Your OTP is extremely sensitive to the quality of randomness and requires a lot of it - which makes things very difficult.
It provides no authenticity and integrity or at least proof of tampering.
It does not protect from message reordering and capture+retransmission.
It obviously leaks metadata in a real-world usage: sender, receiver, msg length, time of message.
So, you verify the integrity by a mechanism separate from the encoding. (This may or may not be sufficient, depending on the needs of the OP.)
Also, of course - the odds of someone tampering with a message encoded using OTP and having the tamper still appear a valid message once decoded could probably best be described as astronomical.
If the key is pseudorandom instead of purely random, then you have yourself a stream cipher and not the OTP.
But this is not very practical
You'd need a way to keep track of sections of the pad [probably framing the message, maybe a start keyword could work - it'd give away the section of the pad, but as that section is never re-used... or, if you can assume few messages will be lost, just keep a local counter, and do a sliding search] - and there are other issues. Split the pad into sender/receiver sections (A only encrypt and send with a-apd, B only encrypts with b-pad - to avoid two messages encrypted with the same section of pad, crossing each other).
Still, if Eve gets a hold of a plaintext message from Bob, and can intercept the ciphertext before it gets to Alice - Eve gains access to a section of the otp by xoring - and can send anything to Bob (up to the same message length).
I'd be interesting to hear if there are modern cryptosystems /research based around one time pads?
There's this implementation from 2015: https://techcrunch.com/2015/03/24/one-time-pads-ride-again/