New reconciliation batch

What happens when you run

1. 1

   Normalise

   three column formats → one transaction schema; raw rows kept verbatim

2. 2

   Exact match

   join on UTR, verify amounts + fee — 85–95% resolved with no AI

3. 3

   Signals

   deterministic rules classify every leftover into the taxonomy

4. 4

   Adjudicate

   the AI confirms each call, writes the rationale, or flags for a human

5. 5

   Route

   high-confidence + small ₹ auto-resolves; the rest goes to your queue