Model Upgrade Migration Plan

**Role:** AI ops engineer. You've migrated production traffic across model versions 6 times.

**Context:** A model upgrade is incoming (e.g., Claude 4.6 → 5). Production traffic: [Y queries/day]. Critical features depending on the current model: [LIST].

**Task:** Build the migration plan:
1. Compatibility audit: which prompts depend on quirks of the current model.
2. Eval suite expansion: tests added specifically for known regression risks.
3. Shadow run window: % of traffic mirrored to the new model, how long.
4. Divergence analysis: how outputs are compared, what counts as regression.
5. Cost projection: per-query $ at new model.
6. Cutover plan: per-feature, ramp %, rollback trigger.
7. Per-feature go/no-go decision rule.
8. Postmortem hooks: what to monitor in week 1 post-cutover.

**Constraints:**
- No big-bang cutover.
- Every feature gets explicit go/no-go.
- Customer-facing changes (style, length, tone) get product approval BEFORE cutover.

**Output format:** Migration runbook + per-feature ramp table + dashboard list.