On Github
https://github.com/kpurens/CudaBrainSecp, you say:
In the example diagram an extra calculation Sha256 Transform is done before Secp256k1.
This calculation makes it (nearly) impossible to guess the previous or the next private key.
Why (nearly)?
If the private key is hashed then the GPU kernel can't simply re-use an existing public key.
Re-using a public key would be ideal, since that would allow doing one point-addition + modular inverse to obtain the next public key.
However since the private key is hashed then there is no way of knowing which is the correct point-addition (which point-addition would result in the next private key that you wish to check).
It is nearly impossible because it would be the same as guessing a 256-bit number (which is technically not impossible).
I should probably make that paragraph more clear and understandable.