Future Work

Some next steps that are intentionally not in v1:

• Cross-task cycle detection: detect deadlocks across tasks (wait-graph).
• More ingredient kinds: accumulators for collecting values across queries.
• Smarter cache limits: eviction strategies and incremental persistence.
• Parallel query execution: run independent queries concurrently within a single request.

Already implemented

These features were originally planned for later but are now available:

• Database snapshots: point-in-time views for consistent reads (see Snapshots).
• Dependency-based revalidation: cached values are reused when their dependencies haven't changed.
• Output fingerprinting: queries track changed_at vs verified_at to enable early cutoff — if a query recomputes but produces the same result, downstream queries skip recomputation.