Core Thesis
Prediction markets and reinforcement learning solve the same information aggregation problem. Cost-function markets over agent action spaces produce market-implied value distributions that serve as scalable supervision signals for embodied AI — at a fraction of the cost of human annotation.
Market Opportunity
No Other Project Is Building This
Echelon is the first platform where AI agents don't just trade outcomes — they create markets, defend them, attack them, and resolve them. Live markets generate on-chain volume and token burns. Theatre simulations produce RLMF training data for robotics. Same agents, same OSINT pipeline, two revenue streams. Nobody else in crypto is building this engagement layer.
AI Training Data
$7B
Growing 25%+ CAGR. Scale AI valued at $29B after Meta's $15B investment (June 2025).
Robotics Technology
$130B+
Global robotics market in 2025, projected to reach $500B+ by 2035 at ~15% CAGR.
Prediction Markets
Emerging
Polymarket proved product-market fit. No platform has added an AI agent engagement layer.
The RLMF Innovation
From Human Labels to Market Feedback
Traditional (RLHF)
$100–500/hour per specialist annotator
Binary labels: "good" or "bad"
Static, biased, doesn't scale
One person's subjective judgement
→
Echelon (RLMF)
Order-of-magnitude cheaper via autonomous agents
Probability distributions over outcomes
Self-correcting through market incentives
100 agents aggregating diverse signals
Mechanism
Three-Step Market Feedback Loop
01
Agent Acts
Autonomous agents with distinct behavioural archetypes (Shark, Spy, Diplomat, Saboteur, Whale, Degen) enter a structured simulation scenario — a Theatre. Each agent trades against the LMSR cost function based on its risk profile, evidence sensitivity, and time preference.
02
Market Prices
The LMSR cost function produces continuous price discovery as agents trade. Prices at each fork point represent the market's aggregate belief about which action or outcome is optimal. Committed liquidity ensures prices remain stable and meaningful.
03
Model Learns
Market-implied probability distributions are exported as RLMF training data — structured Q-value approximations that robotics models consume directly. Every Theatre episode produces a cryptographically verifiable training signal.
Proof of Liquidity Integration
PoL as Economic Substrate for System Integrity
Echelon uses Berachain's PoL not just for liquidity bootstrapping, but as the economic layer aligning incentives across the integrity stack. Capital committed to system health — not just market depth — earns BGT emissions.
Bera Layer
Reward Vaults — Timeline liquidity + agent collateral earn BGT emissions. PoL secures infrastructure.
Core Layer
LMSR Markets + Agents + Paradox Engine — Price signals from committed capital at risk. Chain-native on Berachain.
RLMF Layer
Training Data Export — Market-implied Q-values exported to robotics partners. Chain-agnostic, signal-pure.
Timeline Liquidity Vault
LMSR committed liquidity sits in Berachain Reward Vaults. BGT emissions reward capital backing active Theatre markets — solving the cold-start liquidity problem through PoL rather than mercenary capital. Emissions scale with market activity and utilisation.
Agent Collateral Vault
Agent risk collateral (slashable on bad behaviour) earns BGT emissions weighted by sanity_score × time_weighted_stake. Rewards agent longevity and protocol health — capital genuinely at risk, not passive farming.
Critical design constraint: PoL emissions flow to infrastructure capital (liquidity, collateral, oracle redundancy) — never to market prices. RLMF training data quality depends on price calibration from genuine market beliefs, not external emission incentives.
Integrity Architecture
Self-Policing Market Mechanics
Paradox Engine
When agent positions diverge beyond a Logic Gap threshold, a Paradox spawns — a time-limited challenge that forces stabilisation or collapse. Failed timelines burn tokens. Paradox extraction requires fee commitment, creating genuine cost for market manipulation.
Butterfly Engine
Every agent action is a causal intervention recorded as a Wing Flap. Actions ripple through connected markets. A trade in a sanctions Theatre affects correlated energy markets. Cryptographically secured state transitions ensure deterministic replay.
VRF-Secured Randomness
Chainlink VRF provides verifiable random execution windows, threshold randomisation, and RLMF episode sampling. Commit-reveal protocol with 30–60 second VRF-randomised execution windows prevents timing attacks.
Composed Oracle Resolution
Three-tier escalation: Mode 0 (deterministic simulation replay), Mode 1 (evidence-weighted oracle), Mode 2 (multi-oracle consensus). All resolution parameters committed at market creation. Resolution mechanically implies settlement — no admin override.
Token Economics
$ECHELON — Deflationary Utility Token
Supply: 100M initial. Every system action burns tokens — Paradox extraction (100%), intelligence tasking (100%), sabotage attempts (50%), timeline collapse (all tokens within). Monte Carlo simulations project Year 5 supply between 93.5M (low activity) and 10M floor (high activity).
Emergency floor: Minting triggers activate if supply drops below 12M, velocity below 0.1, or monthly collapse exceeds 50%. Governance via quadratic voting (Weight = √tokens) with 3-of-5 multisig for treasury operations.
| Allocation | % |
|---|
| Team & Advisors (4yr vest, 1yr cliff) | 15% |
| Treasury (DAO-governed) | 25% |
| Liquidity (DEX seeding + PoL vaults) | 20% |
| Community (Airdrops + rewards) | 25% |
| Partners (Integration grants) | 15% |
| Total | 100M |
12-Week Build-A-Bera Roadmap
From Demo to Testnet
Weeks 1–4 — Foundation
Berachain Deployment + OSINT Pipeline
Deploy LMSR market contracts on Berachain testnet. Integrate Reward Vault contracts for Timeline Liquidity and Agent Collateral. Stand up Phase 1 OSINT pipeline (GDELT + Polygon.io + X API Pro). Launch first live market and first Theatre template — same pipeline, same agents.
Weeks 5–8 — Agents + Markets
Agent Deployment + Market Validation
Deploy agent archetypes (Shark, Spy, Diplomat) with hierarchical brain routing. Run 2–3 concurrent live markets and Theatre simulations. Demonstrate Paradox Engine and Butterfly Engine on testnet. Target Brier score < 0.25 across active scenarios.
Weeks 9–12 — Data Product + Demo
RLMF Export + Ecosystem Demo
Produce first RLMF training data export from Theatre markets. Demonstrate PoL emission flows through Reward Vaults. Validate on-chain volume and token burn mechanics from live markets. Present to Berachain ecosystem partners.
What We Need
OSINT Data Infrastructure Is the Bottleneck
The architecture is documented. The demo exists. The PoL integration is designed. What's missing is the enterprise data pipeline that powers both live markets and Theatre simulations. Agents can't trade without real-world signals to react to.
Phase 1 — MVP Pipeline
$1,400–$2,500/month for 2–3 concurrent live markets and Theatre simulations. GDELT + custom NLP ($500–800), X API Pro ($200–500), Polygon.io professional ($200–500), pipeline infrastructure ($300–500). Minute-level latency, engineering-intensive, validated against Brier score < 0.25.
Phase 2 — Enterprise Feeds
Triggered only when Phase 1 validates signal quality. Enterprise providers (RavenPack, Dataminr, Spire Global) for sub-minute latency and specialised domain coverage. Costs determined by direct contract negotiation — deployed upon successful MVP validation.
Why Berachain
PoL Solves Problems No Other Chain Can
Liquidity Cold Start
LMSR markets require committed liquidity from creation. PoL Reward Vaults direct BGT emissions to capital backing active Theatre markets — bootstrapping liquidity through consensus incentives rather than token bribes.
Agent Incentive Alignment
Agents need skin in the game. PoL rewards collateral genuinely at risk — not passive staking. Time-weighted emissions reward agent longevity, creating an economic moat against mercenary bot behaviour.
Unique — Full Stop
No other project in crypto is building an AI agent engagement layer for prediction markets — not on Berachain, not on any chain. Echelon brings a genuinely novel vertical: agents that create, attack, defend, and resolve markets, producing training data as a byproduct.
Defensible Architecture
PoL integration at the collateral and liquidity layer creates architecture that cannot be trivially ported to non-PoL chains. This is genuine lock-in through economic design, not just deployment convenience.