01
Causal + policy graphs
Author domain models in a typed DSL, compose them with operational telemetry, and run them as first-class artifacts under version control.
Decision Intelligence
Reasoning, governed.
Decision Intelligence converts live operational data into signed, replayable decisions. It combines causal models, policy graphs, and constraint solvers — with human checkpoints wired in where the regulator, the operator, or the contract requires them.
▍ The product
What an operator sees.
Decision Inspector — the workstation an analyst opens when a decision lands. The graph, the inputs that produced it, the operator who approved it, and the controls to replay or override. No part of this view leaves the customer perimeter.
Decision Inspector·DEC-2026-04-29-1041
ApprovedUnclassified // FOUO
Decision graph
Bed re-allocation · Ward 7
23 nodes·31 edges
Illustrative interface — values are design fixtures, not benchmarks
▍ The problem
Most ‘AI decision’ systems are black boxes wrapped in confidence intervals.
In critical environments, that is not acceptable. A hospital cannot deploy a triage recommender it cannot defend in a malpractice review. A utility cannot accept a load-shed decision it cannot reproduce after an incident. Decision Intelligence is built on the inverse premise: every decision is a graph, every graph is signed, every signed graph is replayable on the data that produced it.
▍ Capabilities
What ships in the box.
02
Constraint solver runtime
Pluggable solvers (MILP, SAT, CP-SAT) with deterministic seeds and reproducible execution traces.
03
Human checkpoints
Approval gates with role-aware routing, override capture, and structured annotation — all wired into the decision graph.
04
Signed decision artifacts
Every decision exits the system as a cryptographically signed graph with input snapshots, model versions, and operator chain.
05
Replay & counterfactual
Re-execute any prior decision against today’s data, today’s models, or hypothetical inputs — for audit, training, or post-incident review.
06
Operator console
A workstation interface for the analyst on the receiving end — purpose-built for triage, override, and escalation, not for tinkering.
▍ Use in sector
Concrete deployments.
Reference scenarios — drawn from active design-partner conversations and prior operator engagements.
- DEFENSECourse-of-action ranking for a brigade staff: terrain, ROE, force composition, and intelligence priors fused into a defensible recommendation — with the constraint graph attached.
- ENERGYSub-second load-shed decisions during grid contingencies, with the policy precedence graph and operator override channel signed into the audit log.
- HEALTHCAREBed-allocation and clinical pathway routing across a hospital network, with each decision carrying its causal model version and the clinician who approved it.
- FINANCIALPre-trade compliance and risk decisions, replayable on demand for regulator queries — without re-running the original market data.
Attached to every decision
Provenance
Any prior run, deterministically
Replay
No external egress required
Air-gappable
Decision artifacts as evidence
Signed