User blog comment:Homuhomu123/Trigger Rates of Night Battle Cut-In/@comment-25779449-20150716021718/@comment-2142840-20150722035125

Torp cut-ins appear to be a piece-wise linear function w/ a 50 hard cap for non-flagship, since anything above it doesn't seem to contribute to cut-in chance. There are a few outliers, usually the n<100 samples, which appear to be abnormally high, but I would assume that the decrease in cut-in probability as luck increases is due to sampling error. This seems consistent with the older formula.

Some of these graphs look like the linear function might not be a good fit. Once more values are filled in, I think it would be worthwhile to run a graphic residual scatterplot to see if there is a softcap breakpoint (square rooted luck) somewhere.