Tablekeeper — Built by a BAND Agent Factory

**Tablekeeper — Built by a BAND Agent Factory** Tablekeeper is a restaurant reservation system built through BAND's multi-agent AI factory. Our goal was to demonstrate reliable software engineering through specialized agents, independent QA, failure recovery, and verifiable results. **Our Five-Agent Factory** We configured five specialized roles: - **Architect:** Plans, delegates, and manages acceptance. - **Builder:** Implements features and prepares artifacts. - **Red Team QA:** Challenges code and identifies defects. - **Verifier:** Independently validates approved artifacts. - **Commander:** Coordinates the workflow. The factory follows planning, implementation, adversarial testing, repair, and verification. Human interventions are documented rather than presented as complete autonomy. **The Application** Tablekeeper handles restaurant availability, bookings, cancellations, competing requests, recurring reservations, time zones, and planning operations. Four development stages progressively introduced core reservations, browser workflows, concurrency, recurring bookings, and advanced planning. **Real Failures and Recovery** Independent QA exposed problems including hidden availability grids, stale booking context, and data consistency issues. These were addressed through documented repairs, regression tests, and independent verification. **Verified Results** All four shipped official stage suites passed local tests: - Stage 1: 120/120 - Stage 2: 25/25 - Stage 3: 7/7 - Stage 4: 6/6 Additional browser, concurrency, and isolated execution checks passed. Private judging tests are not included. **Reproducibility** Built with BAND, Codex, Python, Docker, and browser technologies, the public repository includes four runnable stages, reusable agent mandates, factory documentation, and genuine BAND session evidence. Tablekeeper demonstrates collaborative AI engineering with clear responsibilities, transparent failures, and independently verified outcomes.