ifc-0248

17.8.6 Cross-domain transport (CTTE–5)

CTTE–4 constructs a presentation inside one fixed observation and operation language. CTTE–5 asks whether that constructor can survive a change of domain presentation. The source interface is frozen before target generation: eight abstract observation ports, four one-generator constructors, four external operations, and the CTTE–4 thresholds. Four evaluator-held domains then independently permute and rename every observation and operation, change all measurement units and probe-response gains, and use disjoint local vocabularies.

Transport requires more than replacing nouns. A separate bridge split contains answer-independent structural fingerprints and calibration measurements. From these records, the system constructs a typed bijection between observation ports, estimates the target-to-source unit transformation, constructs a typed operation alignment, and estimates response-gain transport. It pulls the target residual back to the frozen interface, runs the unchanged constructor, and pushes the new presentation forward under fresh target names. Evaluation then canonicalizes the result through the inferred alignment. Thus a target object can have a name never seen in the source domain while still counting as the same abstract construction up to typed sketch isomorphism.

The preregistered benchmark contains 2,800 ledgers. In each of four regular domains, 100 examples are generated for each closed, factor, product, exception, memory, and compound family. Two additional controls test the admission boundary. In 200 ambiguous-alignment worlds, two local ports have the same structural fingerprint, so a bijective interface map is not identifiable. In 200 blocked-probe product worlds, the target domain lacks both operations that distinguish the factor and product presentations. These require abstain-alignment and abstain-no-probe, respectively; neither is permission to choose the most familiar theory.

CTTE–5 passed all sixteen registered gates. It recovered both observation and operation alignments, the structural action, and the target presentation up to isomorphism in every regular world. It also returned the registered status in every closed, compound, ambiguous-alignment, and blocked-probe control, with no wrong extension admission. Literal and lexical reuse each achieved only \(30.29\% \) because local names and positions did not preserve the interface. Recovering observation alignment but omitting probe transport achieved \(78.57\% \). The remaining \(21.43\)-point gap localizes the importance of moving experimental operations along with the theory.

System	Action accuracy	Transported artifact
Literal reuse	\(30.29\% \)	none
Lexical reuse	\(30.29\% \)	names only
Aligned fit without probes	\(78.57\% \)	observation interface
CTTE–5 typed transport	\(100\% \)	theory and operations
Transport oracle	\(100\% \)	hidden interface map

The result establishes a narrow finite commutation property: in these generated worlds, constructing after the registered alignment agrees with transporting the abstract construction. The experiment checks typed port and operation correspondences, not preservation of arbitrary identities, composites, or natural transformations; “functor” in the archived protocol is therefore best read as shorthand for this finite interface map. It does not show that scientific ontologies can be aligned from unrestricted prose. The bridge fingerprints are supplied, unusually clean, and explicitly typed; the source grammar remains fixed. This qualification is precisely what separates CTTE–5 from the GLP-1 challenge, where the relevant biological interface is only partially known.

Experiment: CTTE–5: provenance-safe domain transport. Input: 2,800 target ledgers with permuted local vocabularies, units, operations, and disjoint bridge, discovery, and probe sources.
Frozen structure: the CTTE–4 abstract interface, four constructors, one-generator budget, and admission thresholds.
Constructed artifact: inferred typed interface alignments, a transported experimental operation, and fresh target-named finite presentation.
Result: \(100\% \) action, interface-alignment, and presentation-isomorphism recovery; zero wrong admissions; all negative-control abstentions correct.
Baselines: \(30.29\% \) literal/lexical reuse and \(78.57\% \) observation alignment without probe transport.
Verdict: admitted as calibrated typed transport, not open-domain ontology alignment or biological transfer.