I make the following analogy:
You have two walls. Both of them are blocking your path to a big pot of gold on the other side. One wall is camoflaged. The other is glowing neon making it very clearly visible. Both walls have holes.
The holes in the camo wall are large and persistant but half covered in leaves. The holes in the neon wall each have a spotlight on them and are constantly appearing and disappearing. Each time one appears there is a huge BANG and a red arrow pointing to it. But soon there-after they are patched up by dilligent gnomes armed with plaster and bricks.
Neither is perfect because they both have holes. But one wall attracts less attention to the holes. Here's a hint... trial by fire is not acceptable when the trials are occuring in the real world with real data. Might want to paint fewer arrows pointing toward your faults.