Admission and identity
Manifest validation, signatures, non-human identity, proof of possession, replay controls, and registration authorization.
Technical credibility should come from inspectable engineering—not adjectives.
Legation’s sales story is strongest when architecture claims are tied to code, tests, deployment artifacts, threat-model boundaries, and clearly stated limits.
01 / Codebase substance
A real platform workspace—not a topology diagram.
RustMemory-safe implementation language
0Unsafe Rust permitted across the platform
ModularIsolated delivery, governance, transport, and evidence layers
4Deployment profiles, commercial to sovereign
The repository includes runnable services, control-plane components, protocol adapters, customer Embassy deployment, compliance evidence, observability, key custody, and operational interfaces.
02 / Implemented control areas
Defense in depth across the full request path.
Execution governance
Mandates, policy gates, projection, redaction, recall, data sensitivity, dispatch controls, and security sensors.
Evidence and operations
Transparency mechanisms, SIEM events, compliance snapshots, observability, portal surfaces, and deployment automation.
Closed egress gate
A closed, versioned egress schema gate (legation-enclave::evidence::validate_egress) at the single projection chokepoint validates every payload against the closed schema for its kind; an unknown kind fails closed. The MCP-direct path (legation-mcp) is closed the same way with deny_unknown_fields and a mcp_projection_is_closed guard. The first typed payload—ControlEvidenceSet for CMMC / NIST 800-171—carries closed enums and references only (SHA-384 digests, bounded ids), never excerpts or filenames.
AssuranceReceipt
An Ed25519-signed, offline-verifiable receipt (legation-compliance::receipt) binds the verdict, harness digest, sealed-bag scope, authorizing mandate, and hash-chained audit-head. Any tampered field breaks the signature; it verifies with no clock and no network.
03 / Testing posture
Tests are evidence of engineering—not proof of authorization.
The platform maintains an automated test suite, clean formatting and linting, and a workspace-wide prohibition on unsafe Rust. This supports technical confidence, while independent validation remains necessary for the strongest security and compliance claims.
| Evidence | Status represented by repository |
|---|---|
| Workspace tests | 505 passing across 19 crates |
| Clippy | Reported clean across all targets |
| Formatting | Reported clean |
| Unsafe Rust | Forbidden across workspace |
| Independent penetration test | Still required |
| Formal authorization | Deployment and assessor dependent |
04 / Deployment assets
The operating model is expressed as infrastructure.
- Terraform for regional AWS hub infrastructure
- Customer Embassy deployment modules and Helm assets
- Commercial, GovCloud, EU, and sovereign examples
- Identity, secrets, logging, WAF, routing, and audit-retention resources
- Architecture and operational documentation
05 / Claim discipline
Say exactly what the deployed tier can prove.
Recommended language“Designed for,” “supports a deployment path for,” and “provides controls aligned to” are stronger than making unsupported certification or authorization claims.
| Claim | Safe positioning |
|---|---|
| Customer data locality | Customer-side execution and projection are architectural controls; validate each workload’s data path |
| Vendor IP protection | Compiled cargo and contractual controls below Sovereign; hardware-backed host resistance at Sovereign/TEE |
| FedRAMP / IL5 | GovCloud and compliance-aware deployment path—not authorization by default |
| FIPS | FIPS-oriented endpoints and modules require exact version and validation confirmation |
| Post-quantum | Algorithm support requires interoperability, custody, and compliance validation |
06 / Validation roadmap
The next moat is proof under hostile conditions.
- Independent cryptographic design review
- Penetration and workload escape testing
- Multi-region failure and Redis consistency testing
- HSM and TEE end-to-end validation
- Reproducible build and supply-chain verification
- Performance benchmarks with realistic agent and MCP workloads
- Assessor review of generated compliance evidence