Methodology
A result is only useful if its conditions can travel with it.
A passing demo is not a reliable workflow. EvalMechanics will publish the workload, inputs, evaluation method, raw outcomes, failure surface, correction burden, economics, human gate, and retest condition beside every controlled result.
Evidence classes
TESTED means EvalMechanics ran the stated workload. RESEARCHED means primary documentation, pricing, or public source material was checked. UNVERIFIED means a claim has not been independently confirmed. RETEST DUE means material change has made an earlier test potentially stale.
What a test must include
- What was tested and why it matters.
- Workload, inputs, version, and setup.
- Results and raw evidence where lawful to publish.
- Failure modes, correction steps, and correction time.
- Who should and should not rely on the result.
- Limitations and a retest trigger.
We do not turn vendor claims, polished demos, or one-off anecdotes into test results.