Short answer. The rule requires the number of individuals excluded from required calculations because they fall within an unknown category to be reported. A defensible audit should preserve the difference between values that are genuinely unknown, declined, missing, invalid or outside the controlled taxonomy rather than inventing demographic classifications.
What the rule expressly requires
For both selection/classification and scoring workflows, 6 RCNY §5-301 requires the audit to indicate the number of individuals assessed by the AEDT who are not included in the required calculations because they fall within an unknown category.
Unknown is not the same as inferred
DCWP guidance states that imputed or inferred demographic data cannot be used to conduct the bias audit. Demographic characteristics should therefore not be guessed from names, photographs, geography, language, resumes or similar proxies.
Why data-state preservation matters
| Source state | Recommended audit treatment |
|---|---|
| Known category | Map only through an explicit, documented taxonomy. |
| Unknown | Preserve as unknown and report consistently with the rule. |
| Declined / not disclosed | Preserve the source state; do not invent a protected category. |
| Missing / blank | Treat as a data-quality state requiring review, not automatically as a known demographic category. |
| Unrecognized source value | Resolve or document the mapping gap before final calculations. |
What can go wrong
- Collapsing blanks and declined responses into a protected category.
- Guessing race/ethnicity from names or geography.
- Dropping unknown records without reporting the required count.
- Changing mappings between audit runs without version control.
Operational implication
A dataset can contain all required columns and still be unready if the demographic values are ambiguous or inconsistently encoded. That is a data-readiness problem, not merely a formatting problem.
Last legally reviewed: September 27, 2026.