The gap

Model adaptation — fine-tuning, LoRA adapters, RAG embeddings, prompt templates — is ungoverned. There is no standard for cryptographic attestation binding an adaptation artifact to its issuing authority, no sandbox pre-activation certification, no dependency chain management, and no licensing intersection that composably tracks the most restrictive license across contributing sources. Each model provider implements its own adaptation distribution; there is no mesh-distributed, federated-trained, cross-model-portable adaptation primitive.

In defense, medical, and regulated settings, ungoverned adaptation is a deployment blocker: there is no way to verify that a skill loaded at runtime was issued by an authorized authority, was sandbox-certified, and does not violate licensing constraints.

The invention

A runtime adaptation primitive in which signed adaptation artifacts (LoRA, RAG, prompt, MoE, hybrid) carry cryptographic integrity attestation binding them to their issuing authority credential. Artifacts undergo sandbox pre-activation certification in a governance-credentialed execution environment before operational deployment. An admissibility gate doubles as a skill router — the admissibility profile, rather than capability matching, determines which skills apply to which requests.

An always-active personal layer remains active across all generation contexts. Cascade deactivation dependencies deactivate dependent artifacts upon prerequisite deactivation. Cross-model portability enables artifacts to transfer between compatible base models. Federated skill training allows participants to contribute training observations without data leaving their local environment. Composite licensing intersection applies the most restrictive license across all contributing sources.

The inventive step

Prior art loads model adaptations through provider-specific mechanisms with no governance — a LoRA adapter is a file downloaded from HuggingFace, loaded into a model, and trusted on faith. Spatial adaptation artifacts carry governance in the artifact itself: cryptographic authority binding, sandbox certification, admissibility-gated activation, dependency management, and composable licensing — making adaptation a governed mutation in the spatial mesh.

The departure is that adaptation artifacts are not files — they are governance-credentialed mutations in the spatial mesh, subject to the same admissibility evaluation, lineage recording, and licensing constraints as any other governed operation.

Alone, and in composition

On its own, spatial adaptation artifacts enable governed model adaptation for defense field deployment, industrial robotics adaptive updates, medical device software updates, and regulatory-aware LLM fine-tuning — any setting where runtime skill loading must be auditable and governed.

In composition, spatial adaptation artifacts enable the spatial mesh to adapt its behavior at runtime — new sensing capabilities, new actuation modes, new coordination patterns — loaded through the same governance chain as every other mutation.