Human feedback
Structured judgement, not scattered opinions.
Preference data, evaluations, rankings and structured human feedback designed around clear rubrics so results stay comparable.
What we collect
Signals, structured.
Preference pairs
Side-by-side comparisons with recorded rationale.
Rubric scoring
Multi-dimension scoring defined with your team before review begins.
Expert review
Domain reviewers for specialised or technical evaluation tasks.
Written critique
Structured free-text explanations attached to each judgement.
Potential applications
- Preference optimisation and reward modelling
- Model evaluation and regression testing
- Instruction-following improvement
- Domain-specific quality benchmarks
Collection workflow
- 01Rubric and task design agreed with your team.
- 02Reviewer calibration on shared examples.
- 03Review tasks completed through structured workflows.
- 04Agreement measured, outliers re-reviewed.
- 05Labels, scores and rationale delivered together.
Building an AI system that needs better data?
Tell us what your model needs to learn. We'll explore how a purpose-built human data program could support it.