MMP.14:8 - Common Anti-Patterns and How to Avoid Them

Anti-pattern	Repair
Fit the overall mean and infer adequacy for every conditional forecast	Compare the feature on which the receiving use depends.
Call every residual an error in the model	Compare with the model’s permitted variation and relevant numerical or recording error.
Simulate latent states and compare them with selected records	Pass replications through the actual recording law.
Treat a predictive tail fraction as the probability that the model is wrong	Retain the reference distribution and its actual probability statement.
Tune repeatedly on a “test” set and report an untouched test result	Treat that set as development data; qualify or separately assess the adaptive procedure.
Replace an inconvenient observation or inflate noise until a check passes	Examine the observation basis and competing model components; preserve legitimate unusual outcomes.
Improve prediction and claim the causal explanation is established	Retain the predictive claim; obtain identification and subject grounds for the causal claim.
Require a new experiment after every mismatch	Use C.11.DUA to choose among available repair, narrower use, conditional continuation and worthwhile inquiry.