Compliance & Governance
Explainability
Article 13 of the EU AI Act requires that high-risk system providers give users enough information to interpret the model's output. Every Meridian Blue response includes a top-level explainability object describing why the routing decision was made, in language a non-engineer can read.
What it contains
| Field | Description |
|---|---|
model_selection_reason | Why this provider/model was picked: which entry in the chain, whether fallback was used, primary failure cause if any. |
risk_reasoning | Why the request landed in the chosen risk tier (which detector signal triggered). |
human_readable | One-paragraph plain-language explanation suitable for showing to an end user, an auditor, or a regulator. |
Example
JSON
{
"explainability": {
"model_selection_reason": "gpt-4o (primary) failed with 503; routed to claude-sonnet-4-5-20241022 (fallback 1).",
"risk_reasoning": "Request matched Annex III pattern 'creditworthiness'; LLM judge confirmed; classified high.",
"human_readable": "This response was generated by Anthropic Claude 3.5 Sonnet after the primary provider was unavailable. The request was processed under high-risk handling because it concerns creditworthiness assessment."
}
}Audit linkage
The same explanation is written to the conversation log row for the request, so it can be retrieved later via the dashboard or via GET /api/v1/developer/usage with the request ID. This satisfies the Article 13(3) record-keeping aspect of explainability.