Risk and requirements matrix
System risk tiers
| Tier | Description |
|---|---|
| Low-Risk Operations | Beta testing, internal pilots, and proofs of concept where failure impact is limited and human oversight is strong. A single-team internal prototype whose outputs do not inform downstream decisions is a representative example. |
| Medium-Risk Operations | Internal agency use with minimal operational impact, and low-impact public use with strong human oversight. |
| High-Risk Operations | Significant operational impact or high-impact public-facing deployment, including consequential information or external actions. |
Risk tier is the primary determinant of mandatory telemetry. Perform a separate data-handling assessment: a Low-Risk experiment processing restricted data may need stricter telemetry protection than a High-Risk service using public data.
Priority tiers
| Priority | Description |
|---|---|
| Tier 1 | Most foundational telemetry supporting routine production oversight. |
| Tier 2 | Evidence supporting incident investigation and accountability. |
| Tier 3 | Deeper performance, safety, experience, resilience, and governance analysis. |
| Tier 4 | Evaluation-intensive or aspirational telemetry requiring labelled or representative data. |
Obligation matrix
| Priority tier | High-Risk system | Medium-Risk system | Low-Risk system |
|---|---|---|---|
| Tier 1 | Mandatory | Mandatory | Good-to-Have |
| Tier 2 | Mandatory | Mandatory but Waivable | Good-to-Have |
| Tier 3 | Mandatory but Waivable | Good-to-Have | Good-to-Have |
| Tier 4 | Good-to-Have | Good-to-Have | Good-to-Have |
Actions
- Mandatory — must be implemented; no waiver is permitted.
- Mandatory but Waivable — must be implemented unless the authorised oversight function approves documented risk acceptance.
- Good-to-Have — strongly encouraged and selected according to system-specific risk and constraints.
“Not applicable” is not an ORF matrix action. This playbook uses it only as a governed record status when the subject of an item does not exist within the approved system scope; see Select requirements.
Worked matrix examples
High-Risk public assistant
Tier 1 and Tier 2 are mandatory. Tier 3 must be implemented unless validly waived. Tier 4 is selected where, for example, hallucination or fairness evaluation addresses a material risk.
Medium-Risk internal drafting assistant
Tier 1 is mandatory. Each Tier 2 item is implemented or individually/collectively waived through the approved process. Tier 3/4 items may still be valuable for human edit, drift, or quality concerns.
Low-Risk prototype
All items are Good-to-Have under the matrix, but Tier 1 provides a sensible foundation and makes later production progression easier. Data protection obligations still apply.
Changes that may affect the matrix
Reapply the matrix when the system moves risk tier. New capabilities may also make previously inapplicable items applicable—for example adding retrieval, tool use, memory, multi-service orchestration, or human handover.