Identity
Patient matching is the quiet failure in every HIE

The cost is asymmetric
A missed match delays care; a wrong match corrupts two records at once. Design the thresholds so the expensive mistake is rarer than the cheap one.
Demographics lie more than they used to
Moved addresses, married names, shared phone numbers. Deterministic matching on name and date of birth alone fails at scale; probabilistic matching without audit fails the regulator.
Record the decision
Every match and every non-match should be explainable after the fact: which fields, which weights, which reviewer.