Risk State Engine · evaluation runtime v0.4.0
A cleaner control layer for numerical risk state.
Sophisticated quant firms already use incremental systems, caches, snapshots, lineage and reconciliation processes, and periodic authoritative rebuilds. LuxiQuant Risk State Engine adds a narrow state-maintenance and control layer around the scenario-by-position matrices those systems already produce. Your models, market data, pricing logic, scenario generation, trading logic, and methodology stay yours.
Maintain, reconcile, and measure state
Luxi maintains numerical state as the matrix changes and reconciles incremental histories against authoritative fresh rebuilds. Rebuilds remain comparators and reconciliation mechanisms; Luxi does not claim to eliminate them.
State maintenance and reconciliation
Changed positions update hierarchical Neumaier totals, empirical VaR/ES, and tail state on the canonical CPU path. Existing customer systems remain the source of the matrix and authoritative rebuild.
Reproducible final state
Under the Luxi State Equivalence Contract, incremental updates must produce the same Luxi numerical state and authenticated result as a fresh rebuild of the same final matrix.
Measured update and authentication cost
Update shape, state-maintenance cost, and authentication cost can be measured separately. Optional CUDA authenticates changed two-dimensional Merkle tiles; it does not price or compute VaR/ES.
Supporting lineage evidence
A signed receipt can bind the State Equivalence Contract, hashes, and incremental lineage. It supports reconciliation and verification; it is not the product itself or a regulatory opinion.
Measured boundary
H100 tile authentication
On a 1 GB resident state, 10% scatter authentication was about 11.1× faster on an NVIDIA H100 NVL than CPU hashing, with matching roots. Tiny and column updates stayed on CPU.
Internal measurement. H100 NVL only. H200 and B200 are unmeasured for Risk Runtime.
Evaluate directly from OCaml
The evaluation package now includes an OCaml library over the existing C ABI, with handle lifecycle, Bigarray and v0.3 ingest, incremental cell and column updates, signed receipt access, and accelerator enablement through the same interface.
Who this is for
Quantitative infrastructure
Teams that already operate incremental aggregation, caches, snapshots, reconciliation, and authoritative rebuilds and want a cleaner state-control boundary.
Valuation and fund administration
Teams already producing scenario-by-position numerical states and reconciling changes over time.
Model risk and validation
Teams that need reproducible final state under a declared numerical contract without replacing existing pricing or revaluation systems.
Operational verification
Teams that need measurable update cost and independently checkable lineage as supporting evidence around state maintenance.
Two divisions, separate evidence
LuxiQuant
The primary commercial division. Risk State Engine is the institutional evaluation path. LuxiBook remains the public receipt and determinism demonstration.
LuxiEnergy — AI Energy Systems
The secondary division works to reduce waste across AI compute, memory movement, workload behavior, and power delivery.