Skip to content

Research

Enterprise AI research and evaluation methods

Open protocols and blank worksheets for testing enterprise document AI on your own material, plus an atlas of where data goes in each deployment.

01

Evaluation protocol

The Regulated Document AI Evaluation Kit publishes document set design, test families, scoring guidance, permission checks and blank worksheets.

  • Document set and question design
  • Citation and no-answer scoring
  • Permission and access checks
  • Blank evaluator worksheets
02

Architecture atlas

The data-flow atlas compares managed, customer-cloud, self-hosted or on-premise and local-model patterns without treating labels as evidence.

  • Managed cloud pattern
  • Customer-cloud pattern
  • Self-hosted / on-premise
  • Local-model pattern
03

Benchmark boundary

We do not make up benchmark data. Without a document set and a scoring record someone else can reproduce, there is no result to publish.

  • Methods published before results
  • Reproducible document set required
  • Scoring record required
  • No invented benchmark data

What this page does not prove

  1. B1Research methods are not measured product results.
  2. B2We claim no external benchmark without reproducible evidence.
  3. B3Licensing and confidentiality constrain which corpora can be published.

Related pages

Plan an evaluation