How do the statistics of lost BTC come about? What metrics are used to judge?

Just curious when articles cite that certain percentages of BTC that are lost, effectively altering the value of the rest of the supply -- how are these stats generated or gathered?

Please shed some light if you can, as a lot of these sources sound very confident when quoting these numbers but I personally feel like it's not really something that can be proved easily.

If you dig further, the actual source messages are commonly phrased as "estimates", the cited Cane Island research paper does not go further into how the classification was done (unless I did not read thoroughly, happy to be proven otherwise). In which case how does the 4% lost per year as well as the total sum come about?

The only way I can think of reasonably to measure against is activity on the address, however just because a wallet sits for a few years without transactions, does not necessarily mean no one has access to them. Where is the line between "someone HODLing" against "this is probably lost"?