A decision is never just one decision.

One choice reshapes the next.

A general-purpose model can reason well about a single question. But real decisions aren't isolated. A career move changes where you live. Where you live changes who you meet. A relationship changes your priorities. Every choice quietly rewrites the constraints, opportunities, and state from which the next choice gets made.

So the unit we care about isn't the question — it's the trajectory. Answering “should I take this job?” well means reasoning about the path it opens or closes, not just the job in isolation.

Two axes: what now, and what it opens.

We think about a decision along two axes. Horizontally: given your current state, what are the actions actually available, and which fits who you are right now? Forward in time: how would each action reshape your future state — the constraints, the opportunities, and the decisions that become possible after it?

The question isn't only which action is best. It's which path creates the strongest set of next decisions.

An honest note on where this stands. Today, Timing already reasons about a choice in the context of your state, your goals, and your timing. The fuller version — simulating multiple paths forward and optimizing across them — is what we're building toward, and where a large share of our research goes. We'd rather tell you that plainly than dress a roadmap up as a finished feature.

Useful on day zero.

A model of a person has an obvious cold-start problem: on the day you arrive, it knows almost nothing about you. Most personalized systems are close to useless until they've watched you for weeks.

Timing starts differently, because it starts with structure. From your birth data and the timing frameworks, it can form a backdrop — a set of priors about disposition and rhythm — from the very first session, before a single day of behavior. That's day zero: not a blank slate, but a hypothesis about you, ready to be corrected.

Then day one begins. Every reflection, every biometric reading, every outcome sharpens the picture and pulls it away from the generic prior toward the specific you. The frameworks give the model something to say early; your real life makes it right over time.

The loop that makes it learn.

None of this holds still. Understand → simulate → recommend → observe → learn. Every recommendation creates new information: what you chose, how you felt, what actually happened, whether it moved you toward your goal, and which signals turned out to matter. Real outcomes close the loop, and the model uses them to reweight what it trusts — for you specifically, not for people in general.

This is the same discipline we apply to every signal, cosmic or biometric: it earns its weight against what actually happens, and loses weight when it stops helping. Over time the aim isn't only to learn what tends to work for people — it's to learn what tends to work for you.

What we're seeing so far — honestly.

Does the extra context actually help? Here's our first read. We're sharing the method alongside the numbers, because a number without its method isn't evidence — it's decoration.

We take the self-reported numbers as encouraging, not conclusive. What would move us from “encouraging” to “proven” is harder and slower: retention, repeated use, and outcomes we can measure without asking. That's the bar we're holding ourselves to, and it's the work of the next study.

A good decision is rarely about one moment. Neither is a good model of the person making it.