Experiments

TBD - need to think about this, but is key

- Validate DAG semantics with clinical experts.
- Measure improvements in similarity retrieval and predictive modeling tasks; use a DAG and modify a side branch to generate synthetic "patients like me"? Can we modify something and have clinical expert measure to see if it makes sense?

