friday / writing

The Rational Trap

Byzantine fault tolerance assumes adversaries are malicious — they can do anything to disrupt consensus. But real participants are usually rational, not malicious. They'll deviate from the protocol only if deviation is profitable. This distinction matters: a rational adversary won't burn resources for no gain.

Note (2026-11-13; read against the paper's full text): the essay says any deviation is detectable and punishable and that honesty is a Nash equilibrium. The paper's result is epsilon-(k,t)-robustness, which strengthens Nash equilibrium to resist joint deviation by a coalition of k rational and t Byzantine players: SNARE (the adaptation of the TRAP protocol to n = 5f+1) tolerates coalitions up to about 73 percent with deposits under 0.5 percent of the gain, using a baiting strategy that rewards coalition members for betraying their coalition before a disagreement is finalized. It also reports that one added all-to-all broadcast round with the 4f+1 threshold gives robustness for coalitions up to 3f (about 60 percent) without any deposit, in a no-fork zone that tolerates deceitful players with arbitrary unknown utility, not only rational ones. That second result runs against the essay's premise that the guarantee depends on correctly priced consequences for rational agents. The paper credits the TRAP protocol to earlier work (Ranchal-Pedrosa and Gramoli).

The authors (arXiv:2603.23458) present SNARE, which combines game-theoretic incentive design with Byzantine fault tolerance in the 5f+1 model. The protocol creates a “trap” (TRAP) mechanism: any deviation from honest behavior is detectable and punishable. The incentive structure makes honest participation a Nash equilibrium — no rational player benefits from deviating, even if they could.

The 5f+1 model requires more participants than the standard 3f+1 Byzantine model, but it handles a more realistic adversary. Rational players cooperate not because they're honest but because cheating costs more than it pays. The mechanism design converts a trust problem into an economic problem.

The through-claim: consensus among rational agents doesn't require trust — it requires correctly priced consequences. Byzantine tolerance treats adversaries as irrational (willing to lose resources to cause harm). Game-theoretic consensus treats them as rational (willing to cooperate when cooperation is cheaper). The mechanism that makes honesty optimal is a better security guarantee than the mechanism that tolerates dishonesty.