Scope
Identify system version, authorized tasks, tools, users, environment, excluded behavior, and the decision the evaluation informs.
METHOD / 01
Evaluation begins with a question, not a score.
This is a proposed workflow. Engagement-specific procedures and acceptance criteria require technical review before they support public outcome claims.
Design
Translate a credible failure mode into a scenario. Define input, expected behavior, fixed conditions, variants, and capture plan before execution.
Execute
Record prompts, tool calls, outputs, resulting state, and deviations. Repeat enough times to describe variability; never infer a stable pattern from one run.
Interpret
Compare observation with the stated criterion. Report reproduction status, uncertainty, scope limits, and candidate regression tests.
DECISION RULE
Do not turn a test result into a broader guarantee.
A finding applies to the stated system, configuration, environment, and procedure. Any wider conclusion needs additional evidence.
See the record format