Sounds like snapshot test driven development would work well in that case though...

Sounds like snapshot test driven development would work well in that case though:

Write a scenario that uses a new water shader - eyeball it - if it's good then "fix" the snapshot and pass it around.

Repeat with a few other scenarios to see it performing under all relevant circumstances.

It can be a bit tricky if the result of the shader is not deterministic - but there are sensible ways of dealing with that.