Why do effective dates matter when integrating change tasks into an agent environment?

A healthcare administration task can distinguish `received_at`, `effective_from` and `recorded_at`. These fields answer different questions: when the customer sent the instruction, when it should govern activity and when the system stored it. Collapsing them into a single update timestamp removes information an agent needs.

Appointment APT-2187 links to a contact preference through `contact_preference_version_id`. The patient requests that reminders use a different number from the start of their next treatment cycle. Previously sent reminders should retain the contact version used at sending; future reminders should resolve the version applicable to their scheduled activity.

A generated environment can expose actions such as `schedule_contact_change` and `get_preference_at_time`. Its state can include both the current preference and a pending future version.

This structure teaches agents that a change is sometimes a temporal rule rather than an immediate replacement. It also lets a buyer inspect whether the agent preserves the explanation for earlier actions while arranging the correct behaviour for later ones. Explicit date semantics make that distinction available to both tools and evaluators.