Proof-of-stake networks are operated by validators running software that implements the consensus rules. Whether those validators run the same implementation determines how a single bug propagates.
Multiple implementations of one specification
Consensus rules are written as a specification, and independent teams build separate clients that implement it, often in different programming languages.
Correct clients behave identically because they follow the same rules. They differ in internal architecture, performance characteristics and the bugs they contain.
Independence is the point. Two teams working separately are unlikely to make the same mistake in the same place.
Why a supermajority in one client is dangerous
If a client holding most of the stake produces an incorrect block or fails to attest, the network can finalize the error or stall entirely.
Below a supermajority, the same bug leaves the affected validators offline or penalized while the remaining ones continue to finalize correctly.
The distinction is between an incident affecting some operators and one affecting the chain, and client distribution decides which occurs.
Penalties are deliberately non-linear
Networks penalize validators for being offline, and the penalty is designed to scale with how many are offline at the same time.
An isolated outage costs little. A correlated outage affecting a large share costs each participant far more.
The structure makes concentration expensive in expectation, so an operator's client choice carries a financial consequence rather than being purely technical.
Two layers must both be diversified
Validators run an execution client and a consensus client, and these are separate software with separate development teams.
Diversity in one layer does not compensate for concentration in the other, since a fault in either can produce incorrect behavior.
Public dashboards track distribution across both layers, which is how operators and researchers observe whether concentration is building.
These figures are estimates rather than a census, since client identity is inferred from behavior and block contents rather than reported directly by validators.
Voluntary reporting by large operators improves the picture for the stake they control, while the long tail of individual validators remains inferred, so the true distribution is known approximately at best.
What pushes operators toward the majority
The most widely used client typically has the most documentation, the most community support and the best-tested integrations, all of which lower operating effort.
Those advantages compound, which is why concentration tends to increase without deliberate counterweight.
Large staking operations address this by splitting infrastructure across clients internally, and some services publish their distribution so depositors can see how the risk is spread.