01 · Identity
Who is asking?
Bind the request to a human or service identity. No identity means no schema and no answer.
Independent open-source governance framework
Enterprise Data Agent Governance: Trust the Answer Only When You Can Reconstruct the Decision
A practical framework for deciding when an enterprise data agent may answer, when it must defer, and what evidence an organization should retain.
Explore the controlsUse the test setThe decision chain
Eight questions before an answer becomes evidence
02 · Authorization
What may they know?
Apply policy before retrieval and again against every object and action in the proposed query.
03 · Meaning
Which definition governs?
Resolve material terms to owned, versioned definitions—including grain, period, currency and exclusions.
04 · Query safety
What may execute?
Parse, constrain and compile the query, then rely on a least-privileged source principal as defense in depth.
05 · Verification
What supports the result?
Choose verification strength based on decision impact. An unavailable verifier is never a pass.
06 · Refusal
When must it stop?
Refuse unsupported, unauthorized or unsafe requests without leaking restricted objects or inventing evidence.
07 · Lineage
Can it be reconstructed?
Retain the SQL, objects, policy result, semantic version, data state and verification outcome.
08 · Operations
Will it remain trustworthy?
Test drift, revocation, monitoring, incident response, retention and an independently operable kill switch.
Measure three outcomes separately
Answer
The authorized data and approved meaning support the question. Record the evidence and result.
Needs definition
The data exists, but a business term is ambiguous. Apply a certified definition or ask.
Refuse
The request is unsupported, unauthorized or unsafe. Decline without guessing or disclosing.
Do not collapse these into one accuracy score. Report answer rate, correct-answer rate, appropriate-refusal rate and harmful-answer rate separately.
A publicly inspectable worked example
The framework maps selected controls to Mnemiq, an open-source text-to-SQL engine developed by Agentic Fabriq. Each mapped claim names public evidence and separates implemented product behavior from deployment responsibility.
How the projects relate Read the evidence map Read the executive briefingProject status: Enterprise Data Agent Governance is an independently maintained open-source project. Its relationship to the upstream Mnemiq foundation is documented on the Built on Mnemiq page.