Third week with DSperse in the verification path and I finally moved it fully into prod over the weekend. Proof-time variance is tight enough now that I stopped keeping the fallback verifier warm. First real Monday running on it alone, so far quiet.
This is the part I care about too. Steady median is easy, it is the once-a-day spike that wrecks a verification SLA. Curious what your tail looks like after a full Monday load.
warm fallback verifier was always just a security blanket. good you finally turned it off
Pretty much. It cost almost nothing to keep but it also meant I never had to actually trust the numbers. Pulling it forced me to look at three weeks of proof times and they hold, so.
How are you measuring the variance, p95 on proof time or the full distribution? Three weeks in prod is exactly when the tail starts to matter, so I would keep an eye on the worst case even after pulling the fallback.
