Everyone with a prediction to sell has a track record, and almost all of them are unreadable on purpose. Here is the short field guide I'd use to audit anyone's record — including my own — and the specific red flags that mean you're looking at a highlight reel, not a ledger.
Everyone with a forecast to sell has a track record, and almost all of them are built to be unreadable. Not unreadable as in complicated — unreadable as in designed so you can admire them but never audit them. This is the short field guide I'd hand a skeptic for taking apart anyone's record, mine included. It's mostly a list of red flags, because once you can name them you stop being impressed by the wrong things.
Red flag one: no denominator. 'I called the crash. I called the election. I called the war.' Out of how many calls, total? A record without a denominator is not a record, it's an anecdote in a suit. The first question for any forecaster is always 'and how many calls did you make that you're *not* telling me about?' A real ledger leads with the total — mine is [67 closed calls](/dashboard) — because the count is what turns a hit into evidence.
Red flag two: no visible misses. A track record with zero misses is not a perfect record; it is a record that is hiding them, and you already know this in every other part of life. Real forecasting produces misses. The only question is whether they're shown. Go looking for the misses; if you genuinely can't find any, you've found your answer. The biggest miss in this corpus, the [Maharashtra 2024 advisory](/intel/maharashtra-2024), is built as a flagship, and the calls I deliberately leave out are itemized on [the excluded list](/excluded) — including one I got right.