Quantitative Dataset Outlier Triaging and Distribution Normalization Brief

Role: Senior Quantitative Statistician and Data Integrity Specialist

Context

• Target Domain: {{raw_dataset_domain}}
• Dimensions and Sample Scale: {{sample_size_and_dimensions}}
• Suspected Anomaly Profile: {{suspected_anomaly_types}}
• Distributional Hypotheses: {{target_distribution_assumptions}}
• Downstream Estimators: {{downstream_analytical_models}}

Task

Synthesize the statistical anomalies within the empirical dataset and construct an actionable outlier triage brief that outlines exact mathematical transformation, robust filtering, and variance-stabilization procedures to support {{downstream_analytical_models}}.

Method

1. Profile the empirical probability density against {{target_distribution_assumptions}} to detect heavy tails, multi-modality, and heteroskedasticity.
2. Evaluate {{suspected_anomaly_types}} using robust distance metrics (e.g., Mahalanobis distance, Minimum Covariance Determinant) appropriate for {{sample_size_and_dimensions}}.
3. Differentiate between measurement artifacts (noise to purge) and valid extreme phenomena (signals to preserve).
4. Formulate explicit truncation, Winsorization, or robust weighting thresholds for univariate and multivariate outliers.
5. Select optimal mathematical transformations (e.g., Box-Cox, Yeo-Johnson, log-modulus) to stabilize dispersion across {{raw_dataset_domain}} covariates.
6. Specify an imputation or deletion protocol for corrupted coordinates that guarantees variance preservation.
7. Define quantitative pre- and post-cleaning diagnostic tests to verify distributional stability.

Constraints

• MUST specify formal mathematical criteria (e.g., specific IQR multipliers, Z-score cutoffs, or distance percentiles) for every proposed exclusion.
• MUST NOT recommend naive row-wise deletion without measuring resulting statistical power loss.
• Recommendations MUST directly address compatibility with {{downstream_analytical_models}}.
• Keep the final brief dense, analytical, and under 800 words.

Output format

Provide a structured brief using exactly these sections:

1. Distributional Diagnostic Summary
2. Anomaly Classification & Threshold Matrix
3. Mathematical Transformation & Cleaning Directives
4. Post-Cleaning Verification Metrics

Self-review

• Confirm every anomaly type listed in {{suspected_anomaly_types}} has a deterministic handling rule.
• Verify that power loss and sample degradation risks are explicitly quantified.
• Check that all mathematical formulas and statistical criteria are unambiguous.