Principles

What a designer should believe

Patterns document how products solve problems; principles state what to believe about solving them. No principle publishes on aggregate evidence alone — it must hold in at least three distinct contexts from two independent vendors, and its boundaries must be mapped. A principle with no known limits is either trivial or untested.

Validated

“Make trust scopes visible and editable.”

Trust granted invisibly is trust that cannot be revoked deliberately. When what an agent may do — and what it remembers — is a first-class object the human can …

4 evidence · 1 boundary condition

“Prefer reversibility to permission.”

Every ask spends human attention; reversibility spends none until something goes wrong. Where actions can be made cheap to undo (checkpoints, sandboxes, rewind)…

4 evidence · 1 boundary condition

“Show intent before action.”

Humans review the shape of work far better than they review individual operations. An agent that externalizes what it is about to do — a plan, a preview, a narr…

4 evidence · 1 boundary condition

Watching

“Calibrate the judge with a few human labels and show the disagreement.”

Labeling each case is a research project; labeling none leaves the judge unchecked. A handful of human labels on the cases the judge is least sure about tunes i…

2 evidence · 0 boundary conditions

“Grow from five examples to five hundred in one interface.”

Evaluation becomes a habit when it starts small and stays in one place. Five hand-picked cases run the same way as five hundred, with the same comparison and th…

2 evidence · 0 boundary conditions

“Show the judge's prompt, its samples and where humans disagree with it.”

A model may score the answers, and humans will trust it only if they can see how it decides. The prompt the judge runs on, the cases it scored and the verdicts …

3 evidence · 1 boundary condition

“Show quality, cost, speed and safety as a profile, not one number.”

One number is easy to read and hides the trade a change made. A prompt that answers better and costs twice as much is a decision, not an improvement. Four chang…

2 evidence · 1 boundary condition

“Show the prompt change beside the result change.”

A score with no link to the edit that caused it cannot be acted on. The person sees the two prompt versions and the two sets of answers on one screen, and a reg…

3 evidence · 1 boundary condition