- **Partitioned CSV Event Storage**: Collected events are buffered in memory and flushed in batches to `/storage/data/{clientId}/{YYYY-MM-DD}.csv`.
- **Zero Per-Row Database Inserts**: Event telemetry streams are never inserted per-row into PostgreSQL.
- **File Rotation & Pruning**: CSV files rotate daily or when reaching the configured size threshold (`fileRotationMb`, default `100MB`, creating `-part2.csv`), and expired partitions are pruned according to `autoPruneDays`.
- **PostgreSQL Metadata Store (`pg`)**: PostgreSQL is used strictly for administrator accounts (`admin_users`), login attempt tracking (`admin_login_attempts`), registered SDK client credentials (`client_sdks`), and platform configuration (`admin_settings`).
- CRITICAL RULE: This is a multi-IDE project. Whenever any modification, update, or bug fix is completed in our session, automatically update `MEMORY.md` (or provide exact markdown blocks) so project context remains synchronized across VS Code, Google Antigravity, Cursor, and other IDEs (see `.ai-rules`).