Module 1 recap

You started with a statement you had probably never read line by line. You are ending with a month in plain categories, one number named, and a way of working that does not hand your account details to anyone.

The order mattered. Privacy came first: Incognito, a deliberate decision at Settings > Privacy, and a redaction habit that removes what identifies the account and keeps what describes the money. Then the wall almost nobody clears alone: an upload needs an unencrypted PDF, and a locked statement opens with your bank's own convention and prints to a clean copy. Then the win: a fixed category list, transfers separated from spending, every total showing its lines. Then one number, stated plainly, with the moralising banned in the prompt itself.

That last habit is the one to carry. Every prompt here tells Claude what it may not do, and that is what makes the output something you can rely on rather than something that merely sounds good.

Where this goes next

Module 2, Make it true. The three ways an AI silently gets money wrong: screenshots too small to read properly, arithmetic done in its head instead of checked, and totals that look identical whether right or wrong. You get the one sentence to append to every money calculation, and a spot check that takes under a minute.