HTML Report miscalculation between Reference & Test

I've implemented Visual Regression on a production server, that is triggered when the Gitlab pipeline runs. My issue is that the results are not calculated correctly after I change something to the website and then run a test.

1. Run backstop reference
2. Run backstop test
3. Valid test as it should
4. Change something to the website
5. Run backstop test
6. Valid test - but it should be invalid

What's happening:
* Open the HTML Report;
* The reference & test images are opposite to each other;
* The thumbnail of the reference image shows the original version of the site - ok;
* The thumbnail of the test image shows the new version of the site - ok;

* When I click on the thumbnail of the **reference** image - the **TEST** image opens;

I've tried deleting the generated folders that contains the report and the images; I've checked the configuration multiple times;

* On my local computer, the setup works.