Relevance Sample Size and Cost — DEJAN Codebase Explorer

Relevance probing accepts 1 to 100 samples per item-model pair, scaling API calls linearly while metered through usage-based credit billing.

Proactive Suggestion: Higher sample counts on stochastic models significantly reduce score variance, but sampling above 10 yields diminishing statistical returns for standard binary recommendations.

Referenced by

Relevance Probes
Q2B vs Relevance Probes
Calculating Query Relevance Scores
Relevance Probe Sample Methodology
Relevance Probing Mechanism
Relevance Sampling Configuration
Active Model Recommendations
Relevance Score Calculation Formula
Relevance Probe Sampling Configuration
Relevance Model Sampling