🚀 go-pugleaf

For a crypto system to be competetive nowadays, it must withstand the
same kinds of attacks that existing systems have proven their ability
to withstand.  This means all the methods that cryptography users,
including industrial, military, and criminal [ :-) ] organizations are
able to muster---including not just cryptanalysis, but commando raids
on traffic endpoints, plus the use of bribery, seduction, and torture
to get the algorithms and other info out of people.  You have to
assume that both the algorithm and an unlimited amount of encrypted
traffic with known plaintext and known keys are available to the
enemy.  They WILL get this info.  You can then change the keys,
but once they know the algorithm you can never make them stop knowing it.
Your system (with the new keys) must continue to withstand cryptanalysis.
The existing ones do---if yours doesn't, you don't have a worthwhile method.

THAT is why you have to publish your algorithm (or post permission the
sci.crypt commando squads to break into your home) if you want anyone
to take you seriously when you say it is secure.  

For relevant war stories, see "The Codebreakers" by David Kahn.