Skip to main content

Where this stops

The claim on this pillar is large, so the limits belong on the same site as the pitch.

A value system is a model, not you

It captures the rules you were able to articulate. It does not capture the ones you hold and have never put into words — which is most of them, for everybody.

When the AI produces a recommendation you dislike from rules you wrote, the useful response is usually not "the system is wrong". It is "there is a rule I have and never wrote down", and finding it is the single most valuable thing this pillar does.

Rules conflict, and ordering is a real choice

Two reasonable rules will collide on a real case. The product resolves it by order and weight, which means your ordering is itself an ethical position — one most people have never had to make explicit.

That is uncomfortable and it is honest. Any system that hid the collision would be resolving it for you, silently.

It does not make you consistent, it makes you legible

The point is not that you become a consistent agent. It is that your inconsistencies become visible — to you first.

The AI applies, it does not judge

It does not decide whether your values are good, and it does not correct them toward anything. If you write rules the movement disagrees with, it applies yours.

That is deliberate and it is the whole basis of Pillar 3. A system that quietly nudged citizens toward the operator's ethics would be the most dangerous thing in this product, and it would be indefensible coming from a movement whose complaint is that institutions do exactly that.

Values are not evidence

The most common misuse: reaching for a value rule to settle a factual question.

"Was the record withheld?" is not a values question. "How much withholding justifies what response?" is. The product keeps them in different places — Evidence and here — and mixing them produces confident nonsense in both directions.

What it is genuinely good at

Applying a stated standard consistently across many cases you will never personally read — and telling you when two of your own rules just disagreed with each other.