Purpose: Run through this checklist before deploying any AI system to ensure you've built the permission structures that prevent responsibility vacuums.

How to use: Duplicate this page for each AI deployment. Check items as you complete them. Expand sections for guidance on completing each item.

✅ Calibration Checkpoints Built In

□ Define Key Terms Explicitly

Before running AI analysis, we've defined what these terms mean for our specific deployment:

• [ ] "Clear data rights" defined (code ownership, training data, model weights, deployment rights, retraining permissions)
• [ ] "Acceptable performance" thresholds defined (accuracy %, variance limits, confidence score ranges, drift tolerance)
• [ ] "Low risk" clarified across functions (regulatory, reputational, operational, financial)

Who participated: [Add names/roles]

When completed: [Date]

Documentation location: [Link to meeting notes or decision doc]

<details> <summary><b>💡 How to complete this</b></summary>

Schedule a 30-60 minute session with Legal, Product, Compliance, IT, and Data Science.

Go through the key terms one by one. Ask each function: "When you say [term], what specifically do you mean?"

Document divergences explicitly. Don't force consensus—design for variance.

Template questions: Use the Calibration Checkpoints Guide for full question sets.

</details>

□ Reconcile AI Analysis Interpretations