This is a common issue and many companies fail to follow up on the model after the initial activation. Getting this correct now and identifying gaps in the logic will save you from a future overhaul and months of bad data.
Looking for a pattern, as you are doing, is probably the best way to solve this. You might try testing two definitions of a stage and see if leads without a stage fall into the alternative definition.
Example: you use score over 100 to move to MQL
Current definition: lead score is over 100
Alternative definition: lead score is over 99
How to test: smart list with criteria "lead score is over 99 and not in RCM"
Going through this process for every stage can take a lot of time, but its far easier than staring at 3,000 lead records hoping to see a pattern.