What a simulation can and cannot tell you

Review status: draft pending scientific sign-off.

Alive's worlds are simulation models: explicit, inspectable descriptions of a workflow, the people and staff in it, and the systems they use, run many times with controlled randomness. They are not digital twins. A digital twin, as defined by the US National Academies (2024), is continuously updated from its real counterpart and used to inform decisions about it; Alive's worlds are not connected to any real system and are not calibrated to one.

What a run can show

• How a workflow behaves under its stated assumptions: queues, delays, losses, capacity use.
• Software failures that the run actually observed, such as rejected writes or unanswered requests.
• How much a change matters relative to a baseline in the same world, with an interval that reflects simulation noise.
• Which assumptions a conclusion depends on, so you know what to measure next.

What it cannot show

• Real clinical outcomes, health benefits or harms.
• How real people will behave. Synthetic populations, including published generators such as Synthea, diverge substantially from real care data on many measures.
• That a real product works. Mock interventions show what follows if their assumed effects hold.

How to read a comparison

Within a world, every variant sees the same synthetic people and the same per-person random draws. The difference between variants is therefore estimated precisely, but it is a difference in the model. If an interval includes zero, the model cannot distinguish the variants at that number of replications.