ADR 0013 — capability and realization as candidate theories; artifact folded into realization

Status: proposed
Date: 2026-06-07
Builds on: 0012 (the basis and the two tests this ADR applies)
Relates: theory/0003 (the rationale and sketched judgments)

Beyond authority (thesmos) and form (horos), the two strongest missing theories are capability — what an entity affords and requires — and realization — how an immutable declaration becomes a running effect. "Artifact" is not the irreducible semantic: a program is an artifact under an interpretation relation, and the irreducible thing is the realization of potential into effect. Resource, federation, and time are recorded as open, not adopted.

Context

0012 adopted admissible as the organizing judgment and the required + irreducible tests. Applying them (theory/0003) turns up candidates — and also shows that the intuitive triad policy / structure / artifact is mis-cut. "Artifact" is a noun, not a judgment; and a single "program/effect" axis silently conflates two different questions: what does this entity afford for others to bind to? versus what does running it produce? The first is the integration surface; the second is the path from declaration to running reality. They pass the tests separately.

Decision

1. Recognize capability (affordance / interface) as a candidate theory. It answers "what does this entity provide and require?" — provides / requires / operation / message / state / requires_authority / compatible. It is irreducible over horos (shape of a value ≠ "offers this operation / needs this cell / emits on this wire") and over thesmos (may ≠ which requests are meaningful), while referencing both (the DAG point, 0012 §2). It is the most important missing layer for safe late binding.

2. Recognize realization (runtime / instantiation) as a candidate theory. It answers "how does a declaration become an effectful, running, or produced thing?" over the skeleton declaration → immutable artifact → desired state → runtime effect → observation (arche RFC 0004) — manifest / builds / host_supports / realizable / deployment_plan / realizes. The theory defines checkable claims around running; it does not run anything. The runner stays a fact-driven occupant, not a primitive (arche RFC 0001); the instantiation semantics may be constitutional even though the implementation is not.

3. Fix the artifact-vs-program category error: fold artifact into realization. CAS is the byte mechanism; an artifact is the immutable content as a fact-mediated, space-scoped object (arche RFC 0002); a program is an artifact under an interpretation relation. Artifact is not yet proven irreducible as its own theory: for now there is no standalone "artifact theory" — it is the immutable object the realization theory reasons over, with build plans and deployment plans folded in. A later pass may split artifact back out if its judgments prove non-definable from realization's (the production-vs-runtime open edge below).

4. Record the open outsiders without adopting them.

Resource / scarcity — capacity is not permission, not definable from thesmos; a serious candidate. Open.
Federation / translation — mapping authority between roots/polities; needed once "global" is literal. The most consequential open outsider: the no-central-authority thesis structurally requires root-to-root trust mapping, so it is the leading open question for "global", not a peripheral maybe. Open (not adopted).
Temporality / freshness — required, but the current lean is that it is a mechanism/tier or a per-judgment modifier, not a peer theory (time is non-monotone and would poison the query/proof layer; thesmos already runs a freshness tier over an active-set accumulator). Open.

5. Create no repos. capability and realization remain candidates, gated by the promotion path (0012 §3, theory/0004 §1). A repo is minted only on crystallization — a name, a normative spec, central judgments, conformance vectors.

Consequences

The working basis becomes four: authority (may) · form (is) · capability (affords / requires) · realization (becomes / runs) — with ontos/logos as the medium and facts/CAS/cell/wire as the primitive surfaces. The relative-completeness checklist (theory/0004 §3) names capability and realization as the two most-likely-missing core theories.
Naming is deferred; the Greek placeholders (dynamis for capability, energeia for realization) are not reserved by this ADR. constellation.json gains members only at repo creation.
Work continues in theory/: write admissible for real; decide whether resource is core; resolve temporality's status; set the concrete thresholds at which capability or realization earn a repo.
Production / derivation vs runtime realization may split later. The current model folds build and deployment into realization; a later pass may separate making (poiesis) from running if their central judgments prove non-definable from each other. Recorded as an open edge, not a second theory — and the reason fold-3 is phrased as "not yet proven irreducible" rather than "never."
If a later pass shows capability and realization collapse into one theory (their provides judgments are adjacent — the type/term, dynamis/energeia seam), this ADR is superseded rather than edited.
Realization now has a first running instantiation to learn from — not a crystallization. The provisional process.v0 fact profile (arche-runner) realizes this ADR's skeleton — declaration → immutable artifact → desired state → runtime effect → observation — directly on stele facts: a content-addressed manifest, a desired fold by generation, and started/stopped/failed observations asserted by a fact-driven realizer that effectuates while the substrate decides authorization (this ADR's "the runner stays a fact-driven occupant, not a primitive"). An inhabited closed loop ran on staging (2026-06-10): a declared desired became a real running container and an admitted process.v0.started, end to end (evidence). It stays a candidate-stage experiment — process.v0 carries v0 in every predicate so it is superseded, not promoted, when the theory firms up, and it mints no repo and reserves no name (this ADR's "create no repos"). What it adds is empirical pressure on the judgments sketched here (manifest / realizable / realizes) and a first data point for the production-vs-runtime open edge.