The deceptive strings aren’t just skewed in terms of total 1s vs 0s; I was calculating the relative number of each sub-string. string (000, 001, 010, 100, etc.) compared to the naïve expectation given the overall mix of 1′s and 0′s. 121 and 122 are non-standard, even by that metric (for example, 122 has literally zero instances of “0000” instead of the expected 2.29, and both have very different numbers of “010“ and “101” compared to “001” “100” “110″ and “011.”
I do not recommend this method, as it didn’t identify things very well, picking only 30⁄60 fake strings (other than mine and #106) and inaccurately calling a full 19 real strings fake. But seconding Guy’s note: I deliberately skewed my guesses to extremes, figuring that would be the best way to “win,” even though the EV was lower than being better calibrated. (ie 10% chance of 1st place + 90% chance of a negative score > 100% chance of a positive score)
The deceptive strings aren’t just skewed in terms of total 1s vs 0s; I was calculating the relative number of each sub-string. string (000, 001, 010, 100, etc.) compared to the naïve expectation given the overall mix of 1′s and 0′s. 121 and 122 are non-standard, even by that metric (for example, 122 has literally zero instances of “0000” instead of the expected 2.29, and both have very different numbers of “010“ and “101” compared to “001” “100” “110″ and “011.”
I do not recommend this method, as it didn’t identify things very well, picking only 30⁄60 fake strings (other than mine and #106) and inaccurately calling a full 19 real strings fake. But seconding Guy’s note: I deliberately skewed my guesses to extremes, figuring that would be the best way to “win,” even though the EV was lower than being better calibrated. (ie 10% chance of 1st place + 90% chance of a negative score > 100% chance of a positive score)