Decisions

Decisions capture business logic as inputs, rules, and outputs that are versioned, tested, and governed independently of your application code. Because they run on a rules engine rather than a probabilistic model, decisions deliver completely predictable outcomes — identical inputs always produce identical results, and every outcome records exactly which rules fired and why. That determinism and traceability is what makes decisions auditable for regulated industries, where black-box models fall short. Aletyx runs decisions on its enterprise build of Drools and Kogito, with microsecond-level execution that can evaluate thousands of rules in parallel — suitable for high-volume transaction processing and real-time use cases like fraud detection.

​

Authoring approaches

You can express decision logic in three complementary ways. Most teams mix them: DMN for transparent, business-owned logic and DRL where they need more power.

​

DMN (Decision Model and Notation)

DMN is an OMG industry standard that separates decision logic from process flow. Aletyx supports DMN at conformance level 3 — the highest standard — so models remain portable and vendor-neutral while still executing precisely. See the DMN reference.

​

Decision Requirements Diagrams (DRDs)

A DRD is the visual map of a decision: input data, decision nodes, and the information requirements (solid arrows) that show how each decision draws on inputs and other decisions.

​

Decision tables and hit policies

Decision tables lay out conditions and outputs in rows so business analysts can author and review rules directly. A hit policy controls how multiple matching rows resolve. Aletyx supports all ten DMN hit policies:

​

FEEL and boxed expressions

FEEL (Friendly Enough Expression Language) is the DMN standard’s expression language — strongly typed, side-effect free, business-friendly, and rich in built-in functions. Beyond decision tables, DMN offers other boxed expressions: literal expressions, contexts (name-value pairs), relations (tabular data), functions (reusable logic), invocations (calls to business knowledge models), and lists. Aletyx also supports custom, domain-specific data types alongside numbers, strings, dates, and complex structures.

​

DRL (Drools Rule Language)

DRL is the native rule language for logic that goes beyond what decision tables express. Rules follow a declarative when/then pattern and run on the Drools inference engine. See the DRL reference.

• Conditions (when / LHS) — pattern matching that evaluates facts against business criteria.
• Actions (then / RHS) — operations run when conditions match, such as modifying working memory or calling functions.
• Facts & working memory — facts are the data rules evaluate; the engine tracks them in working memory and re-evaluates as they change.
• Phreak engine — Drools’ high-performance pattern-matching algorithm fires rules efficiently as facts change.
• Rule units & queries — organize rules into units and query working memory for matching data.

DMN vs. DRL: reach for DMN when business users own declarative, table-style logic that must stay transparent and portable; reach for DRL for complex procedural inference, chaining, and fine-grained control. See DMN vs. DRL vs. BPMN.

​

Testing

Validate decisions before they reach production:

• Real-time testing in the Playground as you edit a model.
• Natural-language testing — business users validate logic in plain English (for example, “What happens if income is $85,000 and credit score is 720?”) and get immediate feedback with no technical knowledge required.
• Test scenarios that assert expected outputs for representative inputs and guard against regressions.

​

Governance with Decision Control

Decision Control manages decisions as governed, testable assets across their full lifecycle:

• Multi-environment promotion — isolated Development → Test/UAT → Production environments with governance checkpoints and standard, fast-track, or self-service approval workflows.
• Four-eyes principle — the creator of a change cannot approve their own change; separation of duties is enforced.
• Complete audit trail — every change records who made it, when, what changed, and why (a required justification) — regulator-ready documentation and the foundation of Accountable AI.

​

Safe AI integration

Aletyx applies deterministic guardrails so LLM outputs can’t override business policy or produce unaudited decisions. You can also expose DMN models to LLMs automatically through the Model Context Protocol (MCP) — no custom integration code. See Decisions + GenAI/MCP and Connecting DMN decisions to LLMs with MCP.

​

Deployment & integration

Package decisions as cloud-native services and promote them across environments. Integration options include:

• REST APIs for millisecond decision execution.
• Event-driven invocation via Kafka.
• SSO through OAuth2/OIDC.
• Kubernetes-native deployment; builds distributed via Maven and container registries.

​

Next steps