Write down one prediction every day, with a date and a confidence level, and check it later.

Takes maybe ninety seconds. And it would quietly put me ahead of an enormous share of people, because almost nobody does it — not executives, not pundits, not most scientists. Humans run on vibes about their own judgment. They remember their hits, forget their misses, and retroactively edit what they "always knew." So their model of their own accuracy never improves, no matter how smart they are or how much they read.
The person who keeps a scored prediction log gets something rare: a real calibration signal. Within a year they know things like "I'm overconfident about deadlines by 40%," "my first read on people is better than my considered one," "I'm reliably wrong about how markets react to news." That's not trivia — it changes which of your own thoughts you trust, which is upstream of every decision you make. Everyone else is navigating with an uncalibrated compass and doesn't know it.
The reason it outcompetes rather than just improves: the advantage compounds and it's invisible. Nobody can copy it by observing you, the way they could copy a diet or a morning routine. And the moat is behavioral, not informational — the practice is fully public knowledge (Tetlock wrote the book) and people still won't do it, because writing down "70% confident" and later reading "wrong" feels bad in a way vague optimism never does. The barrier to entry is tolerance for small, frequent ego damage. That's a barrier most people will never cross, which is exactly what makes the other side of it uncrowded.
Honorable mention: going to bed on time. Less interesting, probably bigger effect size. But sleep improves the hardware — the prediction log improves the operating system.