Skip to content
DocketDocket
← Your pathLevel 1 · Beginner

Extract every authority from a draft and check each one

Lesson 2 of 6 · Verify before you file

1 · Learn the move · Fact-check list

The check is mechanical and it has to be, because the eye reads past a familiar-looking cite. Pull every authority out of the draft into a list, one per line, before you read the argument. A case, a statute, a rule, a regulation, a secondary source, a quotation attributed to anyone. For each, write where it opens: the reporter and the court's own site for a case, the official code for a statute, the rulebook for a rule. Then open each one and record three things: does it exist, does it say what the draft says, is the pin to the right page. Those are three separate questions and the draft can fail any of them alone. The model is good at the extraction and must do none of the checking; it cannot open anything, and if you let it mark a cite as confirmed it will. The state of every line after extraction is UNVERIFIED. The states after you open it are EXISTS, DOES NOT RESOLVE, or EXISTS, HOLDING UNCHECKED when you found the case but have not yet read the passage. No fourth state. 'The model double-checked it' is not a state.

Here is a draft: [paste]. Extract every authority in it into a table, one row each: the citation as written | type (case / statute / rule / regulation / secondary / quotation) | the proposition the draft uses it for | where I would open it | status. Status is UNVERIFIED on every row; you cannot open anything and you never write 'confirmed', 'correct' or 'verified'. Any quotation gets a second row flagged QUOTATION UNSEEN. End with the counts: total rows, quotations, and rows where the citation form looks incomplete.

2 · Your turn. You write the prompt

The partner at Marchetti & Ostrow hands the associate the forty-minute opposition and a rule: before anyone reads the argument, every authority goes in a table and every row gets opened. The associate has ninety minutes and the table is the first thing on the list.

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.