# Ground truth · Glossary
The verified, correct answer for an example, used to train a model or judge its outputs.
[Glossary](/glossary) · Machine learning and MLOps

# Ground truth

     The verified, correct answer for an example, used to train a model or judge its outputs. Ground truth might be a confirmed outcome, such as whether a transaction was fraudulent, or an expert’s agreed judgement.

## Why it matters

     Every accuracy figure depends on the ground truth behind it. If the ‘correct’ answers are themselves inconsistent, delayed or incomplete, a model can look better or worse than it really is.

     Establishing ground truth often requires process changes, such as recording outcomes that were never captured, or having experts label a sample consistently. That investment is frequently the difference between a demonstration and a dependable system.

## In practice

     For example, a UK retailer measuring a returns-fraud model might find that only a fraction of suspected cases are ever investigated, so it sets up a small, randomly sampled review to establish ground truth for cases the model did not flag.

## Where Rodan fits

     Rodan works with domain teams to define and capture ground truth as part of [Applied AI Engineering](https://rodan.io/applied-ai-engineering) delivery.

## Related terms

- [Golden dataset](/glossary/golden-dataset)

- [Data labelling](/glossary/data-labelling)

- [Training data](/glossary/training-data)

- [AI evaluation (evals)](/glossary#ai-evaluation)

- [Supervised learning](/glossary/supervised-learning)
HTML: https://rodan.io/glossary/ground-truth