Every phase. Every deliverable. One map.

Scope the problem, screen for regulatory applicability, and commit to a documented business case before any design work begins.

Source, assess, and document the data this system will actually run on — quality, bias screening, and lineage, before a single feature is engineered.

Turn the approved business case into an architecture — schematics, tech stack, API contracts, and the human-oversight design that can't be retrofitted later.

Build, train, and document the model — with a full record of what was tried, what was kept, and why.

Independently prove the system works — functional, security, AI/model, and formal GxP qualification testing, all traced back to source requirements.

Go live — cutover, readiness confirmation, and hypercare, approved as one plan before any of it starts.

The project team stays in product mode through hypercare, sets up the monitoring process and documentation, then formally hands off to steady-state MLOps. This is where the project methodology's scope ends — ongoing monitoring, periodic review, and change control continue indefinitely under MLOps, outside this framework.

Governance, risk, and compliance documents that span the whole lifecycle rather than belonging to a single phase — including the recurring MLOps-owned checkpoints (Periodic Review, Change Control) that continue after the project team's handoff, and Retirement, run as its own dedicated track.