§02d Jupiter — Discrete Composition — Cross-model composition
Mars® Spec › §02d Jupiter — Discrete Composition › Cross-model composition
← The model-as-map layered architecture · Section index · Subject-directed determinations →
4. Cross-model composition
Cross-model composition is the composition of discrete representations formed under different formal domain models (§01). Two representations formed under distinct domain models may be composed when they share one or more orders — that is, when both domain models declare the same order, and both representations carry owner-sets governed by that order.
4.1 Order as the composition interface
The shared order is the composition interface. Two domain models that declare the same order — the same order identity, the same order-shape specification — are compatible at that order for the purpose of cross-model composition. The composition proceeds at the shared order’s owner-sets exactly as in same-model composition (§2.2). Two orders of different models are the same order when, and because, their content instantiates the same order shape, irrespective of label or name — sameness is shape identity, not string match.
Two domain models that use different order identities for what may appear to be the same concept are not cross-model composable at those orders under this architecture. Equivalence between distinct order identities may not be asserted by the composing deployment, by a downstream consumer, or by an LLM-assisted equivalence judgment. Where two models appear to share a concept but carry distinct order identities, the composition must proceed through a registered cross-domain typed interface (§02c §3) — not by direct order unification. Attempting direct unification of distinct order identities is a structural violation of R2 (§02c §3).
Modality-agnostic combinatorial analysis of orders. The cross-model composition and the combinatorial analysis of orders defined in this section — the per-order common structure, same-order composition across models by shape identity, cross-order composition, and the construction of higher-order orders across models — are not limited to discrete representations. They apply to the combinatorial analysis, combination, composition, synthesis, or derivation of one or more orders whether the orders, representations, or analyses are discrete or non-discrete — including continuous, statistical, probabilistic, embedding-based, numerical, or neural representations and analyses — the discrete representation genus being one modality. The combinatorial analysis of orders across one or more models is governed by the same order-shape contract and recomputation witness obligations regardless of modality; a probabilistic or embedding-based cross-model composition is not a distinct architecture claimable separately from the order-governed cross-model composition framework. A deployment implementing combinatorial order analysis for non-discrete representations without the per-order common structure, shape-identity sameness criterion, and composition witness is non-conforming.
4.2 Cross-model higher-order outputs
Cross-model composition may produce cross-model higher-order outputs — analysis findings that are expressible only at orders that neither constituent model defines in isolation but that emerge from the join of the two models’ order structures.
A cross-model higher-order output is valid under this architecture if and only if:
- The output is attributed to a specific cross-model order — a named order that is declared in the composition’s order-structure extension record and that is derivable from the shared orders of the two constituent models by a declared derivation rule
- The derivation rule is registered, independently recomputable, and carried in the composition witness
- The cross-model order’s shape specification is declared and enforced: elements in the cross-model order’s owner-set satisfy the same order-shape contract as elements in any constituent order’s owner-set
A cross-model output that is not attributed to a declared cross-model order, or whose derivation rule is not registered, is returned as a typed gap referral with locus domain model.
4.3 Composition witness extension for cross-model composition
The composition witness (§2.4) is extended for cross-model composition with:
| Field | Content |
|---|---|
| Model identity pair | The content-addressed identities of the two (or more) domain models participating in the composition |
| Shared order declarations | For each shared order: the order identity and the shared order-shape specification, confirmed identical across participating models |
| Cross-model order declarations | For each cross-model higher-order output: the cross-model order identity, the derivation rule identity and version, the contributing shared orders, and the order-shape specification for the cross-model order |
| Model-pair compatibility record | A signed record asserting that the identified shared orders carry identical order-shape specifications across the participating model versions; signed by the analysis authority performing the composition |
The model-pair compatibility record is not asserted — it is verifiable. An independent inspector confirms it by: (i) retrieving both model versions; (ii) locating the declared shared order in each; (iii) confirming the order-shape specifications are identical. A compatibility record that does not survive this verification is a composition defect.
← The model-as-map layered architecture · Section index · Subject-directed determinations →