How should I scope and license a schedule dataset or environment?

Start from the agent you are building, not from the volume of data you think you need. Four decisions shape a useful scope.

**Domain and process.** Pick one setting, such as loan servicing, utility payment plans or delivery calendars, and name the adjustments that matter: deferral, re-plan, consolidation, cancellation. A narrow first scope is easier to inspect.

**Form.** Decide whether you need a static dataset, a task and evaluation pack, or a simulated environment with tool actions and reset. Training usually favours datasets and varied traces. Evaluation favours tasks with fixed starting state and checks.

**Use.** State plainly whether you intend to train, fine-tune, evaluate, or only demonstrate. Licence terms can differ by use, and a team should not assume that rights for one cover another.

**Terms.** Ask for the licence in writing, including who may use the material, whether outputs from models trained on it are restricted, and what happens when the agreement ends. MirrorFirm will set these in scoping; do not rely on assumptions.

Also agree how you will inspect the material: your reviewers, your checks and your criteria for accepting it. Acceptance criteria written before delivery prevent disputes about quality afterwards.