"Apparently" (I have no evidence, it's just what we've been told, just like I have no evidence for VAR "checking all contentious events") they do now use thicker lines for the attackers compared to the defenders, but the defender's line is just ONE PIXEL wide compared to FIVE PIXELS for the attacker.
What is the real life distance of ONE PIXEL?
So....
1 - a guess-timated arbitrary point in time is chosen when the ball is played
(very, very rarely the FIRST point of contact as it should be)
2 - a guess-timated vertical line for the furthest back part of the defender
3 - a guess-timated vertical line for the furthest forward part of the attaker
...and they then conclude that they can measure and confidently pronounce that an attacker is definitively offside by 2cm. The whole thing is an absolute farce, it's been introduced far too quickly without any proper statistical analysis completed, and its "down to the mm" implementation by PiGMoL is laughable.
As others have said, where's the benefit of the doubt for the attacker?
The software they use CAN do the job and the frame rate IS sufficient, but the VAR gives the impression that the most important part of the calculation is rushed (When the ball was played) and then attempt to be too accurate.
The real problem, however, is that PiGMoL don't want it to work definitively (ala goal-line tech) they want it manipulatable so that they can create "drama"