War-game a PR crisis before it happens
Crisis plans are usually tested in a meeting room, against people who already know the script. MiroShark runs the incident against a crowd that does not: customers, reporters, regulators, and competitors, each grounded in your scenario and each reacting in their own voice on a simulated Twitter and Reddit.
Then you can test the response. Write the statement, the apology, or the disclosure delay you are weighing into the scenario, and the crowd reacts to it as part of the story. Run a second version with a different response and compare the two reports. Each one cites the posts and trades that show where trust broke and whether it came back.
Who it is for
- Comms and crisis teams building or refreshing a playbook.
- Security and legal teams planning how and when to disclose an incident.
- Executives rehearsing a statement before they have to give one.
- Trainers running tabletop exercises who want a crowd to play against.
What you put in
One seed: a prompt, an article URL, or pasted text. Here is the example this page is built on. It opens prefilled in the run panel, ready to edit.
Our company just suffered a major data breach affecting 2 million users, and disclosure is delayed by 48 hours. Simulate how customers, regulators, press, and competitors react - and how a late CEO apology lands.
Will trust recover to baseline within the simulated week?
The cast it builds
MiroShark pulls the people, groups, and claims in your seed into a knowledge graph and casts grounded personas from it, each with its own beliefs, incentives, and posting style. For this seed, expect:
- Affected customers and users
- Regulators and privacy advocates
- Journalists covering the story
- Competitors and industry analysts
What you get back
Each run puts about 20 grounded AI agents (most runs use 10 to 40) through 10 rounds of a Twitter-like feed, a Reddit-like forum, and a Polymarket-style prediction market, then hands you:
- A cited markdown report written by a report agent that reads the simulated feed and market first, so every claim points back to a real post or trade from the run.
- Belief drift: how each group's stance moved round by round as the story developed.
- The top posts that carried the conversation on the simulated Twitter and Reddit.
- The prediction market trajectory: where the crowd priced the central question, and when it moved.
- A knowledge-graph view of the people, organizations, and claims MiroShark pulled from your seed.
- A public share page with a recap card that unfurls on X, Discord, Slack, and iMessage.
What to read for this job
- The round where trust drops fastest, and the post that triggered it.
- How each group reads the response you wrote into the scenario.
- Whether the story fades or escalates toward regulators and press.
- Where the prediction market sits on the recovery question by the end.
How to run it
- Open the run panel. Start from this example or open /sim with a blank prompt.
- Write the seed. A few tips for this job:
- Describe the incident, how many people it affects, and how long disclosure takes.
- Include the response you are considering, such as a CEO apology and its timing.
- Name the groups that matter most: customers, regulators, press, competitors.
- Pin the market if you have a question. MiroShark writes a sharp, resolvable market question for you by default. Pin your own to track the exact outcome you care about.
- Pay and follow the run. A flat $1 USDC per run on Base, Solana, or Monad, with no subscription and no gas to hold. The report lands in about 10 minutes.
- Or run it from code. Agents and scripts can start the same run through the paid x402 API with no account or API key. See the x402 guide and the docs for how the pipeline works.
Questions
Write it into the seed. Describe the incident and the response you plan, for example a CEO apology after a 48-hour disclosure delay, and the agents react to both as the story unfolds. The report shows how each group read it, citing the posts and trades behind each claim.
Yes. Run each version as its own seed, for example a fast disclosure against a slow one, and read the two reports side by side. Each run costs $1 USDC and takes about 10 minutes.
MiroShark builds a knowledge graph of the people, groups, and claims in your seed and casts grounded personas from it, each with its own beliefs, incentives, and posting style. A short prompt seed gets an automatic live-web research sweep first, so the cast reflects the current landscape.