Things that were considered worthy of being classified (as 1930s-era secrets in the field of cryptanalysis) included what we would now call iterative methods (in algorithm design).
One example of an iterative method the highly-crypt0-centric method of banburismus – the updating of Bayesian estimates concerning likelihood of potential solutions in a search set to help direct search directions. One thinks also of counter-intelligence examples in use in the 1940 era when planes were sent to search, being intended to be observed by the enemy as they conducted searches using known-search methods; that hid more advanced search processes used only in well-protected crypto facilities
Iterative methods used for calculating eigenvalue/vectors from approximations, for quantum mechanics related matrices/operators, were also “guarded secrets”. The secrets concerned the raw “methods” themselves along with the ”means” used to complete the processes in a cost-effective time. One things of processes based on rayley quotients and minimization to find the minimum eigenvalue (and the second least eigenvalue) – given the special role this distance has in certain groups when allowing convergence to only occur after almost the maximum number of iterations.
The special space-relations that exist between solution-spaces and search spaces were known to characterize complexity and difficulty metrics – allowing development of key-wrapping ciphers that would measure the security of a key in terms of how long it WOULD take to reverse the mathematics (assuming the ciphertext data had been compromised by capture). This obviously gives a maximum operational window for use of the keying material, in a worst case analysis world.