I know the not a sudden transition has been explained.
FWIW I don't like the terms "safe" vs "unsafe" for car seats because we're really only ever talking about comparative risk and probability. It doesn't actually make sense to call something unsafe because a lot of the time when people claim a certain car seat is "unsafe" they mean it is less safe than some preferred option that they would recommend. But for example, a 2.5 year old in a forward facing seat is certainly much safer than they would be using just a seatbelt, even if they would be safer still in a rear facing seat.
But in any case - it's hard to give an exact figure because unfortunately, the data on car seat effectiveness comes from actual children who are killed or seriously injured in car accidents, and you can only compare between groups of said children, since no two accidents are directly comparable, which is why you end up with "round number" milestones when in reality, this isn't likely to be representative, it's just how the groups are divided up. It's not like children wake up on their fourth birthday with a magically stronger neck than the day before. You also may want to be cautious of making judgements at the edges of a group. For example, one of the groups often used to compare data is raw figures comparing Germany to Sweden at a time where most parents in Germany would move to FF at around 9-10 months old. The category of 1-2 year olds has a very large, clear difference, whereas the category of 0-1 year olds looks pretty much the same. So you could say OK, whatever German parents are doing up to 12 months is fine/comparable to Sweden. But this is clearly not true, because it doesn't make sense that 9-12 month old babies are perfectly safe forward facing, and then at 12 months it becomes unsafe for them - it's more likely that across all accidents where 0-11 month old babies were killed or injured in Germany at the time the data was collected, only a small number of them were over 9 months old in the first place.
But it seems from the data/information we do have:
Forward facing seats do not provide adequate protection at all for babies under a year of age - injury rates are only slightly better than being unrestrained, and injury due to strain on the neck from the seat itself is common.
Then up to about age 2, forward facing seats have some protective effect but rear facing seats have such significantly better protection that it absolutely makes sense to recommend rear facing. Over age 2 and certainly 3 the data is less clear and it seems to be more context dependent.
Seatbelts/booster seats don't provide much benefit over being unrestrained until around 2-3 years old, and even then it's not great. 5 point harness seats continue to offer a significant benefit over seatbelt/booster until around 4-6 years, after this the data is less clear.
Seatbelt alone is dangerous before age 3 as said, but since there are significant benefits of other seats until 4-6 you could use that as a minimum. In terms of difference between seatbelt and booster seat, the numbers are even fuzzier here because you tend to have big age groups and they often only bother to look at children up to about age 8 and don't look at figures for older children.
To say more simply:
Absolute minimum for forward facing: Around 1 year, better 2-3 years for best risk reduction.
Absolute minimum for booster seat: Around 3 years, 4-6 years for best risk reduction.
Absolute minimum for seatbelt use: Around 6 years, 8+ years for best risk reduction.
Current UK law is:
Forward facing minimum 9kg (approx 9 months) or minimum 15 months depending on the seat. Majority of seats on market have the 15 month minimum.
Booster seat minimum 15kg/100cm (approx 3 years)
Seatbelt minimum 135cm or 12 years (135cm is on average 9-10 years).