Written guidelines
We help turn your label taxonomy or rubric into clear instructions, agreed with your team before the pilot.
Quality System
Quality built into every step, not checked at the end
Labels and ratings are only as good as the guidelines and reviewers behind them. One6 combines written guidelines, calibrated contributors and layered review so that disagreement is caught before it reaches your training set.
Quality controls
Reviewer calibration
Reviewers are calibrated on shared examples against your rubric before production work begins.
Gold-standard items
Examples with known answers are used to measure contributor accuracy.
Reviewer sampling
Samples of contributor output are reviewed throughout the project.
Agreement monitoring
Agreement between annotators and reviewers is measured and monitored for the life of the project.
Contributor qualification
Contributors pass a skills test or trial task and are trained on the project's guidelines before production work.
Performance tracking
Each contributor's accuracy and output are tracked. Contributors below the quality bar are retrained or removed.
Expert staffing
Specialized material is handled by domain experts rather than general annotators.
Quality across the project
- 01
Agree
Quality checks and acceptance criteria are agreed upfront and written into the specification.
- 02
Calibrate
Contributors work through shared examples against your guidelines before production.
- 03
Prove
You see the quality results during the pilot, before committing to volume.
- 04
Sustain
Quality assurance and contributor performance tracking continue through every batch delivered at scale.
Frequently asked questions
How is quality measured?
Quality checks are agreed upfront and typically combine gold-standard examples, reviewer sampling and agreement between annotators. You see the results during the pilot.
How do you keep ratings consistent across reviewers?
Reviewers are calibrated on shared examples against your rubric before production work, and agreement is monitored throughout the project.
Can you audit or re-label an existing dataset?
Yes. Re-labeling or auditing an existing dataset with quality problems is one of the projects our annotation service supports.
More about the platform
- Project WorkflowWritten specification, pilot review, then delivery in regular batches with ongoing QA.
- Contributor NetworkDomain experts, native speakers, generalists and voice talent, matched to each project.
- Trust & SecurityData ownership, confidentiality and contributor consent, agreed before work begins.
Have a data project in mind?
Tell us what your model needs. We'll come back with questions, a proposed approach and a pilot plan.