So consistent in fact, that time these decision can be put into a model, which can they decided automatically if there is a potential mistake by software / AI

E.g. for handballs, the fact the threshold of ball to hand varies to such week on week, game on game is outrageous. However with the datasets being generated we will up a consistent definition

Now the refs are always striving to give the best decision possible based on the rulebook, and they will have an even larger database of decision which they can discuss and refine as to how rules should be interpreted

Allowing even more bias to be removed based on a ref knowing who is playing based on the time of day, like a 12:30 kick off

In addition to this, there are also randomly generated possibilities shown to the refs during the game, or generated from historical games which the refs are being graded on.

This means we would know the decision is a clear and obvious mistake. And 50/50s are left as they were.

But here is the killer. The aim is to not reref the game, and the error to be clear an obvious. Therefore for the on-field decision to be overturned, it needs to cross a 70-80% threshold.

They can anonymise the players and kits. Then the referees vote on what the infringement outcome should be out of two possibilities, similar to Ask the Audience in WWTBAM
1. The on field decision
2. The potential alternative decision

However the replays are regenerated using software similar to what was used in the 2022 World Cup in Qatar