It always frustrates me when I have more trouble with the
Riddler Express puzzle than the Riddler Classic puzzle. Yet that's where I am this week. In about 30 minutes' work in Excel, I found that the number of times needed to flip the coins to correctly identify which one has been doctored is between 130 and 140 (I have the exact number, but I'll keep this post spoiler-free).
Meanwhile, I'm making little headway into the Riddler Express puzzle. My first finding was that when the match is tied at 1-1, there is no incentive to play scissors since double-scissors are objectively more powerful. Also, based on the symmetry of the game, the probability of winning a match at 1-1 is obviously 1/2.
But then I had to extrapolate backwards to an uneven match at 0-1 (or 1-0, depending on your perspective). I wrote the probability of winning the entire match as a bunch of terms based on the relative probabilities for each player using rock, paper, scissors, or double-scissors. There are 7 variables (recall that the player up in the match 1-0 has no incentive to choose scissors) and since the probabilities of playing something must sum to 1 for each player, there are 5 degrees of freedom.
So I wrote out the probability that the player down 0-1 wins the match. I'll let someone else write it symbolically, but this is 1/2 times the probability that they win with paper, rock, or double-scissors (1/2 because they still have to win the final round) plus the probability that they win with scissors (instant win), all divided by the same terms plus the probability that their opponent wins. I called the terms in the numerator W (for "win") and the added terms in the denominator L (for "lose").
Next, I took some derivatives to indicate whether the probability of winning is helped or hurt by throwing more of a particular hand. Again, I'll avoid writing this explicitly, but for example, the derivative of the probability of winning with respect to the rate at which the player throws paper is proportional 1/2 times the rate at which their opponent throws rock times L minus the rate at which their opponent throws double-scissors times W. You can extrapolate this pattern outward to obtain similar expressions for all other variables. These aren't quite derivatives (they omit the denominator squared in the quotient rule) but they must all be zero when an optimum strategy is obtained. I called these values "pressures" since they indicate whether a variable should increase or decrease.
Basically, I took my original values, then added to them some small constant times the pressures I calculated minus the average pressure. I iterated this process, hoping to home in on optimal strategies for both players.
It didn't work. For every initial condition I plug in, I end up with negative probabilities or probabilities that escape to infinity. I also tried a search based on fitting a paraboloid to the probability function, then finding the minimum of this paraboloid. This method is similar to Newton's method for finding the zeroes of a function, except it's a multivariable version. That took a long time to set up and was a complete mess.
So I'm stuck.