Clean a delinquent list without touching a cell
Lesson 1 of 6 · Build the list, then scrub it
1 · Learn the move · Constraints + negative instructions
Ask a model to 'clean up' a pulled list and it will do what cleaning sounds like: fix the spelling of names, standardise addresses, fill a blank owner from the row above, merge two rows it decides are the same person. Every one of those edits breaks the thread back to the county export, and a cell you cannot trace is a letter you cannot defend. The move is to forbid editing outright and ask for new columns beside the untouched rows: DUP_OF, CLUSTER, ENTITY, ESTATE_CHECK, ADDRESS_CHECK. Clustering is where the money is - one owner scattered across six rows gets one letter about six parcels instead of six postcards. And every flag is written as a hypothesis with the text that raised it, because 'Estate of' in a name field is a reason to check the probate index, not a fact about who can sign.
Do not change any cell in the rows below. Not spelling, not addresses, not blanks. Add columns to the right only: DUP_OF, CLUSTER, ENTITY, ESTATE_CHECK, ADDRESS_CHECK. Each flag quotes the text that raised it. Then give me a one-letter-per-owner summary and a verify-first list. [paste rows]
2 · Your turn. You write the prompt
Your VA pulled 900 rows from the Tollard County delinquent file, and the mail merge for your blind offers runs Friday. In a sample you can see the same owner on several rows, a couple of LLCs and a trust, one name field that reads 'c/o Estate of', and a row with no owner name. Write the prompt that tags and clusters it without altering a single value from the export.
Write the prompt for this scenario as given, or swap in a real one from your own work. The score works the same either way.
Remember: the AI sees only your prompt, not this page. If the situation isn't in your prompt, it doesn't exist.
Optional. These shape the output when you run your prompt below, not your score.