The setup
The arithmetic of feature removal always looks safe in the spreadsheet: four percent of users, high maintenance cost, an obvious migration path. The spreadsheet is missing one column — which four percent. If the affected users include the people who answer questions in your forum and vouch for you on social media, you are not cutting a feature, you are cutting your most efficient marketing channel.
This scenario prices that risk before you pay it. The decision is a 90-day deprecation with a reasonable exit package. The audiences are the affected minority, the watching majority, prospective customers encountering the argument from outside, and a competitor community ready to welcome defectors. The trigger is the notice itself; the horizon is eight weeks, long enough to see whether the grievance decays or becomes lore.
Deprecations are ideal simulation material because the real damage is reputational contagion — unaffected users updating their beliefs about what you might do to them next. That updating happens through conversation, which is precisely the thing a multi-agent run models and a stakeholder spreadsheet does not.
The prompt
Copy this into MiroFish as your scenario question, swapping the specifics for your own:
We are removing our self-hosted export option in 90 days; affected users get a migration guide and six months of a discounted cloud tier. Roughly 4% of users rely on it, but they include several long-time community voices. Simulate eight weeks from the deprecation notice across: affected power users, the wider user community, prospective customers reading the discussion from outside, and one competitor's community that courts our defectors. Tell me whether the '4% problem' stays contained, how the story reads to prospects who never used the feature, and what the discount offer does to the tone.
Seed files that help
- The deprecation notice draft and migration guide — the perceived quality of the exit path drives most of the reaction.
- Forum or issue-tracker threads where the feature is discussed today, so the graph knows exactly who the attached users are and how central they sit.
- A note on which competitors offer the feature and at what price, giving defection threats a realistic destination.
What to look for in the report
- Containment: whether the reaction stays inside the affected 4% or jumps the fence when unaffected users adopt it as evidence of direction ('what gets cut next?').
- The prospect read: how the episode looks to simulated outsiders evaluating you — deprecations are read as character evidence, not product news.
- Community-voice amplification: what happens when the two or three high-centrality users in the graph weigh in, versus the counterfactual where they stay quiet.
- Whether the discount is read as generosity or as an admission the removal is about upselling — the report usually renders a clear verdict.