Set the cut at the minimum proficiency the job requires, using a named method. Then check it against job performance, run the four-fifths check, and write down how you chose it.
The short answer: minimum acceptable proficiency, named method
The Uniform Guidelines say cutoff scores should normally be set so as to be reasonable and consistent with normal expectations of acceptable proficiency within the work force (29 CFR 1607.5H). OPM ties passing scores to the minimum proficiency the job's competencies require, set with job experts and assessment experts.
There is no universal good pass mark.
Three ways to pick the number
Modified Angoff. Subject matter experts estimate, item by item, the probability that a minimally competent candidate answers correctly. Sum the estimates, average them, and you have a cut score. A rule of thumb is 10 to 20 experts.
Contrasting groups. Judges sort real test takers into qualified and unqualified. The cut sits where 50% are qualified, higher (67% or 75%) if passing an unqualified person is two or three times worse than failing a qualified one. ETS notes this has the strongest theoretical rationale and is the only method that estimates both error types.
Norm-based cuts. Set the cut against the distribution of scores. This is fast and weak. SIOP warns that basing cutoffs on incumbents' performance can set them too high when the workforce is more capable than the applicant pool.
SIOP states that there is no single method for establishing cutoff scores.
Check it against job performance
A cut score only works if it separates people who can do the job from people who cannot. Link the score to a meaningful performance threshold, then compare scores with later performance data.
Run the four-fifths check before you lock it
Under 29 CFR 1607.4D, a group selection rate under four-fifths (80%) of the highest group's rate is generally regarded as evidence of adverse impact. Smaller differences can still count if statistically and practically significant, and larger ones may not when the numbers are small.
CodeSignal advises against moving the cut score midway through a hiring process to chase pass rates. Group differences grow as the cut rises.
Consider bands instead of a single hard line
Bands group a range of scores so candidates are treated alike. SIOP notes bands can be set ad hoc or by the standard error of measurement, and that banding generally yields lower expected performance and utility than top-down selection.
Candidates experience the line directly. Can you still get hired if you fail an assessment test explains what a fail means for them, and is it normal to fail a job assessment shows how common a miss actually is.
Write down the rationale
Where a procedure with a cutoff has adverse impact, 29 CFR 1607.15 marks as essential a description of how normal expectations of proficiency were determined and how the cutoff was set.
Record the method, the experts involved, the data used, and the date. Tools such as AI Candidate Evaluation apply the same scoring criteria to every candidate. If someone fails, what happens if you fail a pre-employment assessment sets out the candidate experience.
Then review the cut once a year against performance data.