If it was less than 10 blocks, then using 10 blocks for confirming transactions if fine, but if it was more, then 10 blocks is not enough...
This conclusion is not correct.
One does not draw such conclusion from 1 (in words ONE) single example.
Does anybody see the flaw in the following algo:
A) find the latest version of NRS (probably safer to hardcode this to avoid Evil Bob's tampering)
B) determine what fork the latest versions are on using method TBD
C) warning nodes that are not on the expected majority fork that they are not on the majority fork
D) Using above to generate overall NXT network fragmentation and adjust the number of confirmations needed based on this
Yes.
1) People should be encouraged to write their own node code. NRS is just one of them.
2) A) does not work out. Tampering is still possible.