The ConvergeQL data model flow

Make the modeling process executable.

ConvergeQL treats model change as a governed lifecycle, not a chat transcript or a single SQL generation step.

Candidate C*Every required validator returns passAgainst one recorded context and policy snapshot

01
Discover
Resolve the smallest useful repository, warehouse, and policy context.
02
Specify
Make grain, contracts, constraints, and protected outputs explicit.
03
Model
Propose a bounded change against one platform and environment.
04
Validate
Run declared schema, data, cost, lineage, and governance checks.
05
Converge
Reject, diagnose, and revise candidates until required gates pass.
06
Review
Present the model decision, diff, evidence, and unresolved limits.
07
Ship
Create a human-approved GitHub change. No autonomous merge or deploy.
08
Observe
Use later production evidence as new context, not silent authority.

Operating doctrine

Authority is earned through evidence.

The agent is not trusted because it is intelligent. Material actions are bounded, observed, tested, and reviewable.

Reasoning explores candidate designs, selects bounded actions, and interprets validation failures.

Versioned policy, authorized tools, and deterministic validators decide what can proceed.

A reviewer sees the specification, diff, evidence, limitations, and rejected candidates before Git delivery.

Two flows, one run

Model change

Intent - Context - Plan - Candidate - Validation - Review - Git

Evidence

Identity - Inputs - Decisions - Tool results - Policies - Tests - Approval

The second flow is what turns AI-assisted coding into governed engineering.

Design partner preview

Bring one existing dbt-managed model, its declared invariants, and a review problem worth solving.