Published and updated 7 September 2026 · Altari Systems

Durable-state engineering

PostgreSQL Migration Lineage for Autonomous AI Agents

Abstract

Migration lineage is the ordered, verifiable history of database schema changes. In database-backed AI agent orchestration, that lineage protects the authority store itself: every scheduler and worker must understand the schema version it is using, and incompatible or divergent histories must fail closed rather than guess.

01

The database is part of the control plane

When project ownership, leases and evidence live in PostgreSQL, schema correctness affects engineering authority. A missing constraint can become duplicate ownership; an incompatible enum can create an impossible scheduler state.

Migrations therefore need stable identifiers, ordering and an applied-history record.

  • Immutable migration identity
  • Recorded application order
  • Transactional change where possible
  • Compatibility contract
02

Divergent lineage is a safety event

Two deployments can share a schema shape while reaching it through different histories. Future migrations may then make unsafe assumptions.

A fail-closed startup check should reject unknown, missing or reordered migration identities and require explicit remediation.

  • Unknown applied migration
  • Modified historical migration
  • Missing predecessor
  • Binary-schema incompatibility
03

Nexus implementation context

Nexus relies on durable PostgreSQL-backed project state. Its migration discipline exists to preserve scheduler, fencing and evidence invariants across software upgrades.

The database is not merely a reporting store; it is an authority boundary whose evolution must remain auditable.

Nexus launch access

Be first to know when Nexus launches.

Register your interest now and we will let you know when the autonomous AI software engineering control plane is ready.

Express Interest ↗