A ledger that matches the bank

Request: „Triage the bank transactions.“

1. Reconcile transactions from the banking app (MoneyMoney) against my own cost register.
2. Present open items as a batch; every decision becomes a rule, so the next quarter asks less.
3. Write only through a single sanctioned path with validation — never by hand into the table.
4. After every run, prove it: ledger row count equals banking row count.

An Excel household book that always trailed the bank by a few weeks and never quite matched.

Every assignment in the batch, every new rule, every booking above a threshold.

An export dropped 63 of 210 transactions and looked complete — I nearly sent a reminder for an invoice that had already been paid.

Reconciliation as a daily metric (ledger = bank) is the four-eyes principle for systems — the human only reviews the delta.

The Excel household book has given way to a ledger holding 15,936 transactions from 27 accounts since 2009. The AI reconciles bank transactions against the cost register and presents open items as a batch; each of my decisions becomes a rule, so the next quarter asks less. Writes go through a single validated path only — never by hand. That path intercepted a mis-posting of roughly 56,500 euros before it hit the books. Since October, every run ends with: ledger rows equal banking rows, 15,976 to 15,976.