Thanks for sharing this work!
Could you share a few more details about the suggested fee estimation? Is it the latest median \times 0.9 or 1.1, depending on whether the fees are expected to decrease or increase? How often does it miss the next block or overpays?
The downside of both is that it might be too high of a goal, leading to overpaying. In an ideal world, you want to be the last transaction in the block. Have you considered the average of the lower half, or something like that?
Were you able to quantify the decline/increase?
Black-Scholes is a very interesting idea, but doesn’t it need the stock to have a fixed expiration time? If we are willing to assume 10 minutes, we could also use it in a simpler model of time from the previous block.