Find where a forecast fails, not just its average error

From [PREDICTIONS], [ACTUALS] and [BUSINESS COSTS], design an error audit by horizon, segment and season. Define the unit, alignment key and metrics, then show a fictional sample calculation. Include missing actuals, zero denominators and one-off shocks as explicit cases. Recommend a chart and a table that expose both typical error and costly tails. Do not claim a model improvement without a held-out comparison. Check for future information leaking into predictions and report sample size for every slice.