What does a first sample review look like for a bank feed and ledger pilot?

An agent team building a finance-operations assistant asks MirrorFirm for a pilot. Their target is a payment that appears in the bank feed but not in the ledger. They use Aster Financial, MirrorFirm's sample organisation, as the setting.

The team's lead engineer writes a one-page brief: bank feed and ledger as the two systems, the accounts payable officer and the finance controller as roles, three tools to read, two to write, and a rule that adjustments above a stated limit go to the controller.

The first sample contains a handful of cases, each with a different cause: a duplicated feed line `BANKTXN-5521`, a payment posted to the wrong supplier `SUP-204`, and a timing difference that clears itself the next day. One case is a matched pair, and one should be escalated.

The team runs its agent, reads the traces and finds it treats the timing difference as an error and raises an adjustment. That is a failure mode they had not planned for. They ask for more timing cases, with different lags, and a policy clause on waiting periods.

The result is a second sample that is narrower and sharper than the first brief.

**What it teaches:** use the sample to find which failure modes matter to you, then scope the pilot around them instead of asking for breadth up front.