A strong scenario has five parts:
Event + audience profiles + platform choice + run boundary + review questions
| Part | What to define |
|---|---|
| Event | What is announced or changes, when and with which public message. |
| Audience profiles | Which stakeholder behaviours and interests matter. |
| Platform choice | Twitter, Reddit or both, with a reason for each. |
| Run boundary | Configuration, rounds, seed and stop condition. |
| Review questions | What patterns the operator will inspect after the run. |
Platform choice
Choose both environments when the decision depends on the difference between fast public posting and community-thread dynamics. Choose one environment when the real question is tightly scoped to that interaction model.
Do not include unsupported platforms in the configuration or describe them as simulated outputs. If the real campaign also uses another channel, record it as an external limitation and analyse it separately.
Profiles before volume
The reviewed source does not establish an Academy-approved default for agent count or simulated calendar days. Do not copy fixed ranges from an old course version. Use the current product configuration, document the actual profile count and rounds, and compare runs under the same settings.
The profile set should represent the stakeholder behaviours relevant to the decision. It should not be presented as a statistically representative sample of the real population.
Review questions
Good review questions point to observable output:
- Which profile types initiated the highest-reach action chains?
- Which claims were repeated across both environments?
- Which discussion stayed platform-specific?
- Which selected agents explain their actions consistently in interview?
- Which result changes when the seed or profile mix changes?
"Will the launch succeed?" is not a reviewable simulation question. "Which message becomes the main objection in repeated runs, and where does it start?" is reviewable.