ADR-0004: Full Datalog + Pull executing on the peer

Status: Accepted (2026-07-15)

Context

Query capability could be phased (index API first, Datalog later) or delivered whole. Query could execute on a server (client-server model) or on the peer against local immutable data (Datomic peer model). The peer model is already fixed by ADR-0001; the open question was v1 query scope.

Decision

The first query milestone (M3) delivers the full EDN Datalog dialect — patterns, predicates, native function clauses, not/or, recursive rules, aggregates, multiple database args — plus the complete Pull grammar, the entity API, and direct index access. Execution is always local to a peer (thin clients get a server-hosted peer, ADR-0006/protocol.md).

Consequences

• The query engine is a first-class deliverable with its own planner, statistics, and conformance corpus — the second-largest engineering item.
• No interim "index API only" public period; users meet the real API first.
• Horizontal read scaling and query isolation come free (queries consume application-process resources, not database-server resources).
• User-defined cljrs functions in query clauses are the one deferral (post-v1); the native built-in set covers common usage.