Skip to content

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.

What this guidance should change

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.

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.

How to evaluate this workflow

Mortgage QC sampling software should let an examiner reproduce the selected population. That requires more than saving the final loan list: preserve eligibility rules, cycle month, strata, method, parameters, seed, selected order, discretionary additions, and the point at which membership was frozen.

Buying criteria

  • Preserve the input population and exclusion rules
  • Record random seed or statistical parameters
  • Stratify channels when the written QC plan requires it
  • Prevent after-the-fact sample editing

Put this guidance into a controlled operating record.

Translate each requirement or method into a named control with an owner, effective date, scope, required inputs, acceptable evidence, review authority, exception path, and retained output. Link the configured control to the current primary source or approved policy instead of copying a sentence into a checklist with no change history.

Test the workflow with one complete cycle. Another qualified reviewer should be able to identify the eligible population, reproduce the sample or calculation, open the cited evidence, understand the response and rationale, see who approved the disposition, and verify how the result reached corrective action and management reporting.

Guidance, software, and automation do not decide legal applicability or replace agency, investor, institutional, or counsel judgment. Record the source review date, monitor changes, version the control, and preserve which rule and evidence state produced every report.

Why this page exists

Keep decisions human and evidence explicit.

Practical guidance tied to the review evidence a lender needs to preserve.

From evidence to conclusion

Put the guidance inside a complete review record.

Run five post-close reviews with the source evidence, calculations, dispositions, and reporting kept together.

Run five reviews free See exact pricing