Can a rulebook express transitive closure?

Yes. This is the most-asked question in the project's history — and the one most often answered wrongly about us, occasionally even by old copies of our own docs — so here is the precise answer.

The theorem (which is real)

Transitive closure is provably not expressible in first-order logic with aggregates (Hella, Libkin, Nurmonen & Wong, "Logics with Aggregate Operators", J. ACM 2001; Libkin, "Expressive Power of SQL", 2003; extending Aho & Ullman, 1979). The rulebook's formula/lookup/aggregation fragment sits exactly in that class. So no formula field, however large, computes unbounded reachability. That part is settled mathematics.

The conclusion (which does NOT follow)

The wrong conclusion is "...therefore the rulebook can't do closure." The theorem is the reason closure is its own declared primitive — not a wall the rulebook stops at. A closure field is one line in the model — { "name": "PrecedesStepClosure", "type": "closure" } — declared over a relationship, optionally with an edge filter. Every substrate emits that one declared field: a WITH RECURSIVE view in Postgres, an owl:TransitiveProperty in OWL, an explicit graph traversal in Python, Go, and TypeScript. These are emissions of a declared field, no different from a COUNTIFS becoming a SQL aggregate. Downstream calculated fields consume the closure like any other field — and the conformance harness grades every closure cell across substrates rather than assuming agreement.

It runs, today, including the hard cases

• talismans-special-solutions declares PrecedesStepClosure, DelegationClosure, and DerivationClosure; conformance tests assert the inferred pairs, and ordinary aggregation fields count them (6 inferred precedence pairs; 10 total closure pairs).

• procedural-knowledge-ontology runs closure over a cyclic step graph with rework loops (LeadsToClosure — steps on a cycle correctly reach themselves), plus a filtered-edge closure (LeadsWithoutHumanGateClosure) that detects publications which bypassed every human approval gate.

• The latest conformance run graded 473,312 cells, and every substrate with a closure engine agreed on every closure cell.

Exactly like OCL adding closure(), SQL adding WITH RECURSIVE, and OWL having transitive properties: every serious framework grows a closure primitive, because the boundary is a theorem. The rulebook simply declares its primitive in the model itself, where it belongs.