ML Model Evaluation with Task-Correct Metrics
This is a ready-made prompt from the Zayenha Prompt library in Programming & Tech, written for claude-sonnet-4-7. It has a main prompt to copy into the model, built-in quality guards, and a verification prompt to run in a second window to review the answer. The full text needs the Plus plan; the description is open to all.
Design a model-evaluation methodology that picks task-correct metrics and addresses imbalance, generalization gap, and bias, interpreting results, without inventing metrics or unmeasured numbers.
Your current plan: free
Evaluating an imbalanced fraud-detection model before production, choosing correct metrics for a diagnostic model, preparing an auditable evaluation report for a model-risk board.
Built-in quality guards
No fabricated metrics, functions, or unmeasured performance numbers
Avoid printing individual records and respect PDPL in sensitive-group analysis
Cover metrics, protocol, error analysis, generalization, and fairness without truncation
Adherence to the six-section structure, results table, and decision recommendation
Frequently asked questions
How do I use this prompt?
Copy the main prompt and paste it into claude-sonnet-4-7 or the model you use, replace what is in square brackets with your details, then run the verification prompt in a second window with the answer pasted in, and review it before relying on it.
Is this prompt free?
Its full text requires the Plus plan or higher. Free prompts you can copy at once without an account are on the ready-made Arabic prompts page.
Which model was this prompt written for?
It was written for claude-sonnet-4-7 and follows how that model works; you can run it on other models too, reviewing the result, since no single wording guarantees the best result on every model.