It's not all manual guessing, it automatically applies heuristics when possible. You can see an example in this video:

Note that I don't manually guess anything. I simply pick a transaction from a recent block and it automatically identifies the decoys: namely, every Possible Spender except #1 is (according to the automatically-applied heuristics) a decoy. #1 is the "real" spender.

The heuristic applied in that case is called Recency Bias and is discussed as a standard characteristic of monero's decoy selection algorithm in the excellent Breaking Monero series (see Episode 5).

The recency bias heuristic takes advantage of the fact that the decoy selection algorithm used in the most popular monero wallets is biased toward selecting keys from recently created txos (on the principle that actively circulating coins are more likely to be spent than old ones). When you have a group of "new coin" decoys, coins that are significantly older stick out like a sore thumb, and you can plausibly identify them as the real spender's coins.

Consequently, I didn't need to manually eliminate the decoys; my software simply noticed that every decoy in the example transaction was a recently-created txo, but there was one txo that was much older. That one stuck out and, per this heuristic, was the real spender, because it is very unlikely that the decoy selection algorithm would choose that txo.