This seems a little odd. I ran the #Rogue room algorithm for all seeds, filtered to only those with 3 missing rooms. I computed a probability model of the same situation.
The data shown here is the size distribution, and there's a tiny but consistent discrepancy in the distribution. It's _so_ consistent that I feel like it must be a bug in my analysis.
For example, if "truly" random the 5th percentile of room size should be 316 but the actual is between 313 and 314.
The 10th percentile should be 346-347 but it's 343-344.
Room size 500 has rank 0.65976 in the statistical model but 0.67530 in reality.
I think there are some extremal points missing on both sides of the distribution but I need to double-check my probability calculation. The part I added today was compensating for the top rooms having only heights 4-6, not 4-7.