I was counting the size of the inner flame 0 for skinny and 1 for fat. I did have the same problem with flame #102. I counted it as fat and it's needs to be skinny.
I had listed the other 2 correctly I believe. Here are the data streams I posted before.
Here is the data I have from the flames. The order is inner top/inner right/inner bottom/inner left/outer left/outer bottom/outer right/outer top.
Outer Color Yellow - 0 Outer Color Orange - 1 Inner Color Purple - 0 Inner Color Green - 1 Inner Color Skinny - 0 Inner Color Fat - 1 Short Flame - 0 Tall Flame- 1
Outer Color - 0001010100101101110010110000000010111011100011110000101011100001110001011111110 1011100111101001100101001101100110001111000110000101110101100011101110010
Inner Color - 0111001011010100001111000100100000100110111101001011111100110010001110011101010 1100001101111110001011011101000000000100111100000010110100100101101010111
Skinny Fat - 01001100001100011101011100011100001110001101011100010000111111000110011000100011010000011001001101001100011000101001011100111111110100111011011000111100
Tall Short - 0110110110100010110110110011110111110110110111110010100011100010110111110111110 1111100111001101101101101101000101101111100101000111100101100111001101100
Those few errors are not important from this perspective, because once you get the string with few error characters:
B34u7y, truth, andrarity5K8HiK2TP874KWF8Vhxov7tbgbttniB2gxqxP8KMU9TwK8odA18
its' obvious that you are on the right track and have to look up for questionable bits.
In the hindsight, we new that the bitstreams are correlated from their distribution, but if I remember correctly vast majority of solvers decoded them separately for some reason. I know what the reason was in my case: I believed there is no skinny/fat bitstream. Enough to fail.