Ontology Packages

DataScript is an EAV store — [entity attribute value] is a triple, the same decomposition as RDF's subject predicate object. So the question was never "triples or no triples." It's globally-named RDF triples + OWL vs. local EAV datoms + Datalog. Same fact primitive; different everything-around-it.

The one axis it all hangs on: RDF is the across-boundary stack (built for parties who share no runtime). Datalog/EAV is the within-boundary stack (one fact base, one trust boundary). Every Onto decision is a variant of "this happens inside a boundary → Datalog, not RDF."

	Onto — Datalog / EAV	RDF — triples / quads
Fact	`[e a v tx]`, temporal (tx is free 4th dim)	`s p o`, atemporal (quad adds a graph slot)
World	Closed — absence = false, so rules fire	Open — absence = unknown
Execution	Reactive Rete, incremental (push)	Query / SPARQL (pull)
Naming	Local keywords; you are the authority	Global IRIs; no central authority
Compose	Unify — one term, the other an alias	Federate — two terms, bridged (`sameAs`…)
Constraint	O'Doyle `:violation/*` rules (engine-internal)	SHACL shapes (portable, publishable)

The boundaries you draw. One runtime, three composition seams — a different tool at each, and RDF's sameAs at none:

Runtime (one fact base) → Datalog, closed-world, reactive. RDF nowhere near it.
Within a context → unify: collapse to one canonical term; the other is a display alias, no entity. No bridge, no translation, nothing to rot.
Across contexts (a subdomain boundary) → refuse to unify: leave the two distinct and unbridged. Deliberate, not federation's "coexist and glue."
Vendor / API boundary → ACL: translate the foreign model into your term and discard it. Translation, not identity — the constructive alternative to sameAs.

What makes unify-composition actually work (the load-bearing bet): composition is asymmetric — utility domains attach into verticals, never vertical-to-vertical — and overlap entities are roles on kinds (OntoClean rigidity), so one Person carries many roles instead of many competing Person-kinds. Variable-base roles (Buyer/Payee) need a Party supertype (likely already yours as REA's Agent).

The one honest exception. RDF/SHACL earns a seat only if a boundary appears where you currently have none — regulated cross-institutional interchange, or a public registry whose format external tools must consume and validate. Even then it lives at the publication/interchange layer, surfaced as JSON-LD; the EAV/Datalog core stays exactly as it is. Runtime and interchange are separable — that's the seam to watch, not the substrate.

this concept boundary strategies diagram - i thought we were settling on eric evans concept at the boundary being important - what is written for onto / unficiation seems different

Reconciling unification and ACL as Evans-aligned boundary tactics