Current limitations

These boundaries are intentional and should remain visible in release notes.

1. One writer per physical `.asa` file. Flask does not turn AsaDB into a multi-writer engine.
2. Transaction restart behavior. Cross-request transactions work while the same backend process is alive. They are rolled back/marked aborted after a server restart.
3. One public process. Run one Flask/Waitress process for one state/data directory. Multiple WSGI processes are not supported yet.
4. Snapshot replication. Replication is asynchronous `.asb` backup/restore, not log shipping or synchronous consensus.
5. No automatic failover. Replica promotion is an operator action.
6. No distributed SQL. Cluster pool metadata does not shard or merge query plans.
7. Eventually consistent FTS. Rebuild after source-data changes.
8. File API locality. It assumes reliable local atomic rename semantics.
9. Panel transactions. The panel can execute SQL transactions, but remote transaction affinity should use `/api/v1/sessions`. 10. Large query cancellation. Flask can cancel queued orchestration jobs; cooperative cancellation of a running direct SELECT depends on future core hooks. Reservoir imports retain their existing cancellation path. 11. No claim of SiriDB protocol compatibility. Only architectural ideas such as explicit nodes/pools/replicas influenced the layout.