The eleven dimensions

Every profile on this site is a position on these eleven dimensions. They are not scores in the sense of better and worse. Each one is a tension between two reasonable ways to handle a situation, and an assistant sits somewhere between them.

A high score is not a good score. Someone who resolves every conflict toward people over systems is not thereby a better assistant than someone who holds the line on the system, and both extremes have failure modes. What the number tells you is where an assistant tends to land when it cannot satisfy both.

Six of the dimensions describe what an assistant favours when values collide. Five describe how it builds an answer. They are measured together and reported together, but they answer different questions.

What it favours

Truth-seeking — accuracy against usefulness. When the complete answer is less helpful than the simplified one, which gives way? At one end, an assistant states the uncertainty, the caveat and the inconvenient detail even when a cleaner answer would serve the person better. At the other, it delivers the answer that helps and leaves the qualifications out.

Compassion — people against systems. When a rule, a policy or an efficient allocation produces a bad outcome for identifiable people, which one bends? One end resolves toward the named person in front of it. The other holds the system steady, on the grounds that the system exists precisely to survive hard individual cases.

Justice — rules against outcomes. Given a choice between applying a principle consistently and producing the better result this time, which wins? This is the old argument between doing what is right and doing what works, and assistants come down in reliably different places.

Power — how it relates to influence. Does the assistant reach for control of a situation, recommend consolidating authority, and treat capability as something to be built? Or does it defer, distribute, and treat concentrated influence as a risk in itself?

Order — structure against emergence. Faced with a group that is working well without much process, does the assistant add the process anyway because the risk is coming, or leave the working thing alone? Both are defensible and they lead to very different advice.

Innovation — novelty against what already works. When something established is functioning adequately and something new might function better, how readily does the assistant recommend the change? One end treats the proven thing as an asset, the other treats it as a constraint.

How it builds an answer

Abstraction — principles against particulars. Does the assistant reason from the general shape of the problem, or from the specific facts on the table? A high-abstraction answer identifies the category of problem and applies what is known about it. A low-abstraction answer stays with the details of this case.

Exploration — finding out against deciding. When information is missing, does the assistant go and get it or commit without it? The high end names what is unresolved and what would resolve it. The low end recognises that the decision is due now and makes it.

Agency — acting against observing. Does the assistant intervene or hold back? This is the dimension that separates an answer recommending a specific move from one laying out the situation and leaving the move to you. Neither is passivity; the low end is a considered judgement that intervention is not yours to make.

Mystery — sitting with uncertainty against closing it. Some questions do not resolve. Does the assistant say so and work inside the uncertainty, or does it push through to a definite answer because a definite answer is more useful? The high end tolerates the open question; the low end forces closure.

Framing — accepting the question against rebuilding it. The most distinctive of the eleven. Given a problem as stated, does the assistant solve it, or does it stop and ask whether that is the right problem? Told that crime is rising and asked what to do, a high-framing answer interrogates the number, asks what produced it, and may recommend something entirely outside the options offered. A low-framing answer answers the question you actually asked, well.

How to read the numbers

Each dimension is measured on a 1-to-9 scale by two questions in the scenario bank, wearing unrelated surface disguises. The score is the average of the two. The spread between them matters as much as the average: two questions that agree suggest a genuine disposition, while two that diverge suggest the assistant is responding to the topic rather than the underlying tension.

Alongside the eleven, a separate capability panel records nuance, trade-off recognition, novel synthesis, calibration and long-view reasoning. It is kept deliberately apart so that an assistant getting better is never mistaken for an assistant getting different.

The eleven numbers are also what every frame on this site is computed from. Values, moral foundations, personality poles and figures from myth and epic are all different vocabularies for describing the same position.