Verification Is Not Validation
Verification confirms a model implements its specification correctly; validation confirms the specification matches reality — and a flawlessly verified model can be entirely invalid.
Field Notes
Short essays on deterministic execution, scenario design, and operational analysis.
These notes treat simulation as an analytical system, not only a visualization layer. The focus is practical: repeatable runs, explicit assumptions, and scenario definitions that can be reviewed over time.
Each entry starts from one question: what must be true before a simulation result can support a decision?
Verification confirms a model implements its specification correctly; validation confirms the specification matches reality — and a flawlessly verified model can be entirely invalid.
The entity-component-model is a representation decision before it is a performance one — it makes a simulation's ontology explicit, its state inspectable, and its behavior composable.
Game engines and simulation engines are optimized against opposite definitions of correct — which shows most plainly in what each sacrifices under load.
Fidelity is a profile shaped to a decision, not a dial turned up — bounded by what you can validate and governed by your weakest relevant component, not your most detailed one.
Traceability is the backward question you can answer, not the log you keep — and a deterministic core turns it from always-on logging into on-demand reconstruction.
Determinism was never the generator's job — it belongs to the artifact a non-deterministic author emits, and the whole discipline lives at the gate where a draft becomes a frozen, validated, replayable object.
A scenario is a designed experiment, not a description of a situation — and its design fixes what the analysis can conclude before a single run executes.
Calling simulation "decision infrastructure" is a load-bearing claim — it commits the system to obligations a visualization tool never has to meet.
Deterministic simulation is a containment problem — drawing a hard boundary around the model core so every source of variance is either controlled inside it or denied a path into it.
Why deterministic execution matters for repeatable simulation, operational analysis, verification, and defensible decision-making.