RMP.Application:5.8 - Keep the threats connected to actual decisions
Use these distinctions while inspecting the supplied plan and its execution. Each threat changes what the comparison or its result can support.
| Threat | What it changes in this application |
|---|---|
| Initial differences between groups | Pairing and random assignment address the comparison; inspect actual workload and complexity. Selecting only the worst branch for treatment and comparing it with an easier branch needs a different justification. |
| Reaction to being observed | Common recording makes the comparison refer to monitored work. Withdrawing that practice changes the conditions for transferring the answer. |
| Several changes introduced together | Record the actual treatment bundle. Adding a new priority rule as well as an approver estimates that bundle unless another justified comparison separates them. |
| Time, learning or carryover | The concurrent branch comparison and the changed shared-queue case require different dependence models. A later result may depend on earlier training or a seasonal workload change. |
| A convenient indicator replaces the goal | Keep actual sending time distinct from a changed promise or CRM update. Restore the event-to-measure relation before using the number. |
| Judgement affected by the observer | Use event records and an applicable correction rule, with assignment hidden from the assessor when feasible. Unavoidable awareness limits the resulting claim. |
| Repeated records mistaken for independent units | Analyze the assigned branches and matched pairs. Several records from one request, approver or branch do not supply additional independent assignments. |
| Inadequate recording resolution or precision | Inspect the recording operation and the planned uncertainty. More rounded rows may add no distinction; more genuinely informative units may still be too costly to obtain. |
| Selecting outcomes, analyses or stopping after seeing results | Preserve what was fixed beforehand and why later choices changed. An exploratory favorable subgroup can suggest another question; it does not retrospectively become the planned confirmatory answer. |
Before extending the result, check the receiving workload, staffing and interactions that its use depends on. Re-running the same data can expose a computational error; it supplies no new workplace cases. A relevant new-data check earns its cost through RMP.6. The useful finish is the supported answer and the action it can inform, or a precise limitation with a justified next move.