Modern blockchains separate consensus (agreeing on the order of transactions) from execution (computing what each transaction does). This decoupling is an engineering optimization — ordering is fast; execution is slow. By decoupling them, the consensus layer runs at full speed while execution catches up asynchronously. The throughput gain is real.
Baudet, Kelkar, and Sonnino (arXiv:2602.20341) identify a fundamental cost of this optimization: invalid transactions can be ordered before they are checked. A malicious leader can submit transactions that reference assets that don't exist, fill execution slots with work that produces no useful state changes, and waste the chain's computational capacity. The attack is called gaslighting — the leader convinces the consensus layer that meaningful work is happening when it isn't.
The impossibility result: resilience against gaslighting and full resource utilization cannot both be achieved deterministically in the decoupled model. If execution is decoupled from consensus, you cannot simultaneously guarantee that all ordered transactions are valid and that the chain uses its full capacity. Checking validity requires execution; executing before ordering requires coupling.
The intermediate solution: leader-based protocols where the leader pre-executes and attests to validity. This restores enough coupling to prevent gaslighting while preserving most of the throughput benefit. The leader becomes a bottleneck for correctness but not for throughput — an asymmetric coupling.
The general observation: decoupling stages of a pipeline for performance creates a gap between what is ordered and what is valid. This gap is exploitable. The optimization that makes the system fast is the same structural feature that makes it vulnerable. Recoupling must be selective — enough to prevent exploitation, not so much as to destroy the performance gain.