Ask a hedge fund analyst what their forecasting track record looks like and you'll get a vague answer. Ask a Superforecaster the same thing and you'll get a Brier score to three decimal places.
That gap is everything. And closing it is the single highest-leverage thing you can do for your decision-making — without learning any new financial concept at all.
The 5-minute setup that changes everything
You don't need software. You don't need a system. You need a list, a column for the probability you assigned, a column for what actually happened, and the discipline to fill both in.
Here's the entire format:
Date | Question | My probability | Resolution date | Outcome | Notes
That's it. Open a Google Sheet, paste those headers, save it. You've just done more than 95% of people who consider themselves "good at predicting things."
What to track (and what to skip)
Don't track: vague predictions ("the market will be choppy"), things you can't measure ("the team will struggle"), or political opinions where the outcome is contested.
Do track: anything that resolves clearly by a specific date. "S&P closes above 6,000 on Dec 31?" Yes. "Fed cuts rates at the September meeting?" Yes. "My friend Sarah gets the promotion she's interviewing for next month?" Yes — assuming you'll find out.
The best questions for a beginner are short-horizon (resolves within 1-3 months) and binary (yes/no, not "how much"). You want fast feedback. Long-horizon questions teach you nothing for years.
The Brier score, explained without math
A Brier score measures how calibrated your predictions are. Lower is better. A perfect forecaster scores 0. A coin flip scores 0.25. Most people score worse than 0.25.
You don't need to compute Brier scores in your first 30 days. Just track. After about 30 predictions, you'll notice your own patterns. After 100, the patterns become impossible to ignore.
The two most common patterns: people are overconfident at the extremes (90%+ predictions resolve "yes" only 70% of the time), and underconfident in the middle (50% predictions are actually closer to 50% than people think). Once you see your own version of this, you can correct it.
How to write questions that actually teach you
Bad: "Will the economy do well in 2026?" Too vague, never resolves.
Good: "Will Q2 US GDP growth, as reported by the BEA on July 31, be above 2.0% annualized?" Specific, dated, measurable.
Bad: "Will my startup succeed?" Definition unclear.
Good: "Will my startup hit $50K MRR by end of Q4 2026?" Clear bar, clear date.
The pattern: every question needs a number, a date, and a source. Without all three, you can rationalize the outcome after the fact and never improve.
The first 10 questions to add this week
Some easy starters that resolve within 90 days. Write down your probability for each before you check anyone else's prediction.
Will the next FOMC meeting cut rates?
Will the S&P 500 close above 6,200 within 30 days?
Will Bitcoin close above $110,000 by end of next month?
Will the next non-farm payrolls report come in above 150,000?
Will the EUR/USD exchange rate close below 1.10 within 60 days?
Will Polymarket's "US recession in 2026" contract close above 25% in 30 days?
Will gold close above $2,800/oz within 30 days?
Will the next CPI report show core inflation above 3.0%?
Will the next Fed dot plot show fewer cuts than the current one?
Will the 10-year Treasury yield close above 4.5% within 30 days?
Don't research first. Just write down your gut probability. The whole point is to find out where your gut is wrong.
The three rules that prevent you from cheating
One: write the probability down before you check anyone else's view. Once you've seen a Polymarket price, you can't honestly assign your own without anchoring.
Two: don't revise. If you wrote 60% on Tuesday, you wrote 60%. Don't update the cell on Wednesday because new information came in. Start a new entry instead.
Three: resolve every prediction. Even when you forget about it. Even when the outcome is embarrassing. Especially when the outcome is embarrassing.
What changes after 30 predictions
You start to notice when you're being overconfident. The internal sensation of "I'm sure about this" becomes correlated with how often you were actually wrong. You begin to flinch when you hear yourself say "definitely."
That flinch is the entire point. It's the same instinct professional poker players develop. The same instinct prediction market traders develop. It's the difference between someone who thinks they're a good forecaster and someone who actually is.
After 100 predictions, you'll have a sense of which domains you're calibrated in and which you're not. Most people are decent at predicting things in their daily life (will Sarah show up on time?) and terrible at predicting macro events (will the Fed cut?). Knowing which is which is enormously valuable.
The journal as a network effect
Once you have a track record, you can compare yours to others. Most prediction markets and forecasting platforms (Manifold, Metaculus, Good Judgment Open) automatically score you and rank you against everyone else.
You'll discover that some random freelance writer in Singapore is better calibrated than every economist in the US. You'll discover that your hedge fund friend is overconfident on macro and well-calibrated on micro. You'll start hiring forecasters by track record, not by credentials.
This is how the next generation of investing teams will be built. The talent acquisition mechanism is already being assembled — and a public Brier score is the new resume.
What Juno lets you practice
The fastest way to build a journal is on a real prediction market, where every contract you trade automatically resolves and scores itself. Your account balance is your track record. Your spread between buy and sell tells you your confidence interval.
Juno is being built specifically for this kind of practice — small stakes, fast resolution, calibration tracking by default. You can train yourself to forecast without risking serious money, and graduate to bigger positions once your Brier score earns the right.
The two-decimal lesson
Most people who claim to be good at predicting things have no data backing it up. The few who do are calibrated to two decimal places and can prove it.
The difference between those two groups is a Google Sheet with six columns, opened on a Sunday afternoon, and updated maybe ten minutes a week for a year.
You can start today. You'll be better than 95% of the financial press inside a month.