I started judging @Dusk by fork count, but I think that metric can hide the more interesting problem.

Zero visible forks doesn't automatically mean the network is healthy. What happens when latency rises, packet loss appears, and some validators go offline at the same time?

I wouldn't ignore fork count either. It is still a useful outcome metric. The better question is what conditions sit behind it.

If P95 latency reaches 500ms, packet loss hits 5%, and validator availability drops 20%, does DUSK see more forked rounds? Does consensus take longer? Does finality slow down before users notice anything?

I'd also stop treating one explorer's mempool number as the full picture. Comparing pending queues and transaction age across multiple nodes could tell us whether congestion is local or spreading.

Some disagreement is normal in distributed systems. That doesn't worry me much.

What I’m watching is the failure curve: does DUSK absorb combined stress gradually, or does reliability suddenly fall off a cliff?

That answer tells me more than fork count alone.

#dusk $DUSK

$ACE $EDEN

What matters most for DUSK network reliability?