Better - estimate complexity so things like "this algorithm will take 1234 + 2n + 500m operations " for our use case

Much better - research (not always possible)

Best - benchmark it using realistic data from your use case

> Better - estimate complexity so things like "this algorithm will take 1234 + 2n + 500m operations " for our use case

I feel like that is never the level of detail you want.

Either that is way too much detail and big-oh notation abstracts away the meaningless stuff.

Or, if this level of detail matters, then you really need to go much further. Think about cache hit ratios, memory layout issues, different speed of different operations, etc.

Calculating number of operations is a weird middle ground that seems pretty useless.