Audit multivariate process/data-quality anomalies — audit_process

Computes a regularized Mahalanobis distance over selected person-level process metrics. Flags indicate review needs only; they are not cheating, identity, diagnosis, or intent classifications.

Usage

audit_process_anomalies(
  data,
  person = "person_id",
  metrics = NULL,
  alpha = 0.975,
  aggregate = TRUE,
  ridge = 1e-06
)

Arguments

data: Data frame.
person: Person identifier column.
metrics: Numeric process metrics. If omitted, usable numeric columns are selected.
alpha: Chi-square review quantile.
aggregate: If TRUE, aggregate metrics to person level before auditing.
ridge: Diagonal covariance regularization.

Value

An object of class "eye_process_anomaly_audit", stored as a named list, with components "table", "metrics", "alpha", "threshold", "center", "covariance", "caveat". It contains multivariate process/data-quality anomalies and associated metadata or diagnostics needed to interpret the result.