Evaluation foundations
Golden set
A curated set of representative inputs and trusted expected outcomes.
Also known as: gold dataset, gold test set
Definition
A golden set contains examples selected and reviewed by subject-matter experts, including normal, difficult, boundary and high-risk cases.
Why it matters
It translates a generic model comparison into evidence about the work your organisation actually performs.
Related concepts
- Test set
Data reserved for measuring performance after development.
- Held-out set
A private or untouched test set protected from optimisation.
- Human baseline
Performance achieved by an appropriate group of people on the same task.
- Critical-error rate
The share of cases containing a failure with severe consequences.