Writing Strong Predictions
- ·Make predictions falsifiable — they must be clearly true or false at end date.
- ·Be specific: include a measurable metric, direction, and magnitude (e.g. "CTV CPMs will rise 15%+ in Q3").
- ·Avoid vague language like "may", "could", or "might" — commit to a direction.
- ·Set realistic timeframes: 30–90 days for tactical signals, 90–180 for strategic ones.
Calibrating Confidence Well
- ·Confidence is a calibrated probability — 80% means you genuinely believe there is an 80% chance the prediction is correct. It must reflect real certainty.
- ·Top forecasters score 65–75% average confidence — high enough to be decisive, low enough to reflect real uncertainty.
- ·Avoid anchoring at round numbers (70%, 80%). Use 63% or 72% to force genuine thinking.
- ·If you're regularly above 85% and not scoring above 85% accuracy, you are overconfident — lower your stated confidence to match your real hit rate.
- ·If you're below 55% regularly, signals may be too speculative to act on.
Improving Accuracy Over Time
- ·Top-quartile operators score 60–70% accuracy across completed signals.
- ·Review every wrong signal: classify the miss type (Total Miss, Timing Off, Nuance) to find patterns.
- ·Track evidence weekly — signals with active tracking logs have ~20% better outcomes.
- ·Calibrate: if your accuracy consistently exceeds your confidence, you're being too conservative.
Using Why Points & Actions
- ·Write at least 3 "Why" points — this forces you to stress-test the logic before committing.
- ·Include both confirming and counter-evidence in your why points for balance.
- ·Define concrete actions: what will you actually do if this signal plays out?
- ·Leverage/counter steps are most valuable — they turn predictions into strategy.
Input Weights & Data Sources
- ·Weight inputs that are most relevant to your category — SEO signals matter more for audience predictions.
- ·Use at least 2–3 data sources from the Sources library per signal.
- ·Operator Input weight should reflect how much proprietary data you have vs. public signals.
- ·Earnings signals are leading indicators — weight them higher for revenue-linked predictions.