Models, Assumptions, and Evidence

A model is useful because it leaves something out. The important question is whether the omitted detail changes the conclusion we want to draw.

The inference chain

flowchart LR
    Q[Scientific question] --> M[Model choice]
    M --> C[Calculation]
    C --> O[Observable]
    O --> K[Claim]
    A[Assumptions] -. constrain .-> M
    V[Validation] -. tests .-> K

Each arrow is a place where uncertainty can enter. A clean result at the calculation stage does not automatically validate the claim.

A minimal quantitative check

Suppose a prediction depends on a parameter . A useful local sensitivity measure is

If a small plausible change in produces a large change in , the assumption deserves more attention than an additional decimal place in the final output.

Record decisions, not only values

For each model, keep a short decision log:

1. What physical or chemical behavior must the model represent?
2. What does the model intentionally omit?
3. Which observable can test whether that omission matters?
4. What result would cause us to choose a different model?