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 delivery.