Fresh-data testing

Build-along · ~10 min at your own keyboard

A model asked to test its own work generates test cases that flatter its own assumptions. The fix: test with data it has never seen, like a real closed listing from last year or a comp set you know is weird. Fresh data finds the failures polite tests never will.

Your starter: Index diff against last volume

Prompt

Make it yours: Point it at your real index export and the last volume's file, and set the one marker your matters use for privileged material so the stop rule fires on your data, not on an example.

Compare the new production index at [path] against the last volume at [path].

Report, in this order: control numbers new since last volume (number and type as the index has them); documents whose date changed; any line whose author or recipient role is counsel.

If any line carries a name, a caption or a docket number, say so and stop on that line. The parties stay lettered in anything that goes to the next step.

Do not describe a document, decide privilege, or propose a log entry. Where a field is missing write [CHECK THE RECORD: the source document]. Any authority a line mentions is not a citation until you have opened it.
Save the report as index-diff-[date].

The coach checks that what you paste back is your own work. The starter pasted back unchanged won't pass.

Do this

0/4

Share what happened

Paste what fresh inputs you used, what broke, and the category of failure.