borg_validate: Validate Complete Evaluation Workflow

In BORG: Bounded Outcome Risk Guard for Model Evaluation

Validate Complete Evaluation Workflow

Description

borg_validate() performs post-hoc validation of an entire evaluation workflow, checking all components for information leakage.

Usage

borg_validate(workflow, strict = FALSE)

Arguments

workflow	A list containing the evaluation workflow components: data The full dataset train_idx Integer vector of training indices test_idx Integer vector of test indices preprocess Optional preprocessing object(s) model The fitted model object predictions Model predictions on test data metrics Computed evaluation metrics
strict	Logical. If TRUE, any hard violation causes an error. Default: FALSE (returns report only).

Details

borg_validate() inspects each component of an evaluation workflow:

1. Split validation: Checks train/test index isolation

2. Preprocessing audit: Traces preprocessing parameters to verify train-only origin

3. Feature audit: Checks for target leakage and proxy features

4. Model audit: Validates that model used only training data

5. Threshold audit: Checks if any thresholds were optimized on test data

Value

A BorgRisk object containing a comprehensive assessment of the workflow.

See Also

borg for proactive enforcement, borg_inspect for single-object inspection.

Examples

# Validate an existing workflow
data <- data.frame(x = rnorm(100), y = rnorm(100))
result <- borg_validate(list(
  data = data,
  train_idx = 1:70,
  test_idx = 71:100
))

# Check validity
if (!result@is_valid) {
  print(result)  # Shows detailed risk report
}

BORG documentation built on March 20, 2026, 5:09 p.m.