Every industry pays for the representation it chooses.
Most pay in dense form: everything materialized, everything moved, every scale-up bought in hardware. The pages below are where choosing differently changes the bill.
Measure the structure. Choose the representation. Sign the consequence.
Engineered physical systems
Hypersonics, propulsion and thermal systems, plasma and kinetic devices, nuclear and regulated aerospace. Coupled disciplines, expensive physics, and design campaigns where many high-value quantities are infeasible to represent densely.
The engine builds representation-aware response surfaces across topology changes and tested validity regions, propagates uncertainty natively, and returns results stamped in or out of the validated domain, sealed with a post-quantum signature.
AI infrastructure
Two workloads, one buildout. Thermal-fluid simulation of AI factories and high-density datacenters, so cooling strategy, layout changes, and failure scenarios are evaluated before capital commits. And the inference data plane as one signed runtime on GPU infrastructure the customer already operates.
High-consequence records
Compliance evidence, discovery productions, cyber incident artifacts, public-sector case files, financial model outputs, healthcare documentation, utility inspection records. Outputs that get challenged after the fact. Each registered artifact is bound to an ML-DSA-signed receipt with tamper-evident integrity records, and anyone holding the bundle can verify it offline.
Cryptography is distributed across code, libraries, TLS endpoints, certificates, keys, databases, and vendor products, and most estates have no authoritative inventory of what is quantum-vulnerable. The transition runs Discover to Quote to Migrate to Protect to Assure, each stage emitting signed artifacts: CycloneDX CBOM, deterministic signed quotes, ledgered migration events, posture snapshots.
OMB M-26-15 requires covered federal agencies to submit a PQC Migration Plan to OMB and ONCD within 120 days of June 24, 2026, identifying cryptographic-inventory methods and automated tools. National security systems are excluded.
Two workloads. Two representations. One doctrine.
Smooth, coupled, high-dimensional physics. The structure measurement says fold: fidelity-governed QTT, with dense and sparse retained where they measure better.
Pooled final-output embeddings where folding does not pay. The same measurement says do not force it: structure-governed latent-factor plus SQ8 instead.
The pair is the point. The substrate measures the structure and chooses the representation; it does not force one answer onto every workload. Their difference is the doctrine, demonstrated.
Federal is a mission environment. Not a sector.
Every workload above operates inside it: design campaigns, AI infrastructure, records, and the cryptographic transition. What changes is the boundary, the procurement path, and the authorization state.