Simulation and prediction – MetaScience, Scientific Inquiry, and Agents

Epistemic purpose

Separate scenario exploration from a predictive claim. Inferring a latent parameter belongs to estimation even when a forward simulator is used.

Specify governing assumptions and boundary conditions; check numerics; run sensitivity or scenario analysis; produce forecasts and uncertainty; freeze predictions before evaluation evidence is accessed.
Inputs: the question, available evidence and provenance, constraints, prior claims, and remaining budget.
Outputs: inspectable artifacts, updated question/evidence state, and a claim record when a substantive claim is made.

Numerical verification, purpose, and uncertainty support computational claims. Predictive skill requires a suitable baseline and independent assessment; scenario calculations do not become forecasts merely by producing numbers.
Distinguish a warranted decision at the time from a claim’s later assessed adequacy; append follow-up without rewriting the original record.

Enter estimation for unknown parameters, hypothesis assessment for competing models, or robustness when performance depends on conditions; revise model purpose if evidence cannot support the intended claim.
Neighbouring profiles: estimation, hypothesis, robustness, theory.

Treat fit as truth, simulation as a physical experiment, or agreement among related models as independent evidence.

Oreskes et al. (1994) distinguishes partial confirmation from complete verification/validation of natural-system models; Beven and Freer (2001) treats equifinality; Schorlemmer et al. (2018) describes earthquake-forecast evaluation.
Cross-field comparison: Shmueli (2010) distinguishes explanatory and predictive objectives across statistical applications. Borrow the distinction while retaining domain-specific physical checks.
Science of process/impact: No corpus result is available for this profile. Forecast scores quantify particular predictive gains and do not measure every kind of advance.
The actions and transitions above are design hypotheses derived from these sources, not established optimal policies. Missing process chronology remains unknown.

Compare forecasts with a stated baseline on held-out observations; retain scenario usefulness as a separate assessment. Test a model with good fitting error and poor transfer.
Use the historical evaluation to choose a source record, preserve evidence boundaries, and assess a meaningful gain. A synthetic control tests a constructed case, not a historical discovery.
Assessment definitions · Agent implementation