, define stricter standards, and keep outputs private. See [Datanets](/reppo-labs/protocol-mechanics/datanets.md) and [Stake-Assured Human Feedback](/reppo-labs/protocol-mechanics/stake-assured-human-feedback.md). #### Current gaps in AI training data Traditional data pipelines still reward speed more than judgment. That often leads to: * rushed work * weak accountability * shallow labels * expensive QA Reppo replaces fixed-task coordination with stake-backed markets. Publishers bring supply. Voters bring curation, ranking, and feedback. That creates a stronger feedback loop between usefulness, incentives, and downstream demand. #### Why stake does not make consensus circular Reppo does not treat the first large holder as ground truth. Voting is continuous, not one-shot. Participants can vote both for and against. Voting power also decays during the epoch, which rewards early discovery over late herd-following. That means support can be challenged as new information appears. Weak markets still face a real early-stage risk when liquidity is thin. The design goal is not perfect truth by decree. It is faster repricing toward useful signal under economic pressure. See [Adversarial Robustness](/reppo-labs/protocol-mechanics/adversarial-robustness.md). #### Current stage of proof The mechanism is live on mainnet and the core economic loop is operating today. That does not mean every claim is fully proven at scale yet. Broader benchmarking, stronger quality metrics, and third-party validation still matter. Those are explicit product and roadmap priorities, not hidden assumptions. See [How Reppo Works](/reppo-labs/foundations/how-reppo-works.md) and [Product Roadmap](/reppo-labs/roadmap/product-roadmap.md). **Reppo is not an L1 or L2 network yet.** --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://reppo-labs-xyz.gitbook.io/reppo-labs/foundations/readme.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.