Experiments
From a question to documented learning.
An experiment is a way to understand what can reasonably be learned about a practical use of AI. It starts with a clear question and ends with evidence, limitations and documentation that can be shared.
Method
Five stages. One traceable chain.
Each stage constrains the next. The result is not a polished demo but an understandable account of what happened.
- 01 · QUESTION
A practical question
Define one person, task or situation in which a practical use of AI needs to be understood better.
- 02 · SCOPE
Scope
Record the users, material, time, responsibilities, sign of success and what the experiment will not test.
- 03 · BUILD / TEST
Build and test
Make the smallest useful implementation. Keep people in control and record exceptions during the work.
- 04 · EVIDENCE
Evidence
Collect observations, failures, limitations and alternative explanations. Do not fill gaps with assumptions.
- 05 · DOCUMENT / SHARE
Document and share
Separate observation, interpretation and open questions. Publish only what the evidence supports.
Evidence boundary
Documentation also states what we do not know.
A future experiment note should show at least the question, scope, date, observation, limitation and publication responsibility.
We publish what we can support.
A single experiment describes a bounded result. We also document limitations and open questions, and do not generalise beyond the available evidence.