Sim Coherence: Keeping a Synthetic World Self-Consistent
A simulation is only as good as its coherence — the property that the world stays self-consistent from run to run and from moment to moment. An incoherent simulation is worse than no simulation: it produces confident wrong answers that pass the gate and poison the canary. Coherence is the first gate on the promotion ladder (SIM-1), and it is the hardest to maintain as a simulation grows in scope.
What coherence means concretely
Coherence has three faces. Temporal coherence: the state of the world at step N+1 follows from step N under the model’s rules — no teleporting inventory, no resurrected customers. Logical coherence: the rules do not contradict each other — a customer cannot churn and renew in the same tick, a vendor cannot sell below its own floor. Statistical coherence: the aggregate behavior is stable across seeds — rerunning the simulation with a different random seed shifts the outcome within its stated variance, not to a different story. Each face is measurable, and each has a threshold in the fidelity budget (SIM-8).
The coherence score
The kingdom’s answer to coherence is a score: every simulation run produces a coherence metric, and the metric is part of the promotion receipt. The score is computed from invariant checks (temporal, logical, statistical) applied to the run’s event log. A run that violates an invariant is flagged, and a model whose runs repeatedly violate invariants cannot promote. The score makes coherence auditable instead of aspirational — the same discipline the audit layer applies to agent actions in the live stack (S7.7, S7.15).
Coherence as a design target
Coherence is not only measured; it is designed. The simulation’s world model should be built from the same entities and rules the live system uses, so that the sandbox and the stack speak the same language. When the simulated world drifts from the real one, the drift is data — it names the assumption that is wrong. Coherence is the discipline that turns a toy world into a testbed.
Grounded in the SIM series, the SECTOR07a-34 vision-score article, the S7 audit work, and the fidelity-budget article (SIM-8). Third article in the Round D simulation track.
