Temporal Queries and Point-in-Time State

Conventional current-state rows overwrite prior values. Logs may show requests but not accepted domain state, timestamps may use different clocks, and audit tables often omit reducer logic. Incident response then cannot reliably answer whether a decision was correct given information available at that moment.

Temporal correctness requires named coordinate. Stream version is precise within one aggregate. Global WAL position orders accepted writes across store. Event timestamp supports human time questions but equal or skewed clocks need tie-breaker. Effective business dates may differ from record time and should be explicit payload facts, not substituted for append order.

AllSource Core supports point-in-time entity reconstruction and can seed replay from nearest eligible snapshot before applying later events. Query ordered entity events through /api/v1/events/query and use reconstruction endpoints for state questions. Core keeps stored history durable through WAL and Parquet rather than depending on application logs.

Expose as-of coordinate in API response and interface. For a decision audit, store event that records selected policy, input references, and outcome; temporal query can then reconstruct both state and evidence. Prime uses same pattern at application layer to trace recalled facts back to source events and reconstruct what agent memory contained before later corrections.