A universal knowledge framework demotes content taxonomies to defaults and keeps answerability

Type: kb/types/note.md · Status: seedling · Tags: document-system, foundations

A framework that aims to serve any knowledge base cannot keep first-order content taxonomies — closed lists that classify what knowledge is (exactly three content modes, exactly five link labels, one fixed type set) — as universal rules. Every such taxonomy is distilled experience from the knowledge bases its authors have already operated, and a genuinely new kind of KB will break it. What survives at the universal level is second-order: the obligation that every collection declare its contract (quality goal, conventions, maintenance semantics) in a form writers, reviewers, and tools can load — plus one domain invariant that keeps the framework a knowledge framework rather than a document framework: every artifact must be answerable — capable of being wrong or stale relative to something outside itself.

Why closed taxonomies fail at universality

The failure is induction from too small a sample. A framework's first taxonomies are abstracted from the KBs its authors have seen — usually one. The rules are real: they encode what made that KB work. But their scope is the kind of KB they came from, and nothing in the rules marks that boundary. A first-person committed methodology KB genuinely needs claim-shaped titles and "do I still believe this?" maintenance; a stance-neutral evidence map is broken by exactly those rules. The rules were never wrong — they were profile features mistaken for universals, because with n=1 the difference is invisible. Universality makes the mistake visible: each new kind of knowledge that arrives ("everything goes") falsifies another rule that had quietly assumed the old kind.

Closure did real work; guards and defaults recover it

Demote, don't delete. A closed taxonomy earns its keep three ways, and each has a second-order replacement:

  • Routing. "This collection is theoretical" lets an agent infer the writing goal from one word. Recovered by keeping the taxonomy's entries as named default profiles — proven bundles a new collection adopts in one line.
  • Growth brake. "The list is complete" blocks speculative additions. Recovered by a worked-case guard: new entries are admitted only after surviving use in a real collection, never from anticipation.
  • Interoperability. Shared labels mean readers recognize conventions across collections. Recovered by a shared catalogue that collections select from and extend, rather than a closed set they must fit.

The guard is the load-bearing piece: open sets without admission discipline proliferate until no convention is shared, which is the other way to fail at being a framework.

The invariant that stays closed: answerability

Universality cannot mean "any text." The boundary that keeps the framework coherent is that its artifacts are knowledge: each is answerable to something beyond itself and can therefore be wrong or stale. The relation varies — a claim answers to the world, a description to a system, a prescription to the outcomes of following it, an attributed position to what the party actually asserts, a capture to its source — but having such a relation does not. First-person commitment ("do I still believe this?") is one answerability relation among several, not the definition of the framework.

This yields an operational scope test instead of a genre list: a collection belongs in the framework iff its contract can state, non-vacuously, what its artifacts answer to and what makes one stale. A poetry collection cannot fill those fields; a stance-neutral debate map fills them differently than a methodology KB, but fills them. Maintenance is then universal in form — every artifact has revision conditions — while fully local in content.

What else survives as universal

Only what derives from the consumer or from the second-order layer itself, not from any kind of knowledge: bounded context (economy pressure comes from the reader's architecture, so it applies to every contract — since context efficiency is the central design concern in agent systems); and the declaration obligation itself (a collection without a loadable contract is an operational defect regardless of its content).

Evidence

Three instances of the same demotion, all shipped, each keeping machinery while opening the set: types became path references, so the type set is open and collection-local (ADR 018); link vocabulary became collection-owned selections from a shared catalogue, replacing one universal set (ADR 019); the three registers became default text-contract profiles (ADR 042). Status/lifecycle semantics are a predicted fourth instance: the note type's status field currently fuses structural lifecycle with first-person endorsement, which is one answerability relation hardwired one level too high.

Caveats

Not every closed set is a content taxonomy. Sets fixed by the consumer's architecture or by the framework's own machinery (e.g., the syntactic shape of frontmatter, the existence of a lifecycle) may stay closed; the test is whether the set classifies what knowledge can be (open, demote to defaults) or what the machinery is (may stay closed). And the claim is about frameworks that aim at universality — a single-purpose KB loses nothing by hardcoding its profile; the cost appears only when the rules are exported.

Relevant Notes: