Stratum

Definition

What is preference data

Preference data is pairwise or ranked human judgment about which model output should be preferred. For expert programs, the rater should be able to justify the preference in domain terms.

What is preference data?

Preference data is pairwise or ranked human judgment about which model output should be preferred. For expert programs, the rater should be able to justify the preference in domain terms.

Why it matters

Rankings from people who cannot see domain errors become style votes.

Example

Two legal answers compared by a practice-area attorney who records why one reasoning is valid.

When it is used

Post-training or selection needs a ranked signal rather than a single gold answer.

Common mistakes

  • Hiding model identity when you actually need source-aware safety review—or the reverse
  • No justification field
  • Using preferences as a substitute for a gold benchmark

Sources

Citations point to primary technical or institutional documents. They support definitions, not a claim that those organizations are customers.

What kind of experts do you need?

Share the profession, specialty, experience, location, headcount, hours, duration, and project description. We assemble the expert capacity.

  • Need 25 licensed nurses for a clinical AI evaluation
  • Need 15 attorneys in a specific practice area
  • Need 20 PhD scientists for benchmark creation
  • Need 30 senior software engineers for code evaluation