Send one export — sales, invoices, stock movements, whatever you already have — and get back what it says.

You get three things: the findings written in plain English, the working Python that produced them, and an account of what the data would not support. That last part is the one that matters. Anything ambiguous is flagged and left alone rather than guessed at, so a correction can never be mistaken for a fact.

Fixed price, fixed scope. You know the total before it starts, and the job finishes — there is no retainer and nothing runs on afterwards.

Four worked examples are on my profile, each with its data and its code, so every figure in them can be reproduced from scratch.