Vetting
Five gates, then a probation on real work
Tier 1 is automated and never sold. Tier 2 is checked by people and is what clients hire. Tier 3 is earned on the job.
-
Sign-up screen
Residency self-certification, IP check, device and phone check.
-
Native-language test
Written in the language, timed, with register and dialect items and paste detection.
-
Skill test
Annotation, RLHF ranking or a domain quiz, scored automatically.
Tier 1 · never sold -
ID check and proctored retest
Document and liveness check, then a recorded retest on new items.
Checked by a person -
Recorded work sample
A 20-minute task with camera and screen recorded, then a 2-minute spoken explanation in the language, scored by a native reviewer.
Checked by a person Tier 2 · ready to hire
Probation: the first 10 billed hours on every new project are double-reviewed.
Tier 3: 200+ billed managed hours with a QA score of 92% or more.
What you see
The evidence stays on the profile
A named person checks identity, residency, language and skill. Each check keeps its provider, date and result, and every status change is logged.
Reviewers are native speakers paid per review. Until a language has three Tier 3 workers, contracted native linguists review the work samples on the same scorecard.
Evaluator · ID 2417
Kazakh · Russian · Astana
- Identity
- Sumsub · national ID card · 12 Oct 2026 Document number never shown
- Residency
- Kazakhstan · verified 28 Oct 2026 Proof of address · IP check at every login
- Languages
- Kazakh 91 · Russian 94 · both proctored Reviewer A. S. · 14 Oct 2026
- Skills
- RLHF ranking 86 · annotation 90 Label Studio, CVAT
- Track record
- Probation: 6 of 10 hours reviewed QA score shown after 30 days
Quality
One score, one set of consequences
QA score is a worker’s gold-set accuracy over the last 30 days on managed projects. Gold items are mixed into every queue.
| Trigger | What happens |
|---|---|
| Gold accuracy under 85% on 2 working days in a row | Paused, paid recalibration, then back in the queue |
| QA score still under 85% two weeks after recalibration | Down one tier and off the project |
| Fraud confirmed by a person reviewing the flag | Removed and added to the shared ban list |
| QA score of 92% or more and 200+ billed managed hours | Tier 3: eligible for lead and QA roles |
Fraud controls
- Proctored retest on new items, with webcam and screen recorded
- A recorded work sample with a spoken explanation
- Device fingerprinting and duplicate-document detection
- Gold items in every queue, every day
- A shared ban list, only after a person reviews the flag
Fairness
- One retake per test, after 30 days
- Appeals within 14 days go to a different reviewer
- Recordings are deleted 90 days after the decision unless held for a fraud case or appeal
Pilot
Start with a paid two-week pilot
10 vetted workers in your tool, ready within 10 business days (5 for Russian). If a managed team’s weekly gold-set agreement falls below 90%, the hours below it aren’t billed.