Executing and Auditing the Capstone Plan
Explore how to transform ML security research questions into structured capstone plans. Learn to define system boundaries, assets, threat models, metrics, and baselines. Understand how to self-audit for contradictions and prepare evidence-based evaluations. This lesson helps you develop repeatable workflows to assess ML security claims and design robust evaluations.
We'll cover the following...
Now turn the research question from the previous lesson into a capstone artifact, a single structured document that another reviewer can examine line by line. The order matters because each section informs the next, and skipping ahead can lead to mismatched metrics or missing baselines.
Start with system and assets, because assets define what security means for this system:
System sketch that names entry points, outputs, and lifecycle stages such as training, deployment, monitoring, and retraining.
Assets such as model weights, training data, user data, API availability, and decision integrity.
Then write the threat model in fields rather than ...