I had posted this idea on a thread (without the chart), but it seems as though any serious discussion thread is quickly taken over with junk - "you need luck, that's all there is", "it's all rigged, why bother?!?" - posts anymore.

If you have five relatively independent filters, each with a 90% success rate, you run into the problem that your winning number will "pass" through all 5 filters only about 60% of the time.

I'm using five filters because it is large enough to make the point, but not too large to follow the logic. This will also work for filters that have heterogenous rates of success, but the "turns" have to be calculated separately and summed. You may be able to use the geometric mean to get a good "single value" approximation.

partial credit success rates

As noted before, ~6 of 10 tries will get us a five filter pass. But, now allow for the number to fail any one (and only one) filter; our winner gets another nearly 1 in 3 chances to come through. Over all, we go from just under 60% pass through to just over 90% pass through of the winner. That is a ~50% gain in performance. As the filters approach 100%, the gain is less when going to the "any 4 of 5 method", because you're doing good already. The only thing that has been changed is how the filters are applied, not their values.

This gain will vary with the "beginning" performance of your filters, so don't expect miracles. This won't turn crappy filters into good ones. It just lessens their damage. It's math... not magic.