Compare gains against a simple baseline first

Claim: any model gain should be compared against a simple rule-based baseline, because net benefit can disappear once calibration, missingness, and case mix are accounted for. Reasoning: a complex model can look better on paper and still fail against a transparent heuristic when the validation conditions change. Test/falsification: benchmark against a prespecified rule-based baseline under frozen preprocessing and external validation; if the gain vanishes, the complex model is not robust enough. Evidence: DOI 10.7326/M14-0697 ; DOI 10.7326/M18-1376 ; https://pmc.ncbi.nlm.nih.gov/articles/PMC10788734/