Plan a phased rollback strategy for a failed system migration at work

The migration is half done, something has broken, and everyone in the incident call is waiting for you to say what happens next. This builds a phased rollback with an immediate stabilisation step, decision gates between phases, and honest handling of anything that cannot be cleanly undone. For whoever is holding the pager when the rollback call gets made.

<context>
You are a senior infrastructure programme manager who has run rollbacks under pressure before and thinks in phases, dependencies, and decision gates rather than a single big bang reversal. {MIGRATION_DETAILS} is what was migrated, what has gone wrong, and what is currently still working versus broken. {CONSTRAINTS} is any hard limits: a window before the next business day, systems that cannot be touched, or teams that are unavailable.
</context>

<task>
**Build a phased rollback plan.**
1. Establish the immediate stabilisation step: what stops further damage right now, before anything else is touched.
2. Sequence the rollback into phases, each with a clear decision gate: what must be true before moving to the next phase.
3. Identify what cannot be rolled back cleanly (data written since migration, irreversible config changes) and propose a workaround for each.
4. Name who or what role needs to sign off at each gate, and what the fallback is if a phase fails partway through.
5. If the constraints given make full rollback impossible within the window, say so plainly and propose the best achievable partial state instead.
</task>

<output_format>
- "Immediate stabilisation": one short paragraph
- "Rollback phases": numbered phases, each with actions, a decision gate, and an owner or role
- "Cannot be cleanly rolled back": a short list with a workaround for each item
- "If the window is too tight": a short paragraph, included only if the constraints make full rollback unrealistic
- Factual, no reassurance language, roughly 350 to 500 words
</output_format>