MMP.17:8 - Common Anti-Patterns and How to Avoid Them

Invited mistake	Repair
Declare success from small average error while a limit crossing is wrong.	Compare the response and error in the receiving operation, including the affected input region.
Treat an optimization objective as the constructed surrogate.	Supply the fitting procedure and the evaluator; inspect computational failure separately from inadequate cases or family.
Add cheap cases whose model misses the needed regime.	Examine the cheap-to-source discrepancy and compare correction with source-only construction at the same total effort.
Differentiate or invert a value surrogate without carrying its approximation.	Include the required operation in construction and error analysis, or keep the source for that operation.
Use the same cases repeatedly for tuning and claim untouched assessment.	Treat them as construction information and qualify the remaining further-use result.
Read a fitted interval as a guarantee against missing source structure.	Retain its conditions and return to source-model criticism or the subject account when those conditions change.