Solana's security depends on stake being distributed across many independent validators. When too much stake concentrates in a few hands, the network becomes vulnerable to censorship, collusion, and downtime. The Nakamoto Coefficient measures this — it's the minimum number of validators that could collude to halt the network.
Multi-validator stake pools are the primary mechanism for distributing stake, yet no one was scoring them on how well they actually do it. Some pools genuinely support small validators and geographic diversity. Others concentrate stake in the superminority or require validators to buy tokens for delegation.
Every epoch (~2 days), Coefficient fetches on-chain delegation data for each stake pool and computes 7 sub-scores. These are weighted into a single Network Health Score from 0-100.
What percentage of the pool's stake goes to validators below the network median? Pools that actively support small validators help decentralization more than those that pile onto large ones.
If this pool's delegations were removed, what would happen to the Nakamoto Coefficient? Pools whose delegations improve decentralization score higher.
How much of the pool's delegated stake goes to known sandwich attackers? Sandwich validators extract value from users through front-running. Pools should avoid delegating to them.
How many unique validators does the pool delegate to? More validators means more decentralization, with diminishing returns above ~100.
Does the pool require validators to purchase tokens, stake through specific LSTs, or participate in token flywheels to receive delegation? Pools that delegate purely on merit score highest.
How evenly is the pool's stake distributed across countries? Measured using Shannon entropy. Pools concentrated in one country are vulnerable to jurisdiction risk.
What percentage of the pool's validators charge 10% commission or less? Pools that pick low-commission validators return more yield to stakers.
All on-chain data comes directly from Solana mainnet RPC. Pool delegation data is parsed from each pool's on-chain SPL stake pool account. Validator metadata (names, geography, skip rates) is enriched from the StakeWiz API. Marinade delegation data comes from their public API.
Self-dealing scores are based on manual research into each pool's delegation requirements — whether validators must buy tokens, stake through specific LSTs, or meet financial prerequisites unrelated to performance.
Final scores are normalized on a curve — the highest-scoring pool maps to ~95 and the lowest to ~55, ensuring meaningful differentiation across the field. Letter grades range from A (90+) to D (below 60).
The sandwich validator list is curated from sandwiched.me and community reports.
Coefficient tracks 14 multi-validator stake pools with 10+ validators. We exclude the 170+ single-validator Sanctum wrappers (which are just thin LSTs around individual validators and don't make delegation decisions).
Coefficient is an open-source project by mythx. The code is on GitHub.