Your question. Your assistant. Your choices.

You should be able to explore a possible future without learning to operate simulation software. Tell an AI assistant what you want to explore. It works out what the question needs, prepares the simulation, runs the approved work and helps you understand the result. You see and direct the work in your browser.

That is AI's role in Whathappensify's democratic purpose: put specialist analytical capacity within reach of people who have never had it. Your knowledge of a place, your question and your judgment guide the work.

From your question to an approved scenario

Suppose you are choosing among three permitted spots for a pupusa truck. The assistant helps identify which customers you want to serve, when they are nearby, what information is available and which alternatives are worth comparing. You bring what you know about your food and customers. It turns those choices into a proposed simulation you can inspect.

Before the scenario is generated, you review and approve the setup in the browser. You can correct an assumption, change the question or decline the proposal. Approval applies to that version and scope. Material changes come back for review.

Your control continues through the work. Ask for another comparison, explore a different possibility or decide that the next step needs evidence from people who know the place.

For a linked experiment, the assistant prepares the whole bounded sequence: the models, the quantities passed between them, the alternatives, the uncertainty design and the rules for continuing or stopping. Later inputs can come from earlier results within the scope you approve. Each run has a visible record; a failed or cancelled part cannot quietly become a successful comparison.

The assistant can find a suitable shared model, help you author one, or adapt a model you already own. It records the model's source, assumptions and review scope, asks for missing facts and leaves relationships unquantified when the evidence does not support a number. A model of a budget or a queue can use its own time and units without inventing a map or calendar.

Reading a result and pursuing the next question

The assistant explains what the simulation found and helps you follow the reasoning. The Epistemic Protocol Machine connects measurements, comparisons, explanations and revisions while preserving the evidence each conclusion depends on.

You can examine the saved record directly: what ran, which assumptions shaped it and what the results support. Ask why the busiest corner looks less promising at lunchtime, or what changes on a day when more people work from home. The assistant can show the basis for its explanation and propose what to explore next.

You can also use Perspectives to make your priorities explicit or compare the same result through other points of view. A Perspective changes how a result is read, not the result itself. Serving regular neighbors, being able to approach the truck and protecting family time may matter differently to different people. The assistant helps explore those differences; the judgment remains yours.

How the assistant connects

Every assistant reaches Whathappensify through the Model Context Protocol, or MCP. Think of it as the channel through which your assistant asks the platform to do the work. The simulation software performs the calculations and returns results with their supporting records.

The assistant can request only the work it is permitted to do. Its explanations must respect the evidence behind the results. Repeating a claim or passing it through another step cannot make it better supported.

Rosa and her assistant

1. Rosa

   What would happen if I parked my pupusa truck at each of my three permitted spots at lunchtime: by the station, at the library corner or on the market street? I want to know where the most people would walk past.

2. Assistant

   I’ll set up a simulation of people walking near each spot on a weekday lunchtime. You’ll see the setup in your browser before anything runs.

3. Assistant · operation calls over MCP
   • simulations.create opens a private simulation for your question
   • catalog.search finds public data about the streets near each spot
   • worlds.create_from_place builds the street network for the neighborhood
   • scenarios.propose writes the setup for you to review
4. Rosa’s browser
   Whathappensify · Setup for your review

   Setup for your review

   • Three spots: by the station, at the library corner and on the market street
   • A weekday lunchtime, 11 a.m. to 2 p.m.
   • Measured: people walking within two minutes of each spot
   • Assumption: a typical office day, with most people at work

   ApproveChangeDecline

   Waiting for Rosa. Nothing runs until she approves.

   Rosa approved this version of the setup. Approval applies to that version and scope; a material change comes back for review.

5. Assistant · operation calls over MCP
   • scenarios.generate turns the approved setup into a scenario that can run
   • EXPLORE.start starts a run for each spot
   • MONITOR.status checks when the runs finish
6. Returned by the platform

   Three runs complete

   INT:CS1 “suggests” from EXPLORE.start

   Each run suggests how many people walk past its own spot at lunchtime, in this model on these inputs. A run suggests; it does not show what will happen.

7. Rosa

   The library corner looked busiest when I watched the runs. Can you just recommend it?

8. Assistant · operation calls over MCP
   • CHOOSE.rank asks for a recommendation from the library corner’s run alone
9. Returned by the platform

   The platform refuses the shortcut

   refused CHOOSE.rank use instead COMPARE

   A single simulation run cannot be turned straight into a recommendation. CHOOSE does not accept interpretation-grade input, so the runs must be compared first. The call is refused before anything runs.

10. Assistant

    The platform won’t turn one run into a recommendation, and I can’t get around that. I’ll compare the runs first.

11. Assistant · operation calls over MCP
    • COMPARE.receipts computes the differences between the runs, within the model
12. Returned by the platform

    Comparison complete

    RSN:CS2 “indicates” from COMPARE.receipts

    Within the model, the comparison indicates that more people walk past the library corner at lunchtime than past the station or the market street. A difference is not a cause or a preference, and it does not choose between the spots.

13. Assistant

    Here is what the results let me say. Within this model, the comparison indicates that more people walk past the library corner at lunchtime than past your other two spots. Each run on its own only suggests what happens at its spot. None of this predicts sales or how many people would stop for pupusas. If you tell me what matters most to you, such as serving your regulars or finishing in time to collect your child, I can rank the spots against those goals. A ranking would be a recommendation, not a decision.

14. Rosa

    Thanks. I’ll think about my regulars and decide.

15. The decision stays with Rosa.

The platform's controls are its grades, claim limits, refusals and approvals. They govern what the platform returns and records. They do not govern every sentence your assistant writes about it. An assistant can misread your question, leave out an option or summarize a result wrongly. The saved record lets you check what the platform actually returned.

Explain, test and learn

The assistant can trace feedback through a model, test extreme conditions and examine which assumptions influence an outcome. It can compare policy rules, show direct effects and effects on others, and preserve the same samples and chance conditions across alternatives so the comparison stays meaningful.

When more information could help, it can compare the value of a specified observation under the choices, costs and assumptions you state. When new observations arrive, it can update a model while keeping the data used to fit it separate from the evidence used to test it. It reports what can be inferred and what remains uncertain.

You see the models, runs, comparisons and readings together in the browser. The assistant's explanation links to their saved basis, so you can follow a claim back through the work that produced it.

Keep the work, explore further

Your work can continue beyond one conversation. Return to a saved simulation, revise it as conditions change, or share suitable work so someone else can examine and extend it. Someone can make their own version, attach a different Perspective or borrow your approach for another place. You choose what to keep private and what to contribute.

The goal is a capability you can keep using: to test an idea, question a proposal and shape what happens next.

Follow Rosa’s question · Explore what you could do · Read the technical reference