Let's take Jack Flaherty vs. Yoan Moncada as an example, since both players had an incredible 2019 but a down year in 2020.
Flaherty 2019 4.7 fWAR 5.7 bWAR 2020 0.6 fWAR -0.3 bWAR 2021 Steamer 2.9 fWAR
Moncada 2019 5.7 fWAR 4.8 bWAR 2020 1.6 fWAR 0.7 bWAR 2021 Steamer 3.3 fWAR
Now I am not sure which WAR model BTV uses, you referenced the lower of Moncada's 2020 WAR, but BTV links to the players profile on Fangraphs. One would then assume their model is more aligned to FG, which Moncada is considerably more valuable in 2019 and 2020.
Using the logic that they are recency biased, it would be make sense if Flaherty projects to be a less valuable going forward than Moncada, especially since Flaherty has only 3 years of control left, and Moncada has 4 not counting the 2025 team option (and BTV incorrectly listed only 3 years of control left, yet they seem to have counted his salaries for the next 4 years, leading to a negative SV). Now they are projecting Flaherty to produce 103.2 AFV over the next 3 years while only 83.8 for Yoan (the difference here is SV fo a Dylan Cease), so this isn't just a dumb tool is fed with numbers from recent performance and spits out a projected value going forward. The problem here lies they have their own projection model, and it is different than the industry projection models, they also have some data issue where they only projected 3 years of control left for Moncada and counted his salaries for next 4. I would say the tool is a good baseline to check the ballpark estimate of SV for a player or trade, but it does leave room of subjectivity, Moncada is perfect example of an outlier.
And for shits and giggles, if we do not apply subjectivity to the BTV values, we end up with shit like these: