What is a Principle?

A principle is a reusable behavioral understanding extracted from experience.

It helps guide decisions across different situations.

A principle answers: "How should I think here?"

Not: "What exact action must I take?"

In PD, principles come from real agent behavior: recurring deviations are captured as evidence, distilled through reflection, and become principles only after the Owner approves them.

Principle vs Rule

Principles and rules are not synonyms. They are two sides of the same coin — one face points to generalization, the other to specificity. (This distinction comes from our essay The Alchemy of Soft to Hard Rules.)

	Principle	Rule
Purpose	Guide decisions	Control actions
Source	Experience abstraction	Operational requirement
Scope	General	Specific
Question	How should I think?	What should I do?

Saying "a principle is a better rule" misses the point:

Principle guides judgment.

Rule constrains specific actions.

A principle like "understand the current state before acting" applies across projects, tools, and scenarios — it needs interpretation. A rule like "pushing without pulling the remote branch is blocked" is specific, mechanical, and leaves little room for interpretation.

They work together. Where repeated experience proves a bottom line must never be forgotten, the Owner can promote a principle into an executable rule that runs at the runtime layer — shadow-tested first, activated only by explicit approval, and always reversible. Candidate rules may be drafted automatically, but hardening never happens on its own: generation produces proposals; the decision — and the go-live — stays with the Owner.

Principle vs Memory

Memory answers:  What happened?
Principle answers: What should we learn from what happened?

Memory stores and retrieves information. A principle is distilled from experience and changes how the agent acts next time — which is why session memory remains the host's job, while behavioral improvement is PD's.

Principle vs Prompt

Prompt: tells the agent what to do, before execution.
Principle: learned from what actually happened, after execution.

A prompt is an instruction written ahead of time. A principle is understanding abstracted from real outcomes — so it survives across sessions, tasks, and contexts instead of living in one conversation.

Where principles live in PD

An approved principle is applied through governance channels: prompt guidance today; and for the few vital bottom lines, owner-promoted executable rules at the runtime layer. See From Experience to Principle for the full journey.