Work type
RLHF and preference ranking
Native speakers compare model answers, pick the better one and explain why, in the language your users write in.
Get a pilotTypical tasks
- Side-by-side ranking with written rationales
- Rubric scores for helpfulness, accuracy and tone
- Flags on unsafe or culturally wrong answers
Who does it
Tier 2 native evaluators; a Tier 3 lead calibrates the rubric with you.
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.