Do you happen to know if there's any known answer tests floating around for this version of SRPs? Below we have tests for RFC 5054 which include known-answer tests, so presumably you could tell whether or not your given (third-party) SRP implementation corresponds to RFC 5054 by feeding it these vectors... it'd be good to have that for the RFC 2945 implementation as well so we can ensure interoperability!

Searching for RFC 2945 I found this: https://github.com/secure-remote-password/test-vectors/tree/8381017c812a273f30ba866af673e3726b314ab3 Their library seems to implement both RFC 5054 and RFC 2945 - knowing this before, it might have simplified the task quite a bit :)
I can just confirm that my implementation seems to work fine with Apple's after more than 2 months of testing. I might be able to capture some test vectors as well.

\o hello @jimm98y -- I'm not too particular about the source of the vectors, just that there's something we can potentially point to and say, we expect to be compatible with these values; check and see if your SRP protocol matches and if not, we'll need to potentially implement a third. :-)

That said, if those vectors pass with this code, happy to take them!

@cipherboy Now they do! It turned out I've made a mistake porting my original code to BC: I did not realize that HashPaddedTriplet does exactly what the name suggests - it pads to the desired length and then hashes. For RFC2945 M2 the padding must not be performed for M1 and K components, so I had to add a new method CalculateM2RFC2945 that implements a customized hash function.