LAMP (Latent-state Audit with Matched-cohort Protocol) is a framework for auditing whether a model uses future, forbidden, or latent information in time-series prediction. It takes a time-series prediction table plus a YAML audit contract and returns a machine-readable dossier separating valid early signal from leakage, shortcuts, and label-adjacent contamination. Full details: see README on GitHub: https://github.com/nelacye/LAMP_IRIS_operational_physiology