Primitives and shared objects.

Cryptographic primitives

Proof system

Target prover family: transparent hash-based; current harness: a classical simulator. For institutional archives that must remain checkable for a decade or more, the proof family is a day-one choice, not an afterthought.

Post-quantum cryptography

Signatures and key encapsulation follow the lattice-based schemes on the NIST standards track (FIPS 203/204 class); hash-based transparency covers the proof layer. Post-quantum is a design constraint fixed from day one, not a migration promised later.

Recursive proof aggregation

Continuous attestation and multi-policy products need proofs that compose: histories aggregated and verified without rebuilding the world each epoch. We are prototyping a recursive aggregation construction selected for exactly this; the construction, its literature, and benchmarks are shared in diligence materials.

Target shape: verifier work stays flat as history grows (analytical today; measured benchmarks are a later-stage target); applications compose without forcing every vertical onto one monolithic circuit.

Interoperability standards we are shaping

Drawn from continuous-attestation, compliance, and agentic-AI reproducibility work we are prototyping:

• Classified obligations state: liability and asset vectors with horizons, mobility, and entitlement maps; the object continuous PoR actually proves.
• Coverage predicate & ladder: executable classical evaluators and harnesses so "assets cover liabilities" means a ladder under registered policy, not a scalar photograph.
• Regime profiles & tolerances: versioned profiles for Swiss safeguarding, MiCAR, and ADGM-class client-asset rules, with breach taxonomies and tolerance hooks inside the attested object.
• Audience derivation: one proven state; supervisor, bank, and client each verify only their entitled predicate, with sequencing that respects legal notification order.
• Policy-as-code surface: TypeScript policies with custody labels as proof; formal specs of ERC-3643-class transfer rules support differential testing of existing gates.
• Assurance-criteria track: objects meant for open profiles and assurance partners, not locked inside a single vendor's PDF. Auditors are the first channel supervisors typically see; standards bodies receive reproducible profiles and test vectors after independent consumption is demonstrated.
• Policy frontend profiles (planned): a bounded decision-table profile suitable for verifiable execution, with explicit unsupported cases, fixtures, and a verification interface external implementers can reproduce.

AI compliance & reproducibility

Standards objects for the agentic pillar, measurement and vocabulary, not product architecture:

• Agentic Determinism Index (ADI) methodology: open harness spec for serving-stack reproducibility; byte-level and canonical-form divergence metrics; reproducible from published transcripts.
• Serving-tuple vocabulary: the six conditions under which same prompt ⇒ same bytes, request, weights hash, engine mode, sampler, software digest, hardware SKU.
• Regime mapping (traceability objects): EU AI Act Art. 12/19 logging and traceability obligations; financial-services algo rules (RTS 6 class); FDA "locked algorithm" definition, the regulatory surfaces these standards target.

Why standards, not just product

Privacy math without shared criteria does not deploy. Shared criteria without portable proofs do not travel across borders. We publish vocabulary, tolerances, and proof composition as open objects; the applications that consume them are the business.