Headquarters: Rotterdam, Netherlands and Atlanta, United States
Assessed: High-volume pre-employment assessment platform, including the pymetrics soft-skills games.
Rubric: version 1. Search protocol applied and all sources accessed: 2026-09-22.
Documented 2, partial 3, absent 3, of eight criteria. This records what is published about the validity and fairness of the scores. It is not an assessment of the product, the model or the underlying science.
| # | Criterion | Level |
|---|---|---|
| C1 | Technical manual or validity report publicly available without a sales conversation | P |
| C2 | Validity coefficients reported with sample size and uncertainty | A |
| C3 | Adverse impact reported by named subgroup | D |
| C4 | Scored construct named and defined, not described only by outcome | P |
| C5 | Evidence of comparability across languages, versions or delivery modes | A |
| C6 | Human oversight documented, with a candidate route to contest a score | P |
| C7 | Independent review named and the reviewer identified | D |
| C8 | Reliability reported, and reported by group | A |
C1. Technical manual or validity report publicly available without a sales conversation
Partial.
The published bias audit is a complete, dated, downloadable document: "SUMMARY OF BIAS AUDIT RESULTS", "V1.0: 11/04/2025", "PREPARED BY BABL AI INC."
Source: https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
The published document is a bias audit. No validity or reliability manual was found.
C2. Validity coefficients reported with sample size and uncertainty
Absent.
No public source was found for this criterion by the search protocol.
Pages examined include https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
C3. Adverse impact reported by named subgroup
Documented.
Full selection rates and impact ratios with per-cell N, by gender ("Male, 95, 0.926, 1.000"; "Female, 187, 0.850, 0.918"), by race/ethnicity ("Black or African American, 104, 0.846, 0.846"; "Two or more races, 31, 0.839, 0.839") and for intersectional groups ("Non-Hispanic or Latino, Female, Black or African American, 85, 0.824, 0.824").
Source: https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
The document reports the number of applicants, the selection rate and the impact ratio for each group, the intersectional permutations, the threshold applied, the demographic categories that were not quantified, and the counts excluded from the calculation.
C4. Scored construct named and defined, not described only by outcome
Partial.
"evaluating personality, cognitive abilities, and job-specific skills to generate a Matching Score"
Source: https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
Three construct domains are named and the output is defined operationally, including the percentile threshold. Individual constructs are not defined publicly.
C5. Evidence of comparability across languages, versions or delivery modes
Absent.
No public source was found for this criterion by the search protocol.
Pages examined include https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
C6. Human oversight documented, with a candidate route to contest a score
Partial.
An accountable party is named and published: "Accountable party: AI Algorithmic Risk Committee (AIARC)", with a named chair and role.
Source: https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
Internal governance is documented and an accountable party is named. No candidate-facing route to contest a score was found.
C7. Independent review named and the reviewer identified
Documented.
"We have independently audited the bias testing assertions and related documentary evidence of Harver (the ‘Company’) as of 11/04/2025". Auditor: "BABL AI Inc., 630 Fairchild Street, Iowa City, IA 52245". Opinion: "Disparate impact quantification PASS; Governance PASS; Risk assessment PASS; Overall PASS".
Source: https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
A second audit, of the pymetrics Soft Skills Platform by the same auditor, is published alongside it.
C8. Reliability reported, and reported by group
Absent.
No public source was found for this criterion by the search protocol.
Pages examined include https://harver.com/wp-content/uploads/2025/11/Harver-Platform-2025-Bias-Audit.pdf (accessed 2026-09-22).
Absent records the outcome of the published search protocol on 2026-09-22. Where a source is supplied that the search did not reach, the cell is re-scored under the corrections policy, with the date of the change recorded and the previous level retained.
Back to the index, the rubric and the methods note