When to Use the Framework
Four common communication needs
Nobody has to work through the Framework front to back. These four common situations are what the Framework was built for. The Framework does not define a process; most organizations already have review models and escalation paths of their own.
What the Framework adds is that everyone involved names the same things the same way, whether they are proposing a use, documenting a tool, writing guidance, or recording what was done. Chapter 4 of the Framework works through each at length.
4.1
In approval flows
Relevant categories Use Case3.2Tool3.3AI Provenance3.4Operational Characteristics3.5Tool Capability3.6
Someone proposes an AI-enabled use and someone else has to decide about it. The Framework gives both sides the same words for the facts that decision rests on.
Two propose-review patterns come up often. In the first, a technologist finds a capability worth testing but the vendor or the tool is not yet approved, so the request is narrow: document the Tool and the Tool Capability, then add only as much Use Case as the reviewer needs to judge whether a test is reasonable.
In the second, the proposal is comprehensive from the start. Review begins with the Use Case, the benefit sought, the visibility of the project, the release and legal status of the content involved, how the user intends to interact with the AI, and then weighs that against the Tool, its AI Provenance, its Operational Characteristics and the Tool Capability in scope.
The Framework is applicable during the follow-up questions as much as the first submission. A reviewer can ask for more on Training Data, or for precision about which capability is actually proposed, and both the ask and the answer can use the Framework vocabulary, bringing in the precision and thereby shortening the round trips.
4.2
For tool documentation
Relevant categories Tool3.3AI Provenance3.4Operational Characteristics3.5Tool Capability3.6
Different customers ask vendors the same questions. Documenting the answers once, using the Framework vocabulary, turns that into a reusable asset that ships with the product.
Four categories that are defined are useful for this scenario: Tool, AI Provenance, Operational Characteristics and Tool Capability. Between them they say who provides the tool, how it is packaged, how the AI is enabled, what kinds of training data are involved, where it runs, what is collected and kept, whether customers can fine-tune it, and what each capability actually does to media.
Where separate capabilities rely on different models, different deployment conditions, or different controls, those are documented per capability. Otherwise, they could be documented once. The same Framework vocabulary applies regardless of the granularity at which the product is documented.
A vendor package prepared this way can drop straight into a studio’s approval request. The proposer just needs to add the use case and potentially little else.
4.3
For use guidelines
Relevant categories Use Case3.2Tool3.3AI Provenance3.4Operational Characteristics3.5Tool Capability3.6
Guidance that says “no generative AI” usually fails at the first hard case. The Framework gives guidelines enough precision to survive contact with a real workflow.
For documenting guidelines, start with the Use Case category to set the boundary: what may enter as an input, what the output is for, whether it can reach final media. Then draw on selected fields from the other four to say what kind of solution is acceptable inside that boundary, e.g., “a restoration capability that falls under Simple Transformation, deployed in a customer-controlled environment” is a sentence the Framework allows to be precisely expressed and a production user to interpret and act on.
The same structure supports escalation. Rather than a vague warning, a guideline can name the combination of inputs, outputs, final use potential and capability class that requires review.
4.4
For production records
Relevant categories Use Case3.2Tool Capability3.6Tool3.3AI Provenance3.4Operational Characteristics3.5
Guidelines describe what is permitted and Production documentation records what was done.
A production team or service vendor can use the Use Case category for the context, list the inputs and outputs of the actual use, reference the vendor’s existing Tool documentation, and name the specific Tool Capabilities in scope. Where deployment matters, e.g., a particular host environment or a particular data residency arrangement, AI Provenance and Operational Characteristics can be captured for that instance rather than for the tool in the abstract.
This precision in recorded detail is what demonstrates that the work stayed inside the approved boundary. It also lets different versions of an asset be distinguished: one made with a restoration capability and another with something more generative that needed different review.
The Framework stops at the governance level. It does not carry vocabulary to record asset-level provenance, per-asset linkage, event history, or contributor identities. The AI Interaction and Workflow Provenance Tracking fields in Approval Context, and the Provenance field in Outputs, are where a fine-grained technical provenance scheme could extend from.