forecasting

Forecasting

Purpose

Make specific, falsifiable predictions with calibrated confidence levels. Track accuracy over time using Brier scores. Apply superforecasting methodology (Tetlock/Good Judgment Project) to any domain — technology trends, project outcomes, market shifts, competitive moves, risk assessment.

When to Use

• User asks for a prediction or forecast on any topic.
• Strategic reflection identifies a decision that depends on uncertain futures.
• Surplus compute is available and a prediction review is due.
• A previously made prediction is approaching its resolution date.
• Deep reflection surfaces a trend worth formally tracking.

Superforecasting Principles

1. Triage — Focus on questions where effort improves accuracy. Ignore questions that are either trivially knowable or fundamentally unknowable.
2. Fermi decomposition — Break big questions into smaller, estimable components. "Will X happen?" → "What's the base rate? What's different this time? What signals would I expect to see?"
3. Balance inside and outside views — Start with the reference class (base rate from historical analogues), then adjust with specific evidence. Never skip the outside view.
4. Update incrementally — Bayesian updating. New evidence shifts confidence by small amounts, not dramatic swings. Avoid overreaction.
5. Calibration over precision — A well-calibrated 60% is better than an overconfident 90%. Your 70% predictions should come true ~70% of the time.
6. Distinguish noise from signal — Most new information is noise. Ask: does this actually change the probability, or does it just feel important because it's recent?
7. Consider contrarian views — Actively seek evidence against your current position. What must be true for the opposite outcome?
8. Post-mortem every resolution — When a prediction resolves, analyze WHY you were right or wrong, not just whether. Update process, not just beliefs.
9. Express uncertainty numerically — "Likely" is ambiguous. 70% is not. Use the probability scale below.
10. Separate confidence from conviction — High confidence (90%) means high probability. Strong conviction means you've thought deeply. You can have low confidence with strong conviction (you've analyzed it thoroughly and it's genuinely uncertain).