SAMPLING

Mortgage QC sampling: random, statistical, and discretionary

A sample is only defensible if the population, method, seed or selection procedure, strata, and final membership are preserved. Changing the file list after seeing results destroys the audit trail.

Operational guidance, not legal advice. Validate the workflow against your written QC plan, current agency/investor guidance, and counsel.

Random selection

ExactClose ranks each eligible loan with a cryptographic HMAC derived from the locked seed and loan identity. Re-running the same population and seed produces the same order.

Channel stratification

Retail, TPO, and correspondent populations can carry different risks. The default 10% method selects within each channel so a small channel is not lost inside the total population.

Statistical method

The statistical calculator records expected defect rate, precision, confidence factor, finite-population correction, and resulting count. The institution must validate that those parameters match its QC plan and current investor requirements.

Discretionary reviews

Targeted high-risk or adverse-event reviews are valuable but should remain visibly separate from the random sample and its defect-rate denominator.