Skip to content

Research index

Enterprise AI research methods and reports

How we evaluate enterprise search and cited answers, published as methods anyone can run on their own documents, with no vendor ranking attached.

Updated 21 Aug 2026

01

Enterprise AI search evaluation framework

A method for comparing retrieval, citations, no-answer behaviour, permissions and operating fit on a representative document set.

  • Document set design
  • Question set
  • Separate scoring dimensions
  • Reproducibility record
02

Regulated Document AI Evaluation Kit

  • Evaluator worksheet
  • Procurement scorecard
  • Known, synthesis and conflict cases
  • Stale, absent and permission cases
03

Benchmark publication boundary

We will not publish a benchmark result until anyone can see the document set and our rights to it, the reference answers, the configuration, how much the results varied, who scored them, the runs that failed and what the test does not cover.

  • Document set rights and reference answers
  • Configuration and variability
  • Evaluators and failed runs
  • Limitations public for scrutiny

What this page does not prove

  1. B1We do not rank vendors on this index.
  2. B2Publishing a method is not the same as publishing a result.
  3. B3Results apply only to the disclosed test configuration.

Related pages

Plan an evaluation