Methodology
How predictive alerts work
ZRG links leading indicators (export licences, customs statistics, MOFCOM notices, port data) to material risk scores using historical lag calibration. This page documents the methodology — not a performance guarantee.
Step 1
Signal ingestion
Public and licensed feeds: MOFCOM announcements, customs trade statistics, GDELT/news entities and curated regulatory sources. Signals are normalised per material (gallium, REE, graphite, etc.).
Step 2
Lag calibration
For each indicator we estimate the typical delay until spot market or supplier lead times move. Calibration uses historical pairs stored in data/leading-indicators/lag-calibration.json — sample sizes are published in the terminal accuracy widget.
Step 3
Confidence score
Confidence reflects calibration sample size, signal strength and portfolio relevance. It is a relative ranking aid — not a statistical guarantee. Example values in marketing materials are illustrative.
Step 4
Known limits
Black-swan events, unpublished policy shifts and missing tier-2 data can invalidate models. Predictive alerts complement — never replace — your own supplier due diligence and contractual safeguards.
Transparency commitment
We do not publish customer-specific accuracy unless agreed in writing. The /api/v2/prediction-accuracy endpoint shows resolved vs. unresolved alerts for your tenant when sufficient history exists.
Decision support only — not legal or investment advice. Methodology evolves with new calibration data.