What should an evaluation team specify when requesting a scheduling dataset or task pack?

Before generation begins, settle these parameters so the pack matches what you want to measure:

- **Domain and obligation type.** Loan instalments, utility plans, delivery tranches, or patient account plans each behave differently. - **Schedule actions in scope.** Creation, skip, re-date, split, merge, early payoff, cancellation, or a subset. - **Policy set.** Which rules, thresholds and approval paths apply, and whether they change over the as-at period. - **Exception mix.** The share of tasks that should end in refusal or escalation, and which kinds of conflict to include. - **Time and calendar.** The jurisdiction, business-day rules, holidays and cut-off times. - **Starting depth.** Fresh cases, or mid-case tasks that begin with notes and prior versions. - **Ground truth form.** Expected final state, acceptable alternatives, and policy citations for each task. - **Holdout design.** Which tasks are kept apart for later testing, and how variants are grouped to prevent leakage.

The release is generated, so each parameter shapes how the output behaves rather than selecting from existing records. Ask for the parameters to be recorded in a manifest that travels with the data, so that anyone reading the results can see what the pack was designed to test.