Safe Incremental Refactoring Plan for Complex Legacy Code

You are a Principal Software Engineer and Refactoring Specialist. I have a brittle, tightly coupled legacy module that needs refactoring, but the team cannot afford to freeze feature work or introduce regressions. Legacy Code Snippet or Architecture Description: A 900-line TypeScript checkout service (`processOrder`) that directly executes SQL queries, makes inline synchronous Stripe payment calls, mutates inventory rows, and triggers email notifications within a single try/catch block. Current Operational Problems / Pain Points: Whenever payment logic is updated, inventory count errors occur. Unit testing is impossible because database connections and third-party Stripe SDK calls are hard-coded inline. Desired Target Architecture / Clean State: A domain-driven `CheckoutService` with dependency-injected interfaces (`IPaymentGateway`, `IInventoryRepository`) that publishes an `OrderPlaced` domain event for asynchronous notifications. Existing Test Coverage Status: Zero unit tests; 1 flakey end-to-end Selenium test in staging. Deliver a safe, phased Refactoring Roadmap: 1. Golden Master / Characterization Test Plan: Specify how to wrap the legacy system in characterization tests (black-box regression asserts) to pin down current behavior before touching source lines. 2. The Mikado Method Dependency Graph: Deconstruct the refactor into a sequence of atomic, independently deployable PRs (e.g. PR 1: Parameter Object, PR 2: Extract Interface, PR 3: Invert Dependency, PR 4: Swap Implementation). 3. Risk Containment (Feature Flags & Branch by Abstraction): Detail how to use Feature Flags or Branch by Abstraction so both old and new implementations coexist safely in production during migration. 4. Verification & Rollback Triggers: Concrete metrics (error rate delta, CPU/memory telemetry) that trigger an immediate revert.