Label data, train AI, earn PRIV
Complete annotation tasks for businesses. Better work earns exponentially more through the 3D incentive system. Elite annotators earn up to 4.8x base rewards.
Earning potential
Base payout set by bounty poster + ecosystem bonus scaled by 3D multiplier
Task types
Image Classification
Categorize images into predefined classes. Is this a cat or a dog? Solar panel or skylight? The simplest task type.
Bounding Boxes
Draw rectangles around objects in images. Mark every car in a street photo, every defect on a solar panel.
Segmentation
Pixel-level annotation. Outline exact shapes of objects. The most detailed and highest-paying task type.
RLHF
Reinforcement Learning from Human Feedback. Rate AI outputs, compare responses, flag errors. Trains large language models.
Transcription
Convert audio to text. Transcribe voice recordings, label speaker turns, annotate sentiment.
How 3D incentives work
Every approved submission earns a base bounty payout from the company that posted it, plus an ecosystem bonus from the PRIV rewards pool. The bonus is multiplied by three dimensions.
Coverage (1.0x - 2.5x)
The platform tracks supply vs. demand for each data category. Working on underrepresented categories earns a higher multiplier.
Freshness (0.1x - 1.2x)
Annotating data that hasn't been touched recently earns more. Re-annotating data that was just labeled earns less.
Quality (0.3x - 1.6x)
Four metrics weighted into a quality score, then scaled superlinearly (score^1.5) and multiplied by your reputation tier bonus.
Reputation tiers
| Tier | Requirement | Max Disputes | Quality Bonus |
|---|---|---|---|
| Novice | 0-10 submissions | Any | 1.0x |
| Verified | 11+ submissions | <10% dispute rate | 1.1x |
| Expert | 51+ submissions | <5% dispute rate | 1.25x |
| Master | 201+ submissions | <2% dispute rate | 1.4x |
| Elite | 501+ submissions | <1% dispute rate | 1.6x |
Reward scenarios
Common task, average quality
Oversupplied category, standard data, decent but not exceptional work
Rare task, high quality Expert
Underrepresented data type + Expert tier = significant earnings boost
Urgent pipeline, Elite annotator
Critical deficit + active pipeline deadline + Elite reputation = maximum multiplier
Oversupplied, just re-annotated, Novice
Well-supplied category + just annotated + poor quality = minimal reward
Rules and nuances
Accuracy is everything
Your quality score uses superlinear scaling (score^1.5). This means 95% accuracy earns disproportionately more than 80%. The difference between good and great is exponential.
Gold standard checks are hidden
Some tasks in your stream are pre-labeled test items. Your performance on these directly affects your quality multiplier. You won't know which ones are gold standards.
Consensus matters
When multiple annotators label the same item, your labels are compared. Consistent agreement with other annotators raises your consensus score (40% of quality).
Disputes reduce your multiplier
Failed submissions (rejected by approvers) count as disputes. Keep your dispute rate below thresholds to maintain your reputation tier and quality bonus.
Payout delay for new annotators
New annotators have a 30-day payout delay (held in escrow). This decreases as you build reputation: 1 day off per successful submission, up to 15 days off. Staking PRIV removes up to 15 more days.
Skill badges gate specialized tasks
Some high-value tasks (medical imaging, solar analysis) require domain-specific PrivBadges. Complete qualification tests to earn badges and access premium bounties.
Ready to annotate?
Start with classification tasks to build reputation, then unlock higher-paying specialized bounties.
Join Waitlist