What the Simulation Owes You: Fidelity Budgets and Honest Labels
A simulation that does not declare its assumptions is a confidence engine for bad decisions. The fidelity budget is the discipline of naming what the simulation does not model, how much that omission matters, and what would have to change for the answer to move. Every simulation owes its reader three honest labels: what is modeled, what is approximated, and what is absent. The fidelity budget is the difference between a testbed and a fairy tale.
The three labels
Modeled: the entities, rules, and dynamics the simulation actually simulates — measured, if possible, against real data. Approximated: the parts represented by a simpler stand-in — a flat cost curve, a homogeneous customer segment, a fixed churn rate — with the direction and size of the approximation error stated. Absent: the dynamics deliberately excluded — regulation, competition beyond the modeled set, seasonal effects — with a note on which decisions they could flip. The three labels are the fidelity budget.
Fidelity as a budget, not a virtue
More fidelity is not always better; it is always more expensive. The budget is the allocation: how much fidelity each part of the model earns, given the decision it supports. A pricing test (SIM-2) needs customer behavior fidelity; it may not need logistics fidelity. The gate (SIM-1, SIM-7) passes when the budget is declared and the absent dynamics are not load-bearing. An honest low-fidelity simulation beats a dishonest high-fidelity one, because the honest one can be improved and the dishonest one cannot be trusted.
The receipt carries the budget
The fidelity budget is part of the promotion receipt — it travels with the model into the canary (SIM-5) and into the live deployment. When the canary disagrees with the simulation, the budget names the likely culprit: an approximation that turned out load-bearing. The loop then re-budgets, re-simulates, and re-tests. The fidelity budget is what makes the simulation layer a learning system instead of a one-shot oracle.
Grounded in the SIM series, the coherence article (SIM-3), the promotion criteria (SIM-7), and the S10 research-debt methodology. Eighth article in the Round D simulation track.
