That's not actually the case. While you're right that after compiling it to the native executable (on iOS this will be a Mach-O binary) you do lose a lot of higher level contextual information the source may have (comments, non-externally linkable variable names, preprocessor or other labels, even non-externally linkable function names can be stripped out). But you still have access to everything else that is left, the code in the form of the assembly (disassemble the machine code), the data, etc. And that's just regarding C, in Objective-C there's even more information kept in the executable since it needs to support its runtime introspection features (so method/class names, method argument types, instance variables and their types, etc. is all kept around in the executable).
The general process for reversing something like this would be to get your debugger/disassembler, locate the important sections of code that are of interest, and begin to understand what it going on. After you understand you can start writing your own version of it. Although technically that isn't even needed, another option is to just rip the relevant assembly code for the function you're interested in, make some fixups so it'll work, and just put it into your own project.
As for the applications being encrypted and code signed on iOS, this can all be easily bypassed too. There's a few popular tools to do just that.
Technically nothing that's on the client is safe. A skilled reverser with enough incentive to reverse it, will be able to. So how does RAP stop someone from reversing their core diagnosis algorithms, their only option is to keep it on the server (which is what I believed they were doing anyway, only stream that wouldn't do that would be the humanitarian offering). In some cases it's actually possible to reverse an algorithm you don't even have access to, however all you have to analyse is then the data being sent out and the data being received; in some scenarios this is enough to actually work it out while in others it will not be.
However it also needs to be stated that a number of the relevant details of the algorithm are already available. RAP has released research papers in the past detailing some of their research.
Either way if the algorithms do get implemented by a third party RAP is still far ahead. As the ML had been trained on more data, have completed and continue to do clinical trials, will get FDA approval early next year, already have been in talks with numerous third parties, etc.
Expand