Skip to content
dent8dent8

0001: Postgres From Start

Date: 2026-06-26

Status

Section titled “Status”

Accepted. Updated 2026-06-30: the Postgres-first decision held — Postgres was the first adapter. An embedded SQLite backend was later added as a second adapter behind the AsyncEventStore boundary (see storage.md); this ADR records the original first-backend choice, not a standing exclusion of SQLite. “SQLite is not part of the initial runtime plan” below should be read in that historical context.

Context

Section titled “Context”

dent8 needs an append-only event log, projection updates, contradiction checks, supersession lineage, replay, and auditability. The original note considered SQLite for a local-first MVP and Postgres later.

Decision

Section titled “Decision”

Use Postgres as the operational source of truth from the start.

SQLite is not part of the initial runtime plan.

Consequences

Section titled “Consequences”

Positive:

  • Runtime semantics match the intended production architecture earlier.
  • Append and projection updates can be transactional from the first adapter.
  • Multi-user and audit-heavy workflows do not require a later storage rethink.
  • Postgres constraints can enforce important invariants near the data.

Negative:

  • Local setup is slightly heavier.
  • Integration tests need a disposable Postgres instance.
  • SQLx compile-time checking may require .sqlx metadata or a build-time database.

Follow-Up

Section titled “Follow-Up”
  • Implement dent8-store-postgres.
  • Decide whether to use SQLx checked queries immediately or begin with dynamic queries.
  • Add Postgres integration tests for constraints and concurrent writes.