The AI doctrine reference
DoctrineBase is a structured home for AI doctrines — principles, project rules, and emerging practice, each written in one consistent shape and kept current as the field moves.
Every entry, one shape
What every entry gives you
Six things every entry carries, so a doctrine reads the same whoever wrote it.
Each doctrine is a named principle with a statement, rationale, and status — not a loose note buried in a doc.
See an entry →Draft, active, deprecated, superseded. Every entry says exactly where it stands and what replaced it.
See the statuses →Alignment, coding, governance, safety, evaluation, prompting — browse doctrine by domain, not by scattered folders.
Browse domains →See which doctrines refine, depend on, or supersede which — so you read a principle in the context of the ones around it.
See relations →Every claim links back to where it came from — papers, posts, and prior doctrine — so nothing rests on memory alone.
Follow a source →Every change is tracked, so you can see how a doctrine evolved — and why it says what it says today.
View history →How it works
Write the principle down in one consistent shape: a clear statement, the rationale behind it, and its sources.
File it under a domain, link the doctrines it relates to, and set its status — draft, active, or beyond.
Review on a cadence: promote, deprecate, or supersede entries as practice moves — with the history kept.
The body of knowledge
Doctrine lives in domains, each holding a set of entries at different stages of maturity — some settled, some still forming, all in one place and kept current.
Active
Emerging
Under review
“We stopped keeping doctrine in scattered docs and Slack threads. Now there’s one place that’s actually authoritative.”
Head of AI Practice · Meridian Labs
Browse the base, or contribute your first doctrine. Open to your whole team.
Browse doctrines