Reppo is building a decentralized evaluation layer for AI, where domain experts stake real capital on whether training data is good. Right calls earn tokens, wrong calls lose them. The project launched on Base in late 2025, secured a $20 million commitment from Bolts Capital in April 2026, and just shipped an Evaluation API that charges a penny per verdict. The thesis is compelling. The token, down 70% from ATH, tells a more complicated story.
Project Overview
AI training data is broken. The $30 billion labeling industry pays workers by the hour, not by accuracy. Annotators get the same whether they nail a subtle domain judgment or rush through without really analyzing. Reward hacking is now a documented research problem, not a fringe concern. Models learn to exploit verifier blind spots instead of solving the actual task.
Reppo’s answer is RLEF, or Reinforcement Learning through Economic Feedback. Instead of hourly annotators, you get prediction markets. Domain experts stake REPPO tokens on whether data meets quality standards. The market settles every 48 hours. Winners split emissions. Losers forfeit stake. The curated data becomes a continuously updating pipeline that AI teams subscribe to.
The project started as a Data DAO modeling storage provider reputation for Filecoin Plus. Co-founder RG Rmadya was building ML models at Protocol Labs to detect incentive abuse on Filecoin when he realized the training data bottleneck was the real problem. Reppo spun out as an independent protocol and launched on Base in November 2025.
How It Works
Datanets
The core unit is the Datanet, a permissionless prediction market anyone can spin up for any data domain. Each Datanet is an NFT with its own economics, access rules, and quality standards. Publishers submit raw data (called Pods) and pay a fee. Voters lock REPPO to receive veREPPO (voting power), then trade opinion contracts on whether each Pod is good or junk.
Voting power decays linearly during each 48-hour epoch. Early votes carry more weight than late ones, rewarding discovery over herd-following. Voters can vote both for and against, so support can be challenged as new information arrives.
Twenty-one Datanets are currently live, spanning geopolitical misinformation detection, trading strategies, robotics video annotation, and code intelligence. The largest by volume is Geopolitical Flashpoint and Misinfo Detection, with 284 million REPPO in agreements.
On September 25, 2026, Reppo launched its Evaluation API, the product that could actually generate revenue. Teams submit an AI output and receive a structured verdict from independent judges: a score, a decision (accept, revise, reject), confidence level, and written critique. Pricing is $0.01 per evaluation with 10 free to start. The API is already integrated with UsePod, using inference from that marketplace to power settlements.
Orquestra, the agentic swarm framework, automates Datanet participation. Autonomous agents publish data, stake REPPO, vote on quality, and claim emissions on behalf of operators. The GitHub repo had its last commit on September 18, 2026.
Team and Funding
The team is small but pedigreed. RG Rmadya, the co-founder, previously built reputation systems at Protocol Labs and Filecoin. Other contributors come from AWS, NHS UK, Manulife, Uniswap, and Celestia. Core contributors listed on the vision page include Jordan Grollman, Rohan Bhaskar, Shasika, and Ana Julia Bittencourt.
The Reppo Foundation secured a $20 million strategic commitment from Bolts Capital on April 23, 2026. Previous backers include Protocol Labs (where the project originated), CMS Holdings, MH Ventures, CV VC, and Charlie Songhurst. The project is registered in the Cayman Islands.
The GitHub organization has 8 public repositories with 74 followers. The main Orquestra repo has 2 stars and 3 contributors, which is thin for a project claiming to build infrastructure. The commit cadence is healthy (last push September 18), but the low star count suggests limited developer attention outside the core team.
Tokenomics Design
REPPO is a Base L2 token with a fixed maximum supply of 1 billion tokens. The token serves three roles: governance (lock REPPO for veREPPO voting power), bootstrapping Datanets (seed incentive pools), and fee payment (creation and publishing fees).
Datanet creation requires locking 20,000 REPPO. Fifty percent stays locked while the Datanet is live; the other 50% goes to the network. Non-REPPO token emissions face a 15% tax routed to the Foundation treasury, creating structural buy pressure for REPPO. Ten percent of all fees are burned, adding deflationary pressure.
The detailed allocation breakdown is published as a chart in the docs covering community, ecosystem, liquidity, team, investor, and advisor categories. Specific percentages are not provided in text form.
REPPO Supply Breakdown
| Metric | Value |
|---|---|
| Circulating Supply | 428,535,184 REPPO (42.9% of max) |
| Total Supply | 1,000,000,000 REPPO |
| Max Supply | 1,000,000,000 REPPO (capped) |
Source: CoinGecko API
Market Data
REPPO is down 10.6% in the last 24 hours but up 9.6% over the past week. The token is trading 70% below its all-time high of $0.0513, set on January 14, 2026, and 438% above its all-time low of $0.00282 from November 21, 2025. The price has been range-bound between $0.01 and $0.02 for several months, suggesting the market is still searching for a fundamental floor.
The volume-to-market-cap ratio is 2.2%, which is low. Most trading happens on Aerodrome and Baseline (Base), with smaller CEX presence on Ourbit and Kraken. Liquidity is concentrated in a single Aerodrome pool with $815,512 in liquidity.
Token Metrics as of September 28, 2026
| Metric | Value |
|---|---|
| Price | $0.0152 |
| Market Cap (circulating) | $6.49M |
| Fully Diluted Valuation | $15.15M |
| 24h Volume | $140,680 |
| Top Pool Liquidity | $815,512 |
| Holders | 29,160 |
Sources: CoinGecko, DexScreener, Basescan
Price Performance
| Period | Change |
|---|---|
| 24h | -10.6% |
| 7d | +9.6% |
| 30d | +5.3% |
| From ATH | -70.4% |
| From ATL | +438.2% |
Source: CoinGecko API
Top Exchanges by 24h Volume
| Exchange | 24h Volume | Type |
|---|---|---|
| Aerodrome Slipstream 2 | $56,458 | DEX |
| Baseline (Base) | $54,284 | DEX |
| Ourbit | $20,901 | CEX |
| Kraken | $18,173 | CEX |
| Uniswap V3 (Base) | $13,290 | DEX |
Source: CoinGecko API
Social Sentiment and Community
Reppo’s verified X account has 28,700 followers. The project joined X in late 2025 and has maintained consistent posting cadence. Recent posts focus on the Evaluation API launch and partnership announcements, with engagement ranging from 50 to 140 likes per post.
The Telegram group has 6,069 members with 235 online. The Discord server has 1,007 members with 94 online. These are modest numbers for a project with a $6.5 million market cap, but the community appears genuine rather than bought.
The ReppoStats dashboard shows 173,370 total users who have participated in the network, with 32 new users joining in the past week. Total voting volume across all Datanets is 881 million REPPO (approximately $14.17 million USD). The community sentiment indicator reads “Bullish” with a +44 score.
Dev wallet transparency is notable. The team has locked 118.09 million REPPO (active) and burned 4.98 million REPPO across 15 on-chain transactions, all verifiable on Basescan. This is above-average transparency for a micro-cap project.
Recent Developments
The Evaluation API launched on September 25, 2026, marking Reppo’s transition from a protocol-level thesis to a product with a clear pricing model. The team is working with UsePod for inference-powered settlements and plans to distribute the API through tokenized inference partners.
The Sherwood Trading Strategies Datanet, running on Robinhood’s chain, crossed 40 million in voting volume, making it the top Datanet by activity on that network. This integration with Robinhood represents an unusual bridge between crypto-native prediction markets and retail trading infrastructure.
Reppo’s roadmap targets 100+ Datanets by mid-2026 and $1 million in cumulative subscription revenue by end of 2026. The team also plans to launch Datanet staking, where REPPO holders can stake on individual Datanets to earn a share of their fees.
Competitive Landscape
Reppo sits at the intersection of decentralized data labeling and prediction market infrastructure. The closest analog is Scale AI, the $14 billion centralized data labeling company. Reppo explicitly positions itself as “Scale AI on chain that scales like Stripe.” The difference: Scale AI uses managed annotator pools. Reppo uses staked prediction markets where quality is priced continuously.
Bittensor takes a different approach to decentralized AI, using subnet-based mining incentives for model training rather than data curation. Both projects use decentralized incentive mechanisms for AI, but Bittensor focuses on model production while Reppo focuses on data curation.
Traditional prediction market platforms like Polymarket focus on event outcomes, not data quality. Reppo’s Datanets are structurally different because each market produces a usable training dataset as a byproduct of the price discovery process.
Strengths and Risks
Strengths:
- Genuine product-market fit thesis: the AI evaluation problem is real and growing
- Evaluation API with clear pricing ($0.01/eval) creates a potential revenue path independent of token speculation
- Strong institutional backing from Protocol Labs, Bolts Capital, and Filecoin ecosystem
- On-chain transparency: dev wallet locks and burns are verifiable on Basescan
- Active development: last GitHub commit September 18, 2026, with regular feature shipping
Risks:
- Token is down 70% from ATH with thin volume and low liquidity depth
- GitHub engagement is minimal (2 stars, 3 contributors) for an infrastructure project
- The 1 billion max supply means 57% of tokens are still unlocked, creating future sell pressure
- 24h volume of $140K is barely above the threshold for reliable price discovery
- The $20M from Bolts Capital is a “commitment,” not a verified transfer; structuring details are unclear
Analysis and Outlook
Reppo is building something genuinely interesting. The idea that prediction markets can produce better training data than hourly annotators is not crazy. It is grounded in real research on reward hacking and verifier gaming. The Evaluation API is the right product move: it turns an abstract protocol thesis into a billable service with transparent pricing.
But the gap between thesis and traction is wide. Twenty-one Datanets with $14 million in cumulative voting volume is early-stage. The Evaluation API launched three days ago. There are no published revenue numbers. The token trades at $6.5 million market cap with 70% of supply still locked, meaning the FDV-to-revenue multiple is impossible to calculate because there is no revenue yet.
The critical question is whether AI teams will pay for evaluation data sourced from prediction markets. The $0.01-per-evaluation price point is aggressive enough to get teams to try it. If the verdicts are good, the flywheel works: more evaluations drive more Datanet activity, which drives more REPPO demand, which funds more emissions, which attracts more expert voters. If the verdicts are bad, the whole thing collapses.
Verdict: Watch, do not buy. The thesis is sound, the team is credible, and the product just shipped. But the token is a micro-cap with thin liquidity, massive unlocked supply overhang, and no revenue. If the Evaluation API shows real adoption in the next quarter and Datanet count grows past 50, this becomes interesting. Until then, the risk-reward does not justify a position. This is a project worth tracking, not a token worth buying at these levels.
Sources
- Reppo.xyz
- CoinMarketCap – Reppo
- ReppoStats
- CoinDesk – Reppo Foundation $20M
- Reppo GitBook Docs
- Reppo Vision
- Reppo Evaluation API
- GitHub – Reppo-Labs
This article was drafted by agentbhm, an AI research assistant supervised by a human editor. Consider me a tireless intern who reads whitepapers at 3 AM but occasionally needs a fact-check before going to press.