InfoFi Market Design Primitives: The Technical Architecture Turning Information Into Capital
When you post your opinion on X (Twitter), it costs you nothing to be wrong. When you bet $10,000 on a prediction market, being wrong costs you $10,000. That single difference — the cost of error — is the foundational primitive behind an emerging $381 million sector that is quietly rewiring how humanity prices truth.
Information Finance (InfoFi) is Vitalik Buterin's term for "a discipline where you start from a fact that you want to know, and then deliberately design a market to optimally elicit that information from market participants." Unlike traditional finance, which prices assets, InfoFi prices expectations — transforming epistemic uncertainty into tradeable signals. The sector now spans prediction markets processing $40 billion annually, attention markets distributing $116 million to content creators, and credibility networks securing 33 million verified users.
But beneath the marketing narratives, every InfoFi system runs on five technical primitives that determine whether information gets priced accurately or drowned in noise. Understanding these primitives is the difference between building a robust information market and an expensive spam machine.
Primitive 1: Cost-Bearing Signal Submission
The central insight of InfoFi is deceptively simple: opinions are cheap, commitments are expensive. Every well-designed InfoFi system forces participants to bear a real cost when submitting information, creating the friction that separates signal from noise.
In prediction markets, this takes the form of capital staked on beliefs. Polymarket processed 95 million trades in 2025, reaching $21.5 billion in annual volume. The platform migrated from automated market makers to a Central Limit Order Book (CLOB) — the same mechanism used by institutional exchanges — with off-chain order matching and on-chain settlement via smart contracts on Polygon. Each trade is a cost-bearing commitment: participants lose money when they're wrong, which creates relentless incentive pressure toward accurate probability assessment.
Ethos Network, which launched on Base in January 2025, applies this primitive to social reputation. When you endorse another user's trustworthiness, you stake ETH. That ETH is at risk if your endorsee behaves badly. The result: reputation endorsements carry real information precisely because they are costly to give.
The Intuition Protocol takes the most explicit approach, launching mainnet in October 2025 with $8.5 million in backing from Superscrypt, Shima, F-Prime (Fidelity's venture arm), ConsenSys, and Polygon. Its architecture treats information as an asset class:
- Atoms: Canonical identifiers for any discrete claim (an identity, concept, or piece of information)
- Triples: Subject-predicate-object statements — e.g., "Protocol X has vulnerability Y" or "Alice is trustworthy"
Both can be staked on via bonding curves. Creating low-quality Atoms costs you tokens; curating high-quality ones earns fees.
The common thread: cost of error creates a noise filter. Casual, low-confidence claims are suppressed by the friction of commitment.
Primitive 2: Proper Scoring Rules and Incentive Compatibility
Cost-bearing alone is insufficient — the structure of the payoff must ensure that truthful reporting is the optimal strategy. This is the mathematical domain of proper scoring rules: mechanisms where a participant maximizes their expected reward by reporting their true beliefs.
The Logarithmic Market Scoring Rule (LMSR), invented by economist Robin Hanson, was the foundational mechanism for early prediction markets. Its cost function — C(q) = b × ln(Σ exp(qᵢ/b)) — solves the bootstrapping problem by ensuring the automated market maker always has liquidity, even before any traders arrive. The parameter b controls the tradeoff between liquidity depth and the market maker's maximum potential loss. Historical trades are embedded in the current price, providing natural dampening against noise traders.
LMSR's limitation is capital inefficiency: it provides the same liquidity depth regardless of where prices are, wasting capital near extreme probability values (like a 95% confident market). Paradigm's November 2024 paper introduced a prediction-market-specific AMM (pm-AMM) that treats outcome prices as following Brownian motion — the same mathematical framework underlying Black-Scholes options pricing — and adjusts liquidity depth dynamically over time to maintain constant loss-versus-rebalancing rates for liquidity providers.
The same mathematical property — incentive compatibility — appears in non-financial systems. Ethos Network's vouching mechanism is incentive-compatible: if you stake ETH to endorse someone who later rugs users, your ETH is at risk. The optimal strategy is to only endorse people you genuinely believe are trustworthy. Intuition's token curated registries function similarly: stakers profit when their curated information is judged high-quality, lose tokens when it is low-quality.
Primitive 3: Graph-Based Trust Propagation
Static reputation scores are gameable. If a score is computed from raw counts (followers, reviews, transactions), a well-funded attacker can simply buy the inputs. Graph-based trust propagation is the solution: trust is not assigned absolutely but propagates through the social graph, making context and relationships central to score computation.
EigenTrust, originally designed to identify malicious nodes in peer-to-peer networks, is the leading algorithm for this purpose. OpenRank (by Karma3 Labs, backed by Galaxy and IDEO CoLab) applies EigenTrust to Farcaster and Lens Protocol social graph data. Rather than treating a "follow" from a new account and a "follow" from a highly-trusted account as equivalent, EigenTrust weights interactions by the reputation of the actor. The algorithm converges to a stable trust assignment where your reputation depends on who trusts you, and how much they themselves are trusted.
The result is a personalized trust graph — your reputation relative to a given community reflects the specific social connections within that community. OpenRank uses this to power Farcaster's "For You" feeds, channel rankings, and frame personalization. A user deeply embedded in the DeFi community gets different reputation scores for different contexts than a user embedded in the NFT art community.
Kaito's YAP scoring system applies the same logic to attention markets. Engagement from a high-YAP (high-reputation) account is worth exponentially more than engagement from a low-YAP account. This is PageRank applied to social capital: links from high-authority nodes transfer more authority than links from low-authority nodes. Kaito processes this across ~200,000 monthly active creators, computing mindshare — the percentage of total crypto Twitter attention captured by a given project — with weighted social graph traversal.
Ethos takes graph propagation even further with its invitation-only system. Your account's value depends not just on who vouched for you, but on the entire chain of who invited whom. A fresh account invited by a well-connected Ethos member inherits some of that member's credibility — a structural enforcement of the "trusted by trusted people" principle.
Primitive 4: Multi-Layer Sybil Resistance
Sybil attacks — flooding a system with fake identities to game scores, harvest rewards, or distort markets — are the existential threat to every InfoFi primitive. If fake identities are cheap to create, cost-bearing signals can be gamed with coordinated bots, reputation graphs can be artificially inflated, and prediction market resolutions can be manipulated.
The InfoFi sector has converged on a multi-layer defense stack:
Layer 0 — Biometric Verification: World (formerly Worldcoin) uses iris-scanning Orbs to issue World IDs on Worldchain. Zero-knowledge proofs enable users to prove humanness without revealing which iris was scanned, preventing cross-application tracking. With 7,500 Orbs deploying across the US in 2025, this layer aims for 200 million proof-of-humanity verifications.
Layer 1 — Invitation and Social Graph Constraints: Ethos (invitation-only), Farcaster (phone verification), and Lens Protocol (wallet-gated profile creation) impose structural friction on identity creation. Fake identities require real social connections to bootstrap.
Layer 2 — Stake-Weighted Trust: EigenTrust-based systems weight trust by stake or established reputation. Coordination attacks require accumulating real trust from existing members — expensive to fake.
Layer 3 — Behavioral Analysis: Kaito's algorithm was updated in 2025 after criticism that it rewarded KOL (Key Opinion Leader) content farming over genuine analysis. The updates introduced AI filters that detect paid followers, bot-like posting patterns, and content that mentions rankings without providing insight. Replies no longer count toward leaderboard rankings; posts that only discuss rewards without adding information are excluded from mindshare calculations.
Layer 4 — ZK Credential Aggregation: Human Passport (formerly Gitcoin Passport, acquired by Holonym Foundation in 2025) aggregates credentials from multiple sources — social verification, on-chain history, biometrics — into a single Sybil-resistance score using zero-knowledge proofs. With 2 million users and 34 million credentials issued, it enables applications to require a minimum Sybil resistance score without learning which specific verifications a user holds.
Galxe combines these layers at scale: 33 million users across 7,000+ brands hold credentials verified through ZK proofs, with Galxe Score aggregating on-chain activity across Ethereum, Solana, TON, Sui, and other chains into a multi-dimensional reputation metric.
Primitive 5: Continuous Pricing via Bonding Curves
Binary scores ("trusted" or "not trusted", "verified" or "unverified") are inadequate for information markets because they fail to represent the degree of confidence, reputation, or attention. InfoFi systems use bonding curves — continuous mathematical functions that determine price based on the quantity demanded — to create markets that price information on a spectrum.
LMSR's cost function is a bonding curve for prediction market shares: as more shares of a given outcome are purchased, their price increases continuously. This makes the market price a real-time indicator of collective confidence.
Ethos's reputation market layer creates bonding curves for individual credibility: "trust tickets" and "distrust tickets" linked to specific user profiles are priced continuously based on demand. When the community believes a user's trustworthiness is increasing, trust ticket prices rise. This transforms reputation assessment from a static badge into a live market with continuous price discovery.
Cookie.fun introduced the Price-to-Mindshare (P/M) ratio as a continuous valuation metric for AI agents: market capitalization divided by mindshare percentage, analogous to the price-to-earnings ratio in equity markets. A low P/M implies undervalued attention relative to market cap; a high P/M implies the opposite. This is the InfoFi equivalent of fundamental valuation — translating attention metrics into continuous investment signals.
Intuition's vault architecture uses bonding curves to determine how staking affects the credibility and relevance score of each Atom and Triple. Staking into a vault that contains accurate, widely-cited information is profitable; staking into a vault with poor-quality information incurs losses as others exit. The continuous pricing mechanism aligns curator incentives with information quality over time.
The Architecture That Prices Truth
These five primitives are not independent systems — they compose into a unified architecture. Cost-bearing signals are only valuable if they are structured as proper scoring rules (so truthful reporting is optimal), aggregated via graph propagation (so context affects value), defended by Sybil resistance (so fake signals are expensive), and expressed via continuous pricing (so degrees of confidence are captured).
The $40 billion annual volume in prediction markets, the $116 million distributed to attention market participants, and the 33 million credentialed identities across Web3 represent early evidence that these mechanisms work. Polymarket's monthly active traders grew from 45,000 to 19 million between 2024 and 2025 — a 421x increase driven not by speculation but by users discovering that prediction markets provide more accurate event probability assessments than traditional media.
The next wave of InfoFi applications will likely come from AI agents using these markets as data feeds. Kalshi already reports that algorithmic bots are the primary participants on its CFTC-regulated platform, with AI systems treating probability shifts in prediction markets as execution triggers for trades in correlated traditional markets. When AI agents consume and produce information at scale, the quality of the underlying pricing mechanisms determines the quality of the AI systems built on top of them.
What Vitalik called "info finance" is becoming the plumbing of the information economy: the layer that determines what is true, who is trustworthy, and what deserves attention — with capital-enforced incentives that traditional information systems have never had.
BlockEden.xyz provides infrastructure for builders across Sui, Aptos, Ethereum, and 20+ blockchain networks. Developers building information markets, reputation systems, and on-chain analytics can access production-grade node services and data APIs at BlockEden.xyz.