What This Stage Covers
Sunset trigger met. Users notified. Workflows migrated to successor or pre-AI baseline. Data egress in portable format. Evidence Register archived (relevance persists post-retirement for audit). Risk Register entries closed. Learnings captured for the continuous-learning loop. Resources reallocated. Final ROAI report against the Concept-stage hypothesis.
Exit Gate — Required Evidence
G5
All Register entries closed; data egress confirmed; Evidence Register archived (not deleted; relevance persists post-retirement); post-mortem captured; final ROAI report comparing realised to Concept-stage hypothesis.
Editorial Framing
Sunset is the hardest stage to execute well because it requires acknowledging that a public investment is no longer earning its place. The discipline is the trigger record — five canonical triggers, written down at Deploy (Gate G3), applied honestly at the quarterly review. The post-mortem closes the continuous-learning loop.
Q3 Institutional return is most visibly produced at Sunset — the function that retires capabilities deliberately builds the institutional discipline that compounds across years. The function that retains capabilities past their useful life builds the opposite.
Required Artefacts
- Sunset trigger record
- User notification record
- Data migration plan + completion record
- Model decommissioning record
- Archived Evidence Register (not deleted)
- Post-mortem
- Final ROAI report (realised vs Concept-stage hypothesis)
Common Failure Modes
- Capability retained past its useful life — sunset triggers not applied honestly
- Data egress incomplete — workflows migrated but data orphaned
- Evidence Register deleted instead of archived (audit relevance persists)
- Post-mortem skipped — learnings not fed back into the framework
- Final ROAI report not produced — Concept-stage hypothesis never validated