Building a UGESP Validation File for Every Skills Assessment You Use

A UGESP validation file is the evidence folder you hand to counsel when a rejected candidate, a plaintiff's lawyer, or a federal investigator asks why your skills assessment screened someone out. It is not a certificate and not a vendor badge. It is a set of documents showing what the job requires, what the assessment measures, how you established the link between the two, and when you last checked that the link still holds. Most US employers running pre-employment testing have no such file, and they find out in the week they need it.

Key Takeaways

• Under 29 CFR 1607.3(A), a selection procedure that produces adverse impact is treated as discriminatory unless validated in accordance with the guidelines.
• Content validity is the realistic route for most frontline roles, because criterion-related validation needs a sample size most employers cannot reach.
• Every validation strategy rests on a job analysis, which 29 CFR 1607.14(A) requires.
• The employer using the procedure, not the vendor, is asked to produce the evidence.
• A June 9, 2026 Department of Justice opinion found the EEOC's disparate-impact guidelines unconstitutional, but the guidelines remain codified at 29 CFR Part 1607 and private plaintiffs can still bring disparate-impact claims.

What 29 CFR 1607.3(A) Asks You to Prove

The rule is short and it does most of the work. Under the Uniform Guidelines on Employee Selection Procedures (29 CFR Part 1607), any selection procedure with an adverse impact on the hiring or promotion opportunities of a race, sex, or ethnic group is considered discriminatory unless it has been validated in accordance with the guidelines. Adverse impact is the trigger; validation evidence is the answer.

Employers routinely miss a second obligation. Section 1607.3(B) says that where two procedures are substantially equally valid, use the one with less adverse impact, and that a validity study should investigate suitable alternatives.

The 2026 Federal Shift, and Why the File Still Matters

On June 9, 2026, the Department of Justice's Office of Legal Counsel issued a memorandum opinion concluding that the US Equal Employment Opportunity Commission's guidelines on disparate-impact liability under Title VII, including the Uniform Guidelines, are unconstitutional. On July 31, 2026, the Office of Personnel Management removed references to the guidelines from federal civil service regulations, while keeping its job-analysis requirement intact.

Three things follow, and employers are getting all three wrong. An Office of Legal Counsel opinion is executive-branch guidance and does not bind federal courts. The guidelines remain codified at 29 CFR Part 1607 today. And disparate-impact liability sits in the statute at 42 U.S.C. 2000e-2(k), added by the Civil Rights Act of 1991, resting on precedent starting with Griggs v. Duke Power Co., 401 U.S. 424 (1971). Federal enforcement appetite has changed; private plaintiffs and state law have not.

The Three Validation Strategies in 1607.5

Section 1607.5(A) accepts three types of validity study, and your choice is driven by the role and your headcount rather than by preference. Technical standards for each sit in section 1607.14.

• Criterion-related - What the evidence must show: Empirical data that scores predict or significantly correlate with important elements of job performance.; When it is practical: High-volume roles with many incumbents, a stable job definition, and a usable performance measure.; Usual blocker: Sample size. The guidelines call this technical feasibility and do not require you to hire people to reach a sample.
• Content - What the evidence must show: Data showing the content of the procedure is representative of important aspects of performance on the job.; When it is practical: Most frontline, skilled trade, and administrative roles where the work can be observed and sampled directly.; Usual blocker: Only fits observable work behaviors and demonstrable knowledge, not abstract traits.
• Construct - What the evidence must show: Data showing the procedure measures identifiable characteristics determined to be important to successful performance.; When it is practical: Rarely. Mostly research settings, or where an established measure is already tied to the job.; Usual blocker: Heaviest evidentiary load of the three. You need both measurement evidence and a job link.

Why content validity is the route for most frontline roles

Content validity wins on availability, not on prestige. A criterion-related study needs enough incumbents, a performance criterion not contaminated by rater bias, and time. A 40-person distribution operation cannot produce that for its picker role and is not obliged to try. What it can do is document that the picking task in its assessment mirrors the picking task in the aisle, which shows you sampled the job rather than guessed at a trait. Generic aptitude batteries carry the opposite problem, as Avoiding Adverse Impact IQ Testing Legal Risks sets out, and the measurement side is in Ensuring Accuracy: Validity and Reliability in Skill Assessments.

Running a Job Analysis That Will Hold Up

The job analysis is the load-bearing document, and it is the one employers skip. Section 1607.14(A) requires a validity study to be based on a review of information about the job, and any method may be used that yields what the strategy needs.

A workable process for a single role takes about a day. Observe the work for a full shift and write down what people actually do, in task language rather than job-description language. Interview two or three strong incumbents and their supervisor separately, and have them rate each task on frequency, importance to the outcome, and the consequence of doing it badly. Keep the tasks that are frequent, important, or high-consequence, then write the link explicitly: this task appears in the job at this frequency, therefore this assessment item is in the test. Date it and name the contributors.

What a Vendor Technical Manual Must Contain

A usable manual states what the instrument measures and what it does not, the validation strategy used, the reference population with its size and composition, reliability figures with the method behind them, adverse impact results by group on the vendor's own data, recommended cutoff scores with reasoning, and whether the evidence supports pass or fail use only or also supports ranking. Section 1607.5(G) treats ranking as requiring more evidence than screening when it produces greater adverse impact. It also needs a transportability discussion: the job conditions under which the evidence can be borrowed for a similar role. If your warehouse is nothing like the study sites, the evidence travels poorly.

The buying signal when a vendor cannot supply one

Treat the absence of a technical manual as a finding, not a formality. A vendor offering a case study, a fairness badge, or an accuracy percentage instead of documented validation evidence is telling you the evidence has never been audited. Ask three questions and listen to the pauses: may we see the full technical manual; what were the adverse impact results by group; and will you support an audit against our own applicant flow data. Role-specific tests in our skill tests library are built so content maps to observable job tasks.

What Goes in the File

Keep one file per selection procedure per job family, with one named owner.

• Dated job analysis with raw task ratings and named contributors - Why it is in the file: Required basis for any validity study under 1607.14(A).
• Task-to-item mapping for the assessment - Why it is in the file: Carries the content validity argument in a single readable page.
• Vendor technical manual, full version - Why it is in the file: Supplies the validation evidence you did not generate yourself.
• Cutoff score decision memo - Why it is in the file: Section 1607.5(H) expects cutoffs consistent with normal expectations of acceptable proficiency.
• Annual adverse impact calculation on your own applicant flow - Why it is in the file: Section 1607.15(A)(2) contemplates adverse impact determinations at least annually for groups at 2 percent or more of the relevant labor force or applicable workforce.
• Record of alternatives considered and rejected - Why it is in the file: Answers the lower-impact alternative question in 1607.3(B).
• Administration protocol, rater training, accommodation log - Why it is in the file: Section 1607.5(E) expects standardized administration and scoring.

Employers with 100 or fewer employees get a lighter path. Section 1607.15(A)(1) allows simplified recordkeeping: counts of persons hired, promoted, and terminated for each job by sex and where appropriate by race and national origin, applicant counts on the same basis, and which procedures were used.

When to Re-Validate

Section 1607.5(K) is honest about this: there are no absolutes in determining whether a validity study is outdated, and the relevant factors are the validation strategy used, changes in the relevant labor market, and changes in the job. A two to three year review cycle is a reasonable working default rather than a figure drawn from the regulation.

Re-validate ahead of schedule when the job is redesigned, when new equipment or software changes the core tasks, when you merge or split roles, when you change the cutoff score or move from pass or fail to ranking, when the vendor updates the instrument or its scoring model, and whenever your annual impact numbers move. A tool quietly re-trained by a vendor is a new tool, which applies with particular force to AI skill assessments, where the model can change without the test looking different.

The same discipline applies across the selection process. If you run structured automated reference checks as a gate that can eliminate candidates, those questions are a selection procedure and belong in a file of their own. Pooling stages into one measurement hides which gate is doing the damage. Further employer guides sit in our HR and pre-employment resources.

This article is general information for US employers and is not legal advice. Confirm your obligations with employment counsel in the states where you hire.

Related Resources

Uniform Guidelines on Employee Selection Procedures (29 CFR Part 1607)

US Equal Employment Opportunity Commission

Ensuring Accuracy: Validity and Reliability in Skill Assessments

Avoiding Adverse Impact IQ Testing Legal Risks

skill tests library

AI skill assessments

HR and pre-employment resources

Frequently Asked Questions

Is a validation file legally required?

No statute names the file itself. The obligation is to show that a selection procedure producing adverse impact is job related, and 29 CFR 1607.5(D) and 1607.15 describe documentation an employer should maintain and have available.

Do the Uniform Guidelines still apply after the 2026 DOJ opinion?

They remain codified at 29 CFR Part 1607. The Office of Legal Counsel opinion of June 9, 2026 is executive-branch guidance and does not bind federal courts, amend Title VII, or remove disparate-impact claims brought by private plaintiffs or under state law.

Is the employer or the vendor responsible for validation?

The employer using the procedure has to produce the evidence. A vendor supplies the technical manual and the underlying study, and both belong in your file, but buying a tool does not move the obligation onto the seller.

How often should a skills assessment be re-validated?

Review every two to three years, and re-validate sooner on a trigger such as role redesign, a cutoff score change, a vendor model update, or a shift in your annual impact numbers. The guidelines set no fixed interval.

Does a small employer need one?

Employers with 100 or fewer employees may use the simplified recordkeeping route in 1607.15(A)(1), which is lighter but not an exemption. Title VII generally reaches employers with 15 or more employees, and a small employer whose test screens out a protected group still faces the job-relatedness question.

The Cheapest Insurance in Hiring

The validation file is the cheapest insurance in hiring, built from a job analysis, a task-to-item map, a vendor technical manual, and an annual impact calculation you can run on a spreadsheet. Start with the role where you reject the most candidates, because that is where your exposure concentrates. Give the file a named owner and a review date, and a legal risk becomes a maintenance task. If your vendor cannot hand you a full technical manual, put that first on the list.