How we work

The experiments are the evidence.

Our method decides what that evidence can support. We record the exact question, the evaluator, and the conditions, we keep the original output, and we keep what happened separate from what we think it means.

The instruments are standardized and still being validated. That means we report what we observed. It does not mean we certify it.

How a run is conducted

Six steps, the same way every time.

1. 01

   Start with a commercial question

   The run has to connect to a real decision, a company being ruled out, a requirement, or an evidence problem.

2. 02

   Define the conditions

   We record the exact wording, the evaluator, its version, access mode, date, and sampling conditions.

3. 03

   Run it fresh, more than once

   Repeated runs reduce the chance that one favourable or unfavourable answer gets mistaken for a pattern.

4. 04

   Keep the original output

   The raw answer stays available for inspection. A summary is never treated as the source.

5. 05

   Record only what occurred

   Inclusion, recommendation, exclusion, evidence use, and next-step behaviour are recorded separately.

6. 06

   State the limitation

   A system's stated reason is evidence of what it said, not proof of how it actually works.