This is odd.
I have a mono voice track that ACX check reports the noise floor passes @ -64db. If I duplicate that track - without modifying a thing - ACX check reports the duplicate track’s noise floor fails @ -41db.
I am running Audacity v3.1.3 on Fedora but the same thing happens on my Windows machine. I’ve done dozens of recordings and this is the only one that this has ever happened to. I have no clue why. Has this happened to anyone else?
How exactly are you doing that? Are you importing the MP3, or do you mean that this is all in one project?
Yes it is, so precise details about what you are doing are essential.
Thanks for your response. For the original track, I am live recording voice from mic into a new mono track. I duplicate the track by (menu) Select → All then Edit → Duplicate
Which version of Audacity, and how did you download / install it?
Which version of ACX Check?
Audacity version stated in the OP is : v3.1.3
Installed via DNF, the platform’s package manager.
Name=ACX Check
Version=2.4.2-1
Since this is the only one of dozens sessions where this has happened, I’m willing to just call it a fluke and move on. I’d still like to understand why it happens though.
So would I 
How big is the project? Would you be able to share it with me via dropbox or similar?
Do you have a copy of the project before you duplicated it?
If you opened it in a fresh Audacity, duplicated it again, does the same thing happen?
Audacity noise measurement is not a scientific marvel. I believe it still rips down the show looking for the quietest snippet. I don’t remember how long the snippet needs to be, but it’s not much. If you have a lead-in to the first word that is passing quiet but the whole rest of the chapter is a noisy wreck, Audacity will pass the chapter. There is a production problem. If you insert Generate Silence anywhere in your chapter, Audacity will see that surgically pure silence and ring alarm bells.
There is an inspection you can do. Divide the weird show in half and ACX-Check each half. No prediction what’s going to happen. I’m just curious.
You didn’t measure both the main and the duplicate at the same time, did you? If you go back and measure the original, does it still pass?
Koz