FORECAST AUDIT / LOCAL DATA ONLY
See what the average hides.
Reconstruct an illustrative multi-model forecast, expose disagreement, and find the assumption worth checking next.
DIRECTIONAL RANGE
MODERATE AGREEMENTPlausible range -9 to +45
Balanced sample loaded. Stress inputs to inspect sensitivity.
AUDIT VERDICT
Useful direction, limited confidence.
CEILING
What changed
The ensemble leans positive, but moderate vote spread limits confidence.
Fragile assumption
Momentum carries the largest weighted influence.
Model contributions
vote × reliability × independenceNext checks
highest information value firstInspect the scoring method
Center = weighted mean of votes. Range widens with vote dispersion, volatility, liquidity stress, event risk, horizon, stale data, and correlated models. Confidence is capped by agreement, freshness, independence, and market stress. These are teaching assumptions, not calibrated probabilities.
METHOD
Confidence should fall when evidence conflicts.
An ensemble is not automatically stronger than one model. Similar training data can create correlated errors. Old inputs can make precise-looking votes irrelevant. A transparent audit keeps the disagreement visible and turns uncertainty into a concrete verification plan.