HomeCompareHarver vs Juicebox (PeopleGPT)

Side-by-side · Rubric v1.0

Harver vs Juicebox (PeopleGPT)

Export comparison (CSV)Pro

§ The result

1 point separates Juicebox (PeopleGPT) and Harver.

Juicebox (PeopleGPT) (Sourcing) edges Harver (Assessment) by 1 point under rubric v1.0, 57 to 56. Juicebox (PeopleGPT) leads 3 of the 7 criteria, including Post-Market Monitoring, while Harver leads 1, including Article 11 Technical Documentation, and 3 are level. The heaviest weighted swing runs the other way — Harver takes Article 11 Technical Documentation at 20%, Harver 65 to Juicebox (PeopleGPT) 52, 100% of the weighted advantage Harver holds — while the widest raw gap sits on Post-Market Monitoring, Harver 40 to Juicebox (PeopleGPT) 55.

§ The record

Harver is filed under Assessment AI, Juicebox (PeopleGPT) under Candidate sourcing AI. They record no secondary category in common.

Both are in scope of NYC LL 144, Illinois HB 3773, EU AI Act and Colorado AI Act. Only Harver is in scope of Mobley v. Workday.

RecordHarverJuicebox (PeopleGPT)
Price tierEnterpriseMid
HeadquartersNLUS
Founded20152022

§ What decides it

The criteria that move the gap, heaviest swing first.

  1. 20% weight · 49% of the movement so far

    Article 11 Technical Documentation

    Harver 65 · Juicebox (PeopleGPT) 52

    2.6 weighted points to Harver

  2. 12% weight · 83% of the movement so far

    Post-Market Monitoring

    Harver 40 · Juicebox (PeopleGPT) 55

    1.8 weighted points to Juicebox (PeopleGPT)

  3. 18% weight · 97% of the movement so far

    Bias Audit Transparency

    Harver 80 · Juicebox (PeopleGPT) 84

    0.7 weighted points to Juicebox (PeopleGPT)

  4. 15% weight · 100% of the movement so far

    Data Governance Disclosure

    Harver 55 · Juicebox (PeopleGPT) 56

    0.1 weighted points to Juicebox (PeopleGPT)

Weighted delta is the raw difference times the criterion weight, so the 4 figures above sum to -0.1 — the difference between the two totals before either is rounded. Shares are of 5.3 points of movement counted without direction. The table below rounds each side to a whole number first, so its Δ raw total reads -1 beside that -0.1.

§ Category-by-category

Raw score per criterion, with weighted delta.

Δ is Harver minus Juicebox (PeopleGPT).

CategoryWeightHarverJuicebox (PeopleGPT)Δ rawΔ weighted
Article 11 Technical Documentation20%6552+13+2.6
Bias Audit Transparency18%8084-4-0.7
FRIA Support15%3535--
Data Governance Disclosure15%5556-1-0.1
Human Oversight Design12%5050--
Post-Market Monitoring12%4055-15-1.8
Customer Documentation8%6060--
Total100%56 (F)57 (F)-1-0.1

§ Evidence

Cited sources and documented absences, criterion by criterion.

An absence is a finding, not a blank: the reviewer looked for a named artefact and recorded that it is not located. Where a source sits is a fact about hosting only.

CriterionWeightHarverJuicebox (PeopleGPT)
CitedAbsencesAudit / hostingCitedAbsencesAudit / hosting
Article 11 Technical Documentation20%30Sources on the vendor's own domain11Sources on a separate domain
Bias Audit Transparency18%20Audit reportSources on the vendor's own domain30Audit reportSources on a separate domain
FRIA Support15%10Sources on the vendor's own domain11Sources on a separate domain
Data Governance Disclosure15%10Sources on the vendor's own domain20Sources on the vendor's own domain
Human Oversight Design12%10Sources on the vendor's own domain11Sources on a separate domain
Post-Market Monitoring12%1020Audit reportSources on a separate domain
Customer Documentation8%10Sources on the vendor's own domain21Sources on the vendor's own domain
All criteria100%100Audit reportSources on the vendor's own domain124Audit reportSources on the vendor's own domainSources on a separate domain

Harver has 10 evidence rows across 7 of 7 criteria — 10 with a source URL, 0 documented absences; Juicebox (PeopleGPT) has 16 evidence rows across 7 of 7 criteria — 12 with a source URL, 4 documented absences; some documentation or audit rows for Juicebox (PeopleGPT) sit on a domain separate from the vendor's own site.

Provenance

The scores above are documented by 26 evidence rows on file for these two vendors: 10 for Harver, across 1 distinct source domain, and 16 for Juicebox (PeopleGPT), across 4 distinct source domains.

Rubric
v1.0
Harver last reviewed
May 21, 2026 · HireAIScore editorial team
Juicebox (PeopleGPT) last reviewed
June 7, 2026 · HireAIScore editorial team

HireAIScore scores are editorial research, not legal advice, and no vendor pays for placement, scoring or removal — see the methodology, conflicts policy and terms.