Skip to main content

Reconciliation & Cutover

1. Reconciliation layers​

Validate at multiple levels:

Structural​

  • object counts
  • schema
  • keys
  • data types

Data​

  • row counts
  • control totals
  • min/max dates
  • distinct key counts
  • null counts

Semantic​

  • measures
  • units
  • currencies
  • hierarchies
  • filters
  • aggregation behavior

Business​

  • critical reports
  • KPI values
  • exception populations
  • user acceptance

2. Cutover models​

Choose:

  • big bang for small, isolated workloads
  • phased cutover for domain migrations
  • parallel run for high-risk financial workloads
  • consumer-by-consumer transition for shared semantic layers

3. Rollback​

Define objective rollback triggers before cutover:

  • reconciliation threshold exceeded
  • critical pipeline failure
  • unacceptable latency
  • security defect
  • consumer breakage
  • unresolved data-quality defect

Rollback must be technically executable and time-bounded.

4. Hypercare​

Monitor the migrated workload at increased frequency until stability evidence supports normal operations.