Effectiveness

Hmm... interesting... Question though, how are you normalizing the danger to the 0-to-1 range? Area, or peak?

If you're normalizing area, and the x axis ranges from 0 to 1, then the raw hit probability for the base danger makes a lot of sense to me. If you're normalization before integrating over botwidth is by area, then the danger should represent the probability of being hit. Then assuming that your surfing reliably gets you to what is normally near-zero danger most of the time, the additional baseline probability of being hit should be the hitrate... The way of handling it I see as making the most theoretical sense would be (hitProbabilityFromStats + actualHitProbability * (1 - hitProbabilityFromStats)). If you one wants to make it even more accurate... take the average "hitrate from stats" for where your robot ends up at each wave, and remove that from the actual hitrate. I think that should result in even better modeling of that baseline danger...