Principles

SSOT is the top priority. Copies breed operational cost explosions. DB contradictions, circular refs, and ambiguous ownership are the most dangerous tech debt.

• Single Source of Truth for every domain concept
• Normalization eliminates contradictions, not for show
• Price, inventory, state—always ask “who is canonical?”
• Schema reviews before feature reviews

Idempotent operations, transactional consistency, and clear state management. If an operation can fail halfway, it will. Design for it.

• Idempotency by default on every write path
• Transactional boundaries explicitly designed
• Retry-safe, crash-safe, restart-safe
• State machines over boolean flags

Not an afterthought bolted on later. Enforced at the DB and policy level. Audit trails, IP restrictions, admin tooling—manageable security for real-world ops.

• Row-level security and policy-level enforcement
• Audit trails on every mutation
• Admin tooling with IP restrictions and approval flows
• Secrets management as a first-class concern

Dev Standards, PR/review routines, checklists, and decision records. Consistent systems always outperform individual heroics.

• Standards codified, not memorized
• PR templates and review checklists enforced
• ADRs for every non-trivial decision
• Onboarding measured in hours, not weeks

Not decoration. PRDs, playbooks, decision records, migration scenarios, and checklists. They recur because they work.

• Runbooks for every critical path
• Migration playbooks before every deploy
• Incident postmortems as learning infrastructure
• Living docs that evolve with the system

Documentation and structure must be machine-readable. AI doesn’t replace engineers—engineers who structure knowledge for AI access multiply their output.

• Machine-readable schemas and documentation
• AI-friendly code structure and naming
• Automated workflows that AI agents can extend
• Knowledge as a compounding competitive advantage