System architecture
The system deliberately gives each tool one primary job instead of asking one AI tool to do everything.
System layers
The handbook and Athena help people understand and retrieve context. GitHub and executed evidence remain authoritative for current product and release state.
Responsibility map
Authority order
When information conflicts, use this order:
- actual code, config, schema, runtime, and immutable Git state
- canonical product/specification artifacts
- repository engineering metadata
- executed deterministic test and CI evidence
- AI Work Packet current coordination state
- Athena / derived documentation
- conversation history
This order prevents a stale chat, roadmap, or wiki page from silently overriding the current product.
Two repositories, two purposes
datarelay-labs/engineering-system
Canonical and machine-consumed. It contains standards, adapters, templates, validation logic, /resume, and session-continuity rules.
datarelay-labs/engineering-system-docs
Human-readable. It explains why the system exists, how the pieces fit together, how to adopt it, and what problems were solved.
If this site disagrees with the canonical Engineering System repository, the canonical repository wins.
Project-local adoption
Base Engineering System adoption requires the small repository entrypoint used by agents and CI:
Session-continuity files such as .cursor/commands/resume.md and the AI Work Packet Issue template are added when that workflow is used.
The repository should not copy every Engineering System standard into itself. It should pin and reference the canonical system.