HMAC SHA256, inner hash works but outer hash fails

I'm experiencing a problem integrating the output of my SHA256 with HMAC. The first/'inner' SHA256 matches the expected output but the second/'outer' does not.

I am following the first example here:

https://csrc.nist.gov/CSRC/media/Projects/Cryptographic-Standards-and-Guidelines/documents/examples/HMAC_SHA256.pdf

I know:

• My K0 is correct
• My K0 xor ipad is correct
• My K0 xor opad is correct
• My Hash((Key^ipad)||text) is even correct

The problem is the ASCII representation from the final/outer sha256() is incorrect.

I get 218135127ad7a8e5967ce7b47499214d1df46a9589eb0bec7637c86021fd928d but it should be 8BB9A1DB9806F20DF7F77B82138C7914D174D59E13DC4D0169C9057B133E1D62

Code:

// Up to this point ipad, opad, K0, both XORs all match the expected values

const std::array<uint32_t, 8>& h_output_1 = Hash<BUFFER_SIZE>(&xor_1[0], 64, &message[0], message_length);
// This is correct
LOG("H_1: " << sha256::GetASCIIRepresentation(h_output_1));

const std::array<uint32_t, 8>& h_output_2 = Hash<BUFFER_SIZE>(&xor_2[0], 64, &h_output_1[0], 32);
// This is NOT correct
LOG("H_2: " << sha256::GetASCIIRepresentation(h_output_2));

Hash() appends the two buffers and calculates the SHA256.

Given the first SHA256 works I think the problem is how the 'inner' 8x uint32_ts SHA256 output is appended to K0 xor opad for the final 'outer' SHA256 calculation.

Given the first SHA256 works, can anyone suggest anything?

NB:

I cannot post GetSHA256() but it just returns the final state of the 8x uint32_ts used during the SHA256 transformations.

GetASCIIRepresentation() splits each uint32_t in to 8x 4-bit integers using masking and shifting, then converts to ASCII.

GetSHA256() and GetASCIIRepresentation() are unit tested and work correctly on message inputs beyond 512 bits.