Generation Engines

Five generation engines covering statistical synthesis, physics-based simulation, constrained generation, and industrial security. Engine selection is fully automatic — the platform routes every job to the correct engine based on the sealed job specification and user intent.

Engine Overview

Engine	Compute	Trigger	Use Case
Mock Engine	CPU	Description-only requests with no training data	Instant synthetic data from natural language descriptions
Synthesize (AI)	CPU/GPU	Training data uploaded or referenced	High-fidelity tabular synthesis via our trained generative model
Constrained Synthesis	CPU/GPU	Hard domain constraints specified in the sealed job spec	Constraint-safe sampling — violations impossible by construction
Virtual SCADA	CPU	SCADA/OT telemetry or plant simulation requested	Physics-honest industrial OT telemetry with six live protocols
ICS Security	CPU	Attack dataset or cyber-range scenario requested	MITRE ATT&CK ICS attack datasets with ground-truth labels

ℹAutomatic engine selection

You never choose an engine directly. The orchestrator analyses the sealed job specification, uploaded data, and user intent to select the correct engine automatically. All plans, including Free, have access to all engines. Credits are the only generation gate.

Mock Engine

Instant data generation from natural language descriptions alone. No training data required. The Mock Engine uses a deterministic, seed-reproducible row generator seeded from the sealed job spec, with correlation-preserving lift that enforces cross-column correlations, foreign key integrity, and nullable patterns.

✦When Mock Engine activates

The platform routes to Mock Engine when the user provides only a description (no uploaded dataset) and the request does not require trained model fidelity. Ideal for prototyping schemas, testing pipelines, and generating development fixtures.

Synthesize (trained AI engine)

The primary trained synthesis engine. Our generative model achieves industry-leading accuracy on public tabular benchmarks. It learns the joint distribution from uploaded training data — marginals, correlations, and tails — and generates statistically faithful synthetic records.

Path Selection

Path	Status	Description
Trained AI	Primary	Distribution-aware training with multi-stage refinement for faithful joint structure
Statistical fallback	Fallback	Classical statistical synthesis for datasets under roughly one hundred rows
Auto-Selector	Active	Routes between the trained and statistical paths based on dataset size and shape

Performance

Generates hundreds of thousands of rows in seconds on CPU, with fidelity measured per run and reported in the utility artefact of the evidence bundle.

Constrained Synthesis

Constraint-safe generation for tabular data that must obey hard domain rules. The platform automatically routes to this engine when the sealed job spec includes domain constraints that require guaranteed satisfaction.

Violations impossible by construction: the sampler respects constraints at every step
Constraint types: box bounds, rate limits, monotonicity, equality/inequality sum constraints
Purpose-built for financial, clinical, and actuarial data that must obey business rules
Cross-column physics relationships (e.g., pressure–temperature coupling) supported

⚠When Constrained Synthesis activates

Constrained Synthesis is selected when the sealed job spec contains explicit domain constraints (box bounds, monotonicity, sum constraints, rate limits). If no hard constraints are present, the platform uses the AI engine or Mock Engine instead.

Virtual SCADA Simulator

Physics-honest OT telemetry simulation with six live industrial protocol stacks. A layered composition architecture covers a deep library of plant templates across the full breadth of critical-infrastructure sectors.

See Virtual SCADA for full protocol and plant documentation.

ICS Security Simulator

Ground-truth labelled ICS attack datasets mapped to MITRE ATT&CK ICS techniques. Generates reconnaissance, credential access, lateral movement, privilege escalation, command and control, and impact traffic with full causality chains. Can wrap a live Virtual SCADA simulation for integrated attack scenarios.

See ICS Security for attack categories and technique mappings.

Supporting Components

Component	Role
AI Orchestrator	Translates natural language into a sealed job spec and routes to the correct engine
Agentic Data Scientist	Autonomous multi-step pipelines spanning any combination of engines

See AI Orchestrator and Agentic Data Scientist for full documentation.