A working model of you.

Not the version you describe at dinner. The one your decisions have been arguing for all along.

The problem

Every tool you use models your work. Your calendar models time. Your task app models output. Your notes model information.

None of them model you.

So the largest decisions of your life — what to build, who to build it with, what to walk away from — get less decision support than booking a flight. Insight arrives in fragments, at 2am or in a hard conversation, and then it evaporates. Nothing accumulates.

How it weighs you

What you say about yourself is the weakest evidence there is.

Attach evidence to a claim and watch what happens. This is not an illustration — it is running the same scoring code the product runs.

claim · behavioural0 of 5 attached

You avoid conflict with people whose approval you want.

confidence0.00

Nothing attached yet — the model holds no opinion and won’t guess one. The marked line is as far as anything you say about yourself can take a claim.

What you say about yourself

Weakest evidence in the system. Anchored, rehearsed, and free to give.

What you actually did

A real choice, where picking one thing meant losing another.

The first number on each row is its weight in the table the whole system runs on. The second appears once you attach it: what that piece was actually worth here, after age and repetition. The top two happened in one conversation, so the second is discounted — saying it twice in one sitting is one observation, not two.

The part that matters

Most tools that claim to understand you are built to agree with you. This one is built to notice the gap between what you say and what you do — and to say so.

tension · stated vs revealed

You describe independence as the thing you’ve organized your life around. In four of the five decisions where independence and approval were genuinely in conflict, you chose approval. Both exceptions came in years you describe as your best.

confidence 0.68 · 9 pieces of evidence · you can dispute this

That sentence is the product. Everything else is scaffolding for producing more sentences like it, more accurately, for longer.

How it knows it’s right

Feeling understood is easy to fake. So the model is scored.

Before you make a decision you’ve logged, it writes down what it thinks you’ll choose and how confident it is. Then reality settles it.

predictionsealed 2027-04-02

Will choose the smaller company.0.71

resolved 2027-04-19 · correct · brier 0.08

You can see its accuracy. You can see whether it’s overconfident. When it’s wrong about you, that’s on the record too — and it’s often the most interesting thing on the page, because the places it can’t predict you are usually the places you’re changing.

Ways in

There is no gate.

You reach something real on the first visit. None of these is a prerequisite for another — they build models of different shapes, and it will tell you which shape yours is.

60 secOne forced choice. Two things you value, in genuine conflict. Pick one, get a single non-obvious observation back. No account.0.15

5 minImport what you’ve already written. A ChatGPT, Claude or Gemini export is years of you thinking out loud. First claims inside twenty minutes.0.50

25 minFive decision autopsies. What you told yourself at the time, and what you did instead. The richest thing you can do in under half an hour.1.00

2.5 hrsThe full interview. Resumable, in any order. It reaches what you avoid and how the last person angry at you would describe you — which no import will ever contain.full

ongoingLog decisions as you make them. It predicts before you choose, and scores itself when reality settles it. This is the part that makes it get better.1.20

The right-hand column is what each is worth as evidence. It is not a coincidence that the cheapest thing to do is worth the least.

What it isn’t

Not a journal. Journals store; this derives.
Not a chatbot. It holds a position about you across years and defends it with evidence.
Not a coach. It doesn’t tell you what to do. It tells you what this looks like next to everything you’ve shown it.
Not therapy. It has clear limits and it says so.
Not a personality test. Nothing here is a type. Everything here is a claim with a date on it.

Your data

The honest version.

Your words are encrypted before they are stored, with a key that is unique to you. Steal our database and you get scrambled text, not your inbox. Delete your data and we destroy your key, which makes every copy of it — including the ones in backups — permanently unreadable. Not marked deleted. Unreadable.

We can read it. A model that cannot read your life cannot model it, and anyone telling you otherwise is either lying or building something that doesn’t work. What is true is that a stolen database cannot, and that we will show you the exact list of what is encrypted and what isn’t.

Names of other people who appear in your sources are encrypted too. They never agreed to be here.

What we promise

You own it. Full export, full deletion, no negotiation.
You can argue with it. Every claim shows its evidence and can be rejected — and a rejection is recorded, not obeyed. If your behaviour later disagrees with you, it will bring the claim back and show you the receipt.
It won’t flatter you. A mirror that tells you what you want to hear is worthless. We track how much of your model is uncomfortable, and if that number starts falling, we treat it as a bug.
We don’t optimize for your attention. Time spent here is not a metric we track. A model that’s working should sometimes send you away.

Get early access.

We’re letting people in slowly. Three fields — and the last one is the one we actually read. Not what you do or where you work: the question you’re living with right now.