Merron · Accuracy and scope
Understand the basis of your report.
Merron organises the facts you confirm, links findings to documents and screens supported AI-product uses. Its legal and factual accuracy has not yet been independently measured. We do not claim 99% accuracy or certified compliance.
When you receive a screening result
Your relevant answers must be complete and the use must fit the supported screen. It currently covers basic scope questions and common transparency routes for providers and deployers, including distinctions between private text generation, public-interest publication and deepfakes.
Potentially prohibited uses, sensitive or regulated-product uses, general-purpose model providers, other operator roles and asserted legal exceptions receive an unresolved result with the reasons and next questions. A completed questionnaire does not guarantee a risk tier. An unresolved result does not establish low risk or permission to continue an activity.
How evidence is checked
Document quotations must exist in the supplied text. A separate AI check compares draft findings with the evidence, including whether a control is planned or implemented and whether statements conflict. A draft that fails the check is withheld; the failed generation is returned to your account allowance. Conflicting facts prevent a final classification.
Both AI passes can make mistakes. A document is evidence of what its author says; it does not prove that a control works. Merron does not inspect your live product or establish the authenticity or completeness of your documents.
What we are measuring
We are developing tests for incorrect classifications, missed obligations, unnecessary obligations, unsupported claims and outdated dates. Development tests are not independent validation. We will measure the share of cases that receive a complete result separately from correctness, so unresolved cases cannot inflate an accuracy claim.
Reports preserve the facts and rules used at generation. Earlier versions remain historical records. Update your case when its facts change and check current requirements before acting on a dated report.
Project checks through MCP
The MCP connection inspects a limited selection of text supplied by your assistant. Pattern matches identify possible integrations and questions; they do not establish deployment or legal applicability. Its screening tool uses the same bounded rules as the workspace, conditional on user-confirmed answers. It does not run the document evidence checker described above or synchronise private cases.
Read the connection’s scope and data handling