Forecast Attribution
What moved each served forecast: additive driver terms that reconcile exactly to the served q50, and the decomposition of what changed since the previous issuance. These terms explain what moved the forecast — not what moved the market.
Measured limit (D3 negative control): the size of a driver contribution is not evidence that the driver is informative. At 24 months, most cells' attributions are no larger than what the same machinery produces on an innovation-matched noise target — so driver breakdowns render mostly at the 6-month horizon and are progressively refused at longer ones, per each record's own measured flag. The forecast's skill is a separate claim, measured separately.
Pick a metro to see why its forecast is what it is. Attributions exist only as explanations of served forecast records — everything else states its own absence.