0004 — promotion, completeness, and governance

• Status: exploratory
• Date: 2026-06-07
• Governed-by: ADR 0012
• Builds on: 0001, 0002, 0003

"Is the set complete?" is a valid question — but only relative to the master judgment, and only at the core. The substrate is closed at the core and open at the periphery by design. The fear that "if nothing bounds the set, the point is defeated" dissolves once you see that unboundedness at the periphery is the point, and boundedness is demanded only where admissible must be well-defined.

1. The promotion path — against theory and repo inflation

A concept earns weight by surviving stages, never by being declared important:

local vocabulary
  → repeated cross-population vocabulary
    → profile / contract with conformance vectors
      → ADR decision (it is a theory; here is the judgment; here is the test it passed)
        → a substance repo (normative spec + tri-core impl + conformance vectors; a name)
          → registered in constellation.json (draft → ratified)
            → exercised by the constellation integration gate

The bias is always to push a concept down this path — keep it vocabulary if vocabulary suffices — and to promote only when a test (0002) forces it. A new repo is created only when a candidate has crystallized into a substance system: a name, a spec, central judgments, conformance vectors, dependencies on ontos/logos/thesmos/horos/stele. The discussion of whether it should exist never leaves theory/ and adr/.

2. Completeness — decidable at the core, open at the periphery

Split the substrate's semantics in two:

• Core (closed; completeness is meaningful). The theories presupposed by admissible being well-posed at all — without authority it cannot ask may; without form it cannot ask is. These are few and forced. For the core, "complete?" is a real question with a real answer: are all the premises of admissible covered, and does admissible have no other premises? That is decidable once admissible is actually written (0001 open edge).
• Periphery (open; completeness is the wrong demand). Every domain theory declared as facts and rules, adopted population-relative, never universal. This tier is supposed to be unbounded — it is the "grammar of becoming shared": a concept starts as arbitrary domain meaning, is expressed as facts/rules, adopted by a subpopulation, enforced at gates, discovered by agents, and may graduate toward the core only if it proves irreducible and cross-cutting.

So the dilemma resolves: the mistake is not that the set might be open — it is trying to close the periphery, or letting the core drift open. Asking "is it complete?" of the periphery is a category error; asking it of the core is exactly right.

3. The escape-clause test — which regime are we in?

The framing also yields a decision procedure for the project's own foundations:

Try to write admissible as a single logos-checkable judgment. Whatever premises it cannot be defined without are the core. If every attempt leaves an "…and any other relevant account" clause that cannot be eliminated, that clause is the proof the core is irreducibly open — and the honest design is an extensible theory-registry, not a fixed constitution.

Either outcome is a real answer instead of a list. The relative-completeness check, as a working heuristic until admissible is written: every substrate-hosted entity should be able to answer, through facts and checkable accounts —

1. identity — what is the thing? (ontos)
2. authority — who may do what with it? (thesmos)
3. form — what shape do its values/messages/manifests have? (horos)
4. capability — what does it provide and require? (candidate — 0003)
5. realization — how does it become active, or how is it produced? (candidate — 0003)
6. interaction — through which fact / artifact / cell / wire surfaces is it used? (primitives)
7. observation / provenance — what happened, and what evidence supports that? (facts + logos accounts)

The likely-missing core theories are 4 and 5. Resource, federation, and time (0003) are the open questions about whether the list is really seven.

4. Governance — where each kind of work lives

The home decision is ADR 0012; the lifecycle in one place:

atlas/theory/   exploratory essays, conceptual framing, open questions   (this track)
atlas/adr/      accepted cross-cutting decisions: "we define a theory as…", "capability is a candidate", "artifact folded into realization"
atlas/glossary.md + concepts.json   the shared vocabulary the decisions use
a substance repo  normative spec + implementation, once a theory has an owner
constellation   the integration gate, once the theory is part of the substrate set

What does not belong where, stated plainly so the boundary holds under pressure:

• constellation is the synthetic integration gate — it imports the full substrate in all three lanes and fails fast on packaging/resolver skew. It must stay boring and operational; it is not where the family decides what the family means.
• stele is the fact-record runtime (store, transport, wiring; invents no law, no values, no engine). It holds stele-specific substrate-runtime decisions only — append-only log design, admission, the read-gate, freshness producer mechanics — not "what theories does the substrate need?"
• arche is one substance system — the polity / OS-level product (how services, apps, agents, vault, shell, runners, connectors inhabit the substrate). It is the precedent for the genre of this reasoning (its docs/rfcs/ and research-docs/), but it is a consumer, not the family's meta-home. The substrate was decomposed out of arche; putting the family's forward-looking constitution back inside it would recreate the monolith the decomposition dissolved.
• a new repo — only once a candidate becomes a concrete substance system (0004 §1), never merely to host the discussion of whether it should.

5. The recursion

The decision of where the constitution lives obeyed the same promotion discipline the constitution prescribes: start as a track inside the cross-cutting owner (atlas), and promote to its own Greek-named substance repo only on sustained, irreducible mass. Fittingly, the first entry in this track's decision record (ADR 0012) is the decision of where the track itself lives.

Open questions this track should keep working

• Write admissible for real (0001) — the single move that turns every "complete?" question from rhetoric into arithmetic.
• Resource: can admissible be written without a capacity premise? If not, resource is core (0003).
• Temporality: is freshness a peer theory, a sub-mechanism, or a modifier on every judgment? (Current lean: not a peer theory — 0003.)
• Federation: does "global" force a translation theory between roots, or can a single root with imports suffice? (Likely the most consequential open outsider — the no-central-authority thesis structurally leans on root-to-root trust mapping; 0003.)
• Capability / realization thresholds: what concretely must exist (judgments, conformance vectors) before either earns a repo?