About Causality

What makes it different

Ask a chat assistant the same question and you get a paragraph: fluent, confident and hard to check. Causality gives you something you can take apart instead.

• A model, not a paragraph. Six to twelve cards, at most three steps from the cause. Every effect has its own direction, size and confidence, so you can agree with one and doubt the next.
• Evidence is graded against your question. A study about another country or another decade is marked as carried across, not passed off as proof. Evidence that is on topic but does not settle the claim says so. And a claim nobody checked looks different from one that was checked and found nothing.
• Sources are real. Every link comes from a web search and names who published it. The AI writes the reasoning; it cannot write a web address.
• No invented numbers. The model reasons in direction, size and confidence. A figure appears only where a source stated one, shown with the source’s own words.
• Disagreement stays visible. When sources conflict, both are kept and shown, not averaged into one figure.
• Honest stops. If a search runs out of time or budget, you are told that it stopped. That is never presented as “there is no evidence”.
• You stay in charge. Before anything is built, the AI shows what it understood and marks what it guessed. Nothing is modelled until you confirm it.

What you can use it for

Anywhere a change in the world needs thinking through before it is argued about:

• Exploring a scenario. A new policy, a technology shift, a market shock. See which effects are likely to follow, and which of them the evidence actually supports.
• Preparing a discussion. Walk into a meeting, workshop or seminar with a structure everyone can point at and challenge, card by card.
• Finding the weak links. The model shows where the reasoning rests on assumptions and where the evidence is thin, which is where your own research should start.
• Teaching and learning. It shows the difference between an effect that follows and one that is merely plausible, and between evidence that fits and evidence that only sounds relevant.
• Keeping and sharing your thinking. Save a scenario to your account, and share it by link when you choose. You can revoke the link at any time.

What it is not: a forecast, a simulation or advice. It reasons qualitatively, and a single run is not a study. Treat a model as a structured starting point for your own judgement.

How it works

1 · Ask

You write a what-if question in your own words.

2 · Define

The AI reads it back as a scenario: what changes, where, over what period, and what to look at. Anything it inferred rather than read is marked. You correct it and confirm.

3 · Build

It searches for sources, grades how well each one fits your scenario, and builds the causal model. The server then checks the model before you see it.

4 · Explore

Select any card to see why it follows, how sure the reasoning is, and the sources behind it, each with the reason for its grade.

The exact rules — what the AI may and may not claim, how evidence is graded, and what each term means — are on the AI & evidence page.

Try it

Ask your own what-if question, or open the worked example from the front page. The example is a recording of one real run, so it costs nothing and needs no account.