Evaluation, July 2026
We tested ClearRegs Search against 493 questions. Here is what we found.
A compliance register is only as good as the answers filling it, so we measure the
search behind it. We ran a structured evaluation of ClearRegs Search across all indexed Approved
Documents: 493 questions, each answer checked against the relevant source material. We publish the counts and
the method so you can judge the scope of the evaluation for yourself.
Each answer was checked against two separate criteria: whether it included the expected
technical terms, and whether it cited the expected Approved Document. 381 answers met both.
None met neither. The remaining 112 met one of the two.
This evaluation measures expected term coverage and citation targeting. It does
not assess whether an answer is technically or legally correct, whether the cited passage supports every
claim, or whether guidance has been applied appropriately to a particular project. It is run in house
rather than independently, and re-run when the indexed documents are materially updated. Reading the 112
back by hand against the source documents showed that a share of them were answers we would accept, marked
down because they used different wording from our expected term list, so the counts understate rather than
flatter.