When blockchain QA tests pass locally but fail on mainnet — what’s really behind it?

I’ve been deep in blockchain QA for a few years, and one thing keeps nagging me — tests that run perfectly on local forks but misbehave the moment they hit mainnet. I’ve seen failures tied to timestamp drift, gas dynamics, even RPC latency.

My guess is that local forks hide too much “real-chain noise.”

I’m curious what others have observed in production or audits — which subtle differences between local and mainnet setups have cost you the most debugging time?

Let’s surface the recurring culprits and how you mitigated them.