I think either approach is incorrect. You should not be taking an approach just because it works in the past nor should you do something repeatedly if it's shown not to work. You should always be iterating and making adjustments to your approach. If something does not work you should consider if it was bad luck or the wrong approach and that's where you should then make adjustments if necessary.
Interesting... but you could indefinitely adjust on something wrong, isn't it ?.... what about working any approach with large number of events OR very low probability events with a smaller number .....to validate any approach... (same thing used for Artificial intelligence validation...
