I did this sort of genetics work a long time ago, but this is what I'm seeing based on the WHO protocol and the kits 1 2 they say to use.

The testing protocol assumes that no human DNA is present in the sample before running the final amplification/detection process. If intact human DNA (from a random human cell or free-floating) is present, then you will get a false COVID positive. Theoretically, if you removed all human DNA from the sample, the the test should be legit.

The first step of the WHO protocol (using "Kit Extraction NucleoSpin Dx Virus") is meant to eliminate all human cell/DNA contamination through a "NucleoSpin gDNA Removal Column" which is basically like a filter that pulls out everything that's not RNA.

The big problem with that is that no procedure is ever 100% efficient. Maybe sometimes you luck out and no DNA molecules get through, but most of the time there will be some contamination. This is why they refer to output as "high quality" output and not "perfect output".

On top to that, kits are faulty and technicians do fuck things up. When I was in the lab, purification screw ups were not unusual. Some people were better at it than others. Kits do help, but again, nothing is infallible.

Now, if even one molecule of the relevant human DNA gets through the process, it will end up being amplified through the final PCR process, and this may be a huge source of the false positives we're seeing. It probably won't happen all the time, but that's why the test isn't ringing the bell on all test samples.

To sum up:

WHO protocol

1. Obtain human tissue/fluid sample by swabbing.
2. Remove human cells/DNA from sample using a filter column.
3. Assume sample contains only RNA.
4. Transcribe RNA to cDNA using kit.
5. Amplify cDNA using PCR with appropriate primers and detect with probes.

The main point of failure here is steps 2 and 3. If a single molecule of that human chromosome 8 makes it through the filter step (or contamination occurs later), you will get a false positive. Won't happen all the time, but that's why the test doesn't appear to be totally fraudulent.