Business Brain Assurance
How we test Business Brain
These measurements come from held-out evaluation cases and purpose-built test harnesses. They are evidence that the product has been seriously tested — not a promise that every customer question will match the same score.
41 / 42
Held-out cases passed
100%
Required evidence found in top five
23
Purpose-built test harnesses
0.985
Mean Reciprocal Rank
What these numbers mean
- Held-out cases passed — questions and scenarios reserved for evaluation, not used to tune the product toward those exact answers.
- Required evidence found in top five — when an answer depends on specific supporting material, that material appeared among the top five retrieved evidence items in the reported run.
- Purpose-built test harnesses — automated suites that exercise retrieval, abstention, continuity and related behaviours under controlled conditions.
- Mean Reciprocal Rank (MRR) — a ranking quality score between 0 and 1 for how early the first relevant evidence item appears. It is not a percentage of customer questions answered correctly, and must not be read as "98% accurate".
Limitations
Synthetic and held-out evaluations are not the same as proven results on every customer's live corpus. Results can change as models, corpora and product behaviour evolve. When available evidence does not establish an answer, Business Brain is designed to say so rather than invent one — that behaviour is part of what we test.
We do not claim that Business Brain is infallible, hallucination-free, or always correct.