"There is no start date and no end date. The timeline runs backward into forgotten Tuesdays and forward into appointments not yet kept."
Both Directions Are First-Class
Because health has no terminal state, its timeline is unbounded in both directions, and both are ordinary. Past scatter: a 2025 check-up scanned today files under 2025; a symptom Dad forgot moves to last Tuesday. Future scatter: a scheduled appointment or a planned training block sits in next month. This is the Track 3 date-picker mechanic seen from the lifecycle angle — captured_at stays the true moment of writing while target_date ranges freely across the whole line, past and future, and the two never collapse into one.
Interventions Are Just Dated Crumbs
The tempting move, once you name "intervention" as a loop phase, is to build an Intervention entity with a start, a status, and a workflow. Forge refuses. A new prescription, a changed training block, a procedure — each is an ordinary dated crumb, no different in kind from a symptom or a meal. Analysis reads the timeline against these crumbs: it finds the day the training block changed and reads the trend on either side of it. The intervention is a marker in the stream, not a ceremony with its own machinery.
Why This Shape Is Sturdy
An append-only stream of dated events, projected into current state, is a well-worn and durable pattern — it is event sourcing in everything but name. It means the record is never rewritten, only added to; the "current picture" is always a projection you can rebuild; and past, present, and future all live on one axis instead of in three different systems. When a domain has no natural completion, this is the shape that fits it — and it is exactly why the storage layer in Track 7 can rebuild every derived view by replaying the log.