EigenTrust

The core mechanism of EigenTrust relies on trust transitivity. The algorithm aggregates local trust scores assigned by each peer to others and iteratively computes a global trust vector. A peer is considered trustworthy if it is trusted by other highly trustworthy peers. This is mathematically equivalent to calculating the principal eigenvector of the normalized local trust matrix, ensuring a unique and stable global reputation score for every node in the network.

EigenTrust is engineered for Sybil resistance. A malicious actor cannot gain disproportionate influence by creating many fake identities (Sybils) because these identities will initially have no trust connections. The algorithm's design ensures that trust flows from a set of pre-trusted seed peers, and Sybil nodes remain isolated in the trust graph. The only way to gain trust is to receive it from an honest, highly trusted source, making it computationally and socially expensive to subvert the network.

To avoid a single point of failure, EigenTrust uses a distributed hash table (DHT) to store and compute trust scores. The global trust value for each peer is calculated by a score manager responsible for that peer's hash space. To prevent malicious score managers from returning false values, the algorithm employs redundant computation by multiple managers and a majority-vote verification scheme. This ensures the integrity of the reputation score without any central authority.

A strategic extension of EigenTrust, called TrustGuard, introduces a countermeasure against strategic oscillating behavior. Malicious peers might build a good reputation only to abuse it later. TrustGuard maintains a reputation history and uses a fading memory function to weigh recent behavior more heavily. It also incorporates a probabilistic challenge-response mechanism where a peer's trustworthiness is continuously verified through random transactions, making it impossible to predict when a betrayal will be detected and punished.

EigenTrust solves the cold start problem through a set of pre-trusted peers. These are a small, static group of nodes known to be honest, which form the basis of the trust network. The algorithm biases the random walk of trust propagation to always have a small probability (a) of jumping back to these seed nodes. This mechanism prevents the trust score from being absorbed by a malicious collective and provides an initial trust anchor for new, honest peers joining the network.

Platforms use EigenTrust to distribute moderation authority among users. Peers rate content quality, and the algorithm computes a global trust vector that weights votes by reputation rather than treating all users equally.

• Sybil resistance: Fake accounts gain near-zero influence because they lack positive trust links from honest peers
• Transitive filtering: A moderator trusted by another trusted moderator inherits proportional authority
• Example: A community notes system where fact-checking scores propagate through a network of verified experts

EigenTrust was originally designed to combat freeloading in P2P networks like Kazaa and Gnutella. Peers who upload rare, high-demand files accumulate positive trust from downloaders.

• Tit-for-tat extension: Goes beyond direct reciprocity by incorporating indirect trust chains
• Resource prioritization: High-trust peers receive preferential bandwidth allocation
• Bootstrapping: New nodes receive a default trust value and must prove contribution to rise above the threshold

Decentralized oracle networks apply EigenTrust-like algorithms to rate data providers. Oracles that consistently submit accurate off-chain data to smart contracts earn higher global trust scores.

• Slashing integration: Low trust scores trigger economic penalties in proof-of-stake systems
• Consensus weighting: Oracle responses are weighted by reputation during aggregation, reducing the impact of faulty or malicious feeds
• Example: Chainlink's implicit reputation model draws from the same eigenvector trust principles

Marketplaces replace centralized star ratings with EigenTrust-based reputation graphs. A seller's score reflects not just direct reviews but the trustworthiness of the reviewers themselves.

• Collusion detection: Coordinated fake review rings fail because fraudulent accounts lack connections to the honest trust graph
• Category-specific trust: A seller trusted for electronics does not automatically inherit trust for home goods
• Dynamic decay: Older transactions lose weight, ensuring scores reflect recent performance

Connected vehicles share road condition and hazard data over ad-hoc mesh networks. EigenTrust ensures that malicious or faulty sensors broadcasting false data are identified and isolated.

• Pre-voting phase: Vehicles exchange local trust vectors before computing global scores
• Low-latency convergence: The algorithm's iterative computation converges in fewer than 10 cycles for moderately sized vehicle clusters
• Safety guarantee: Emergency brake warnings from high-trust peers override conflicting data from low-trust sources

In privacy-preserving federated learning, a central server must select reliable clients for model aggregation. EigenTrust scores identify nodes that submit high-quality gradient updates rather than poisoned or noisy data.

• Contribution auditing: Clients rate each other's model updates based on validation accuracy
• Byzantine fault tolerance: The global trust vector naturally down-weights adversarial nodes without requiring a central authority to manually blacklist them
• Incentive alignment: Only high-trust nodes participate in the final aggregation round, improving model convergence speed

The logical property that allows trust to flow through a network. If node A trusts node B, and node B trusts node C, then A can derive a measure of trust for C. EigenTrust relies entirely on this principle, iteratively computing global trust by aggregating these transitive trust paths across the entire peer-to-peer graph.

• Direct Trust: Explicit trust from one peer to another based on direct interactions.
• Indirect Trust: Inferred trust derived from the recommendations of trusted intermediaries.
• Convergence: The algorithm propagates trust until the global reputation vector stabilizes.

The capability of a network to defend against attacks where a single adversary creates multiple pseudonymous identities to gain disproportionate influence. EigenTrust provides Sybil resistance by ensuring that new, malicious nodes have no pre-existing trust relationships with honest nodes.

• Attack Vector: An adversary creates thousands of fake peers to subvert voting or reputation.
• EigenTrust Defense: Malicious nodes start with zero trust and cannot easily earn it from the honest subgraph.
• Contrast: Unlike simple averaging, EigenTrust's global view prevents a flood of fake peers from boosting a malicious score.

The foundational link analysis algorithm used by Google Search that EigenTrust conceptually extends. PageRank measures the importance of web pages based on the quantity and quality of links. EigenTrust adapts this eigenvector centrality concept to measure peer reputation based on the quantity and quality of satisfaction ratings.

• Random Surfer Model: PageRank's teleportation factor is analogous to EigenTrust's pre-trusted peers, preventing rank sinks.
• Matrix Computation: Both use power iteration on a stochastic matrix to find the principal eigenvector.
• Key Difference: PageRank scores authority; EigenTrust scores trustworthiness based on transaction history.

The process of assigning initial trust values to new entities that lack historical interaction data, addressing the cold start problem. EigenTrust handles this by designating a set of pre-trusted peers whose global reputation is hardcoded and never decays.

• Pre-Trusted Peers: A small, static set of nodes (e.g., the system's founding members) with an initial trust distribution.
• Function: These peers act as an anchor, preventing the global trust vector from collapsing to zero and giving new, honest nodes a path to earn trust.
• Bias: The choice of pre-trusted peers introduces a necessary, centralized bias into the otherwise distributed algorithm.

A measure of the influence of a node in a network. The core mathematical operation of EigenTrust is computing the principal eigenvector of the normalized trust matrix. A peer's score is high if it is trusted by other high-scoring peers.

• Formula: The reputation vector r converges to the left principal eigenvector of the normalized trust matrix C.
• Power Iteration: The algorithm starts with a uniform vector and repeatedly multiplies it by the trust matrix until convergence.
• Significance: This ensures that a recommendation from a highly trusted peer carries exponentially more weight than one from a distrusted peer.

A peer-to-peer communication procedure used to disseminate reputation updates reliably across a distributed system. In a decentralized EigenTrust implementation like HonestPeer, nodes do not have a global view; they use a gossip protocol to aggregate trust scores.

• Epidemic Spread: Each peer periodically exchanges its local trust vector with a random subset of other peers.
• Eventual Consistency: The global reputation score converges probabilistically across the network without a central server.
• Scalability: This method avoids the bottleneck of a single aggregation point, making the system robust to churn.