Ledger writes invisible after rollback

A blue-green release moves the ledger service to green, which writes to a NEW table layout (ledger_v2); blue writes the old layout (ledger_v1). To allow rollback, the team cuts over to green at 12:00 but keeps blue's pods on standby. At 12:25 a bug is found in green; they roll back to blue at 12:28. Now reconciliation alerts fire: ~25 minutes of transactions written by green to ledger_v2 are invisible to blue (which only reads/writes ledger_v1), so balances are wrong and some transactions appear lost. Dashboards: green wrote ~12k rows to ledger_v2 during the window; ledger_v1 has a 25-minute gap; no app errors. How do you triage, mitigate, and what was the design flaw?