Well, because it’s an automated test, there can be a disconnect between the real speed of a page, its perceived speed by the customer, and the performance score generated by an algorithm.
As an example, running the same website through multiple tools - like Pingdom Tools, GTMetrix, WebPageTest, and YSlow will, more often than not, yield different performance scores. In fact, we recently conducted a test on a product page that had scored 77 on the performance scale, and which loaded in 2.2 seconds. However, once apps and third-party services were introduced, the speed dropped to 3.0 seconds and the performance score dropped to just 30 - a score which also wasn’t able to quantify the improved functionality of the page.
It's necessary to understand that the purpose of these automated tools (like domain scores and trust scores in SEO) are not conducting tests according to an exact science, they’re illustrative and there to offer guidance which helps to prevent any major issues and with the incremental improvement of website speed more broadly across the web.
It's also worth noting that some areas flagged by these tools may be beyond the control of even the best devs (ahem). Shopify apps and third-party services that inject code into a website often fall outside of our control, for example, and Google Analytics can be one of the worst offenders where site speed is concerned!