15 FIELD GUIDES · 2 WORKED EXAMPLES
Does the model predict?
Check dimensions, convergence, uncertainty and independent predictions before treating simulation output as physical evidence.
Models & numerical evidence
Start with the question closest to your experiment. Each guide separates the mechanism, an illustrative example, and the evidence needed for the next decision.
How do units catch modeling errors?
Check the dimensions of equations before trusting numerical output. Learn how a unit audit catches conversion errors while leaving physical validity open.
Read the guide ↗FIELD GUIDEWhen do dimensionless groups make models comparable?
Use dimensionless groups to compare physical regimes and scale experiments. See why matching one ratio may not preserve every important mechanism.
Read the guide ↗FIELD GUIDEWhy can boundary conditions dominate a simulation?
Learn how boundary conditions define a simulation problem, how artificial boundaries bias results, and how targeted checks expose hidden assumptions.
Read the guide ↗FIELD GUIDEHow much mesh refinement is enough?
Assess mesh convergence using a chosen output, systematic refinement, and an explicit tolerance. Learn why cell count alone does not establish accuracy.
Read the guide ↗FIELD GUIDEHow do you separate time accuracy from numerical stability?
Check timestep convergence with fixed physical outputs and a controlled spatial model. A stable transient calculation can still miss peaks and timing.
Read the guide ↗FIELD GUIDEWhat should a conservation residual tell you?
Use mass, energy, or charge balances as independent simulation checks. Normalize residuals carefully and distinguish storage from genuine imbalance.
Read the guide ↗FIELD GUIDECan the data determine your fitted parameters?
Distinguish a good fit from identifiable parameters. Use compensating parameter changes and targeted measurements to discover what the data constrain.
Read the guide ↗FIELD GUIDEWhich inputs actually control the prediction?
Use sensitivity analysis to rank influential inputs without confusing influence with uncertainty. Compare local changes and check interactions explicitly.
Read the guide ↗FIELD GUIDEHow does information leak into a test set?
Protect model evaluation from leakage through preprocessing, repeated subjects, and future information. Match the split to the prediction you need.
Read the guide ↗FIELD GUIDEWhat makes a baseline model useful?
Choose baselines that expose whether complexity adds value. Compare simple references on the same data split, metrics, and operating conditions.
Read the guide ↗FIELD GUIDEWhat patterns should you look for in residuals?
Inspect residuals against predictions, time, and operating conditions. Learn how structured errors reveal missing behavior that a single score can hide.
Read the guide ↗FIELD GUIDEHow should you choose between competing models?
Choose a model using the decision, evidence, and evaluation design. Balance predictive performance, complexity, physical constraints, and transfer limits.
Read the guide ↗FIELD GUIDEHow do input uncertainties reach the output?
Propagate input uncertainty through a measurement or simulation model. Include covariance, test linear approximations, and separate omitted model error.
Read the guide ↗FIELD GUIDEWhat does agreement between two solvers establish?
Compare independent solvers without treating agreement as proof. Separate shared assumptions from independent implementations and inspect disagreements.
Read the guide ↗FIELD GUIDEWhen does a prediction become extrapolation?
Identify predictions beyond supported data and physical regimes. Learn how distance, mechanism changes, and confirmation experiments limit extrapolation.
Read the guide ↗PUT THE IDEAS TO WORK