Skip to main content
Trusted Analytics V1 adds an enforcement layer between the agent and your data.

​
Core idea

V1 closes the trust gap with three guarantees:
  1. Scope — only approved tables and joins
  2. Meaning — governed metric definitions
  3. Safety — policy-enforced execution (PII, limits, time filters)

​
V1 building blocks

​
Runtime flow

  1. User asks a question.
  2. Agent calls build_contract.
  3. Policy engine returns allow, block, or needs_clarification.
  4. If allowed, execution runs with contract_id.
  5. Contract is persisted for provenance.

​
Strict vs compatibility mode

policies/policy.yml controls runtime strictness:
  • false: backward-compatible behavior
  • true: strict contract gating (recommended for production)

​
Why V1 matters

  • Consistent revenue-style metrics across LLMs
  • Deterministic PII blocking
  • Query boundary control with dataset bundles
  • Auditable artifacts for compliance and debugging

Validation and testing

Run dazense validate, dazense eval, and dazense eval --engine