Study: Calling Crypto 'Decentralized' Depends on How You Count

Public blockchains are sold on one big promise: no single company or person is in charge. Three researchers tested that promise on two popular crypto ledgers, Chainlink and Uniswap, using 90 days of transaction data. Their finding is uncomfortable. Four perfectly reasonable measurement decisions — which wallets to count, how to group them, and so on — moved one concentration score from 109 to 2,336. That's a 21-fold swing. Same data, opposite verdicts.

There's a second wrinkle. Crypto watchers often cite two scores: the HHI (a concentration score, where higher means a few players dominate) and the Gini coefficient (an inequality score, where higher means wealth is lopsided). Here, the Gini barely budged no matter how they counted, while the HHI swung wildly. The lesson: those two numbers answer different questions and can't be swapped for one another. The team also found a simple coding slip — adding up repeated transfers versus overwriting them — that quietly threw away about 85% of all transaction volume and flipped a conclusion about who holds power.

So what does the data actually show? Both networks are extremely unequal in ownership, with Gini scores of 0.990 and 0.998 — a handful of wallets hold almost everything. But the routing of money is surprisingly spread out. Ownership and movement turn out to be nearly unrelated, meaning a small group owns the tokens while lots of different addresses move them around. The researchers also spotted hidden middlemen: 30 and 18 zero-balance addresses that shuffle funds while holding nothing themselves, 12 of them shared across both networks.

The sharpest warning comes from the governance side. On Chainlink, the addresses allowed to vote on proposals and the addresses actually moving funds barely overlap. That means the open, competitive routing rests on rules controlled by very few players — a 'Nakamoto coefficient of two,' meaning just two parties could effectively shut it down. The authors' recommendation is refreshingly simple: stop reporting concentration as one precise number, and report it as a range with the assumptions spelled out.