The label is a guess
Addresses are pseudonymous. Clustering heuristics group them into entities, and manual labelling attaches names. Both are incomplete, both go stale, and a sophisticated participant can trivially use fresh addresses.
So "smart money bought" means "addresses previously classified as profitable, by a method of unknown accuracy, transacted".
Survivorship bias
Selecting addresses by past profitability guarantees you are looking at winners, including those who were simply lucky. Across millions of addresses, extraordinary records occur by chance.
Testing whether those addresses continue to outperform after selection is the question that matters, and it is rarely published.
You are always late
The transaction must occur, be indexed, be classified and be displayed before you see it. The price has already moved, and the position you are copying was entered at a price you cannot get.
For a large position accumulated over time, you see the end of the accumulation, not the beginning.
There is also a selection effect in what gets displayed. Trackers surface the positions that moved, which means you see concentrated bets and not the diversified book they sit inside. A position that is two percent of someone’s portfolio reads, on a tracker, like conviction.
The one genuinely useful version
Not "what are they buying" but "is this specific token held by addresses with long histories, or only by addresses created last week". That is a concentration and age question, computed from observable data, and it is informative about a token's holder base.
FBT Swap presents on-chain readings with their source and does not convert wallet labels into trade recommendations.