I refer to this document:

https://reference.opcfoundation.org/src/v104/Core/docs/Amendment4/readme.htm

In Chapter "6.8.2 Secure Channel Handshake", this is written:

"In addition, a unique InitializationVector is needed for each Message. This value contructed
from the ClientInitializationVector or ServerInitializationVector where the first 8 bytes are
replaced by the values in Table 49c."

Note the missing verb in the sentence, and the typo in "contructed".
But I am writing this e-mail mostly because I found this apparently deviating
implementation in the .NET Standard stack:

    private static void ApplyChaCha20Poly1305Mask(ChannelToken token, uint lastSequenceNumber, byte[] iv)
    {
        iv[0] ^= (byte)((token.TokenId & 0x000000FF));
        iv[1] ^= (byte)((token.TokenId & 0x0000FF00) >> 8);
        iv[2] ^= (byte)((token.TokenId & 0x00FF0000) >> 16);
        iv[3] ^= (byte)((token.TokenId & 0xFF000000) >> 24);
        iv[4] ^= (byte)((lastSequenceNumber & 0x000000FF));
        iv[5] ^= (byte)((lastSequenceNumber & 0x0000FF00) >> 8);
        iv[6] ^= (byte)((lastSequenceNumber & 0x00FF0000) >> 16);
        iv[7] ^= (byte)((lastSequenceNumber & 0xFF000000) >> 24);
    }

To make that clear, I think implementing this algorithm as it is written in the
amendment would be okay too, but using XOR is certainly a better solution.

Suggested fix: change the spec, or add an errata,
specify that the masking is done by using XOR on the first 8 bytes of the
key value with the mask value.