> For the complete documentation index, see [llms.txt](https://docs.rootcause.ai/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.rootcause.ai/key-innovations/addressing-domain-expertise.md).

# Addressing Domain Expertise

Most causal inference projects stall because they depend heavily on scarce human expertise. RootCause is designed to work in spite of limited expertise, while still letting experts shape and refine the model where it matters.

***

#### Definition & Purpose

Domain expertise is essential to make causal models meaningful, but most organizations can't encode all their knowledge upfront. RootCause addresses this gap by:

* Embedding domain-aware rules directly into the modeling process
* Allowing experts to refine results without acting as a bottleneck
* Keeping outputs explainable and auditable so both technical and business users can trust them

***

#### How It Works

1. Automatic Domain Constraints – Ontology anchors (entity, time, location) and temporal rules ensure only plausible relationships are tested.
2. Built-In Domain Semantics – Where statistical evidence alone cannot settle an edge's direction, an LLM stage proposes an orientation from variable names and real-world semantics, supplying a baseline of domain reasoning even when no expert is available. Each such decision is recorded on the edge itself.
3. Expert Input – Users can declare relationships as *Known* (must exist) or *Blocked* (forbidden). Dependencies can also be added or broken to reflect context.
4. Versioned Refinement – Expert edits are saved as a new version of the model, preserving history; retraining the model applies them, producing an updated causal model that integrates both data evidence and human knowledge.

***

#### Oversight & Flexibility

* Surfaces uncertain or ambiguous edges for review
* Supports relationship rules on ontology concepts — constraints an expert states directly, or accepts from statistical detection and schema structure
* Records per-edge direction provenance — whether an edge's orientation came from the ML classifier, the LLM stage, a user constraint, or was compelled by graph structure

***

#### Outcomes

* Resilient Modeling – Works effectively even when deep expertise is limited or unavailable
* Hybrid Knowledge – Blends statistical discovery with domain context to avoid "black box" results
* Trust & Adoption – Models are more likely to be accepted because they incorporate the reasoning of both machines and experts

***

#### Why It Matters

Causal inference without domain expertise risks producing results that are correct statistically but irrelevant operationally. RootCause bridges this gap, ensuring models are both data-driven and domain-aware, so they can be trusted to guide real decisions.