Hypothesis
An independent critic with actual model evidence can identify issues missed by the modelling role.
LAB 004
Can another agent find mistakes in the model?
An independent critic with actual model evidence can identify issues missed by the modelling role.
Planned
Design or workflow problem
An independent critic with actual model evidence can identify issues missed by the modelling role.
Setup, inputs and constraints
Seed a model with an undersized opening, missing material code and unsupported object. Keep a reference answer set separate from the critic.
Agent and tool architecture
Approved rules + model extraction → deterministic checks → critic explanation → reviewer decision.
Human approval points
A person approves inputs, proposed design choices and release of the candidate. No autonomous production edits or machine operation.
Result
Not run. This record defines the experiment; no performance result is claimed.
What failed / failure tests
Test false passes, false alarms and a fluent explanation contradicting the measured result.
What changed
Baseline protocol. Revisions will be recorded alongside the run that motivated them.
What was learned
No findings recorded. The criteria below define what the experiment must establish.
Acceptance criteria
Report detection, false alarms and not-tested coverage with object IDs and rule versions.