Verification de Faits
Verifies a claim against the workspace knowledge sources. Returns a verdict (verified/unverified) with a confidence score and sources.
Parameters
| Parameter | Type | Required | Variable | Description |
|---|---|---|---|---|
input | dynamic value | Yes | Yes | Claims or text to verify. The system automatically decomposes into individual assertions. |
sources.useKnowledgeGraph | boolean | No | No | Query the Knowledge Graph to verify claims (entity existence, relations, properties). (Default: true) |
sources.useWebSearch | boolean | No | No | Use web search to verify claims (requires a SerpAPI API key). (Default: false) |
sources.customSources | dynamic value | No | Yes | Custom verification sources (documents, reference texts). |
entityTypeFilter | array | No | No | Limit verification to specified entity types (e.g. Client, Project). |
confidenceThreshold | number | No | No | Minimum confidence threshold (0 to 1) to consider a claim as verified. (Default: 0.7, min 0, max 1) |
includeReasoning | boolean | No | No | Include detailed reasoning for each verified or rejected assertion. (Default: true) |
onUnverified | choice (flag, reject, warn) | No | No | Action on unverified claims: flag, reject, or warn. (Default: "flag") |
model.provider | text | No | No | Language model provider for claim analysis. |
model.model | text | No | No | Model identifier to use (e.g. gpt-4o). |
parameters.temperature | number | No | No | Model creativity. Low value recommended for verification. (Default: 0.2, min 0, max 2) |
parameters.maxTokens | number | No | No | Maximum number of tokens in the verification response. (Default: 4096, min 1, max 128000) |
outputVariable | text | No | No | Output variable name containing the verification results. |
violationsVariable | text | No | No | Variable name storing the unverified claims. |
Parameters marked Variable = Yes accept the
{{blockName.field}}syntax.
Output
Output variable : verificationResult
{
"verified": false,
"confidence": 0,
"sources": []
}
Example
Verify a business claim.
Input :
{"claim": "Le fournisseur Acme a un taux de livraison de 98%"}
Output :
{"verified": true, "confidence": 0.87, "sources": [{"id": "doc-1", "title": "Rapport fournisseurs 2025"}]}
Tip
{{verificationResult.verified}} is the boolean verdict. {{verificationResult.sources}} lists the documents used for verification.