Simulation as an Operating Surface

An operating change should be testable before it is made, and the output of the test should be its assumptions, not its headline number.

Care operations make a small number of consequential changes each year: enroll more patients, absorb a departure, move a cohort between programs, change outreach cadence, tighten thresholds, widen autonomy. Each is currently decided in a meeting on the basis of somebody's estimate, and the estimate is not recorded.

A simulation surface makes the estimate an artifact. It states what it assumed, what it projects, how confident it is not, and what could go wrong, and it names the weakest assumption explicitly rather than burying it. In the reference implementation an enrollment scenario names its own linearity as the largest reason to treat the result as directional, and identifies onboarding load, derived from 61 enrollments, as the weakest input.

The second property matters more than the first: simulations are recorded as evidence with their assumptions and model version, so a decision made from one can be re-examined later against what actually happened. That is what converts planning into a learning loop rather than a genre of meeting.

That operational simulation is a consulting service, not a feature; that models built on one organization's history will not transfer; and that showing projections in the operating system will cause them to be quoted as forecasts in board papers.

The last risk is real and largely a presentation problem, which is why the surface should refuse to render a single number without its band.

Retrospective scoring. Every simulation that precedes a real change is compared against the outcome, and the error is published inside the tool. A simulation surface whose errors are not tracked is a persuasion device.

How to communicate non-linearity honestly when the underlying model is linear. Whether simulating autonomy increases requires a different treatment from simulating operational changes, given the safety asymmetry. Whether teams anchor on the first scenario they run.