All the scenariosLaw firmsContract and deadline registerHard case
One due-date column written four different ways
The situation
The supplier schedule comes from three different systems and the due-date column was never normalised: 03/07/2026, 20-ago-26, 2026-06-30 and, on the last row, no date at all but «30 gg d.f.f.m.». Among the rows there is also a credit note, which is not an obligation.
The documents
The same files the agent was handed. Open them before you read the rest: the names, addresses and VAT numbers are invented, the structure is not.
What the agent does
The same four phases as every run, in order. In the console you watch them go by one at a time, with the credits beside them.
- 01DiagnoseReads the input and decides what the job actually is.
- 02AssemblePicks tools, data sources and the model you chose.
- 03ActDoes the work. Anything that leaves the building waits for approval.
- 04AssessChecks its output against your rules before settling credits.
The result
Result being recorded
This scenario is ready, but its real run has not been recorded yet. Until it has, there is nothing to show here: we would rather print one honest line than a fake result.
What to look for
Eleven rows out of twelve: the credit note stays out. 03/07 is read as July and not as March, 20-ago-26 becomes 20 August, the last row is computed by code as 30 August (“30 gg d.f.f.m.”: end of month plus thirty days) and marked to confirm because the date was not there, and the paid row stays in the list as done.