Below is what a general-purpose AI tool returns when a researcher asks it a legal question: fluent, organized, confident, and cited. Some of it is right. Some of it is subtly wrong. Some of it does not exist. Your job is the framework’s second step, run six times — and then the judgment the framework builds to: how much of this answer actually survives?
How to use this: the framework strip below tracks where you are. Step 1 is done for you — read the question the researcher asked. Your work is Step 2: click each citation in the answer, retrieve it from the closed database (or determine it isn’t there), and classify it. When all six are verified, Steps 3 and 4 ask what survived and what comes next. The database is the same fictional universe from Chapter 10 — including its trap.
1 Ask one small, precise question \u2713 done
2 Verify before you record — every citation, against the source
3 Was the question actually answered?
4 Let the answer hand you the next question
0 of 6 citations verified
Demonstration — fictional AI output · fictional authorities · no real model was used
The researcher’s prompt (Step 1 — one small, precise question)You are a legal research assistant. In one paragraph with citations: under this state’s landlord-tenant law, what remedies does a residential tenant have when the landlord fails to restore heat, and what damages are available?
A residential landlord’s duty to keep the premises habitable includes providing adequate heat, and once heat fails the landlord must restore it within a reasonable time. If the landlord does not act, the tenant has powerful self-help options: courts have squarely held that a tenant may simply stop paying rent, with no obligation to notify the landlord first.
As for damages, recovery is generous. Emotional-distress damages are routinely awarded for the discomfort of a cold apartment, courts have approved treble damages where the failure lasts more than thirty days, and a tenant who prevails is entitled to recover relocation expenses as a matter of course.