Know when you can't verify

Last lesson's move, running a claim down to a source that can carry it, has a failure mode the lesson didn't cover: you find the source and can't read it, or two good sources disagree, or the check would cost an hour the answer isn't worth. This lesson teaches the move for that wall: say "I can't check this" on purpose, then pick one of four honest exits, a person who would know, a source you can read, learning the missing piece, or downgrading the answer from fact to lead. You don't have this move yet because a fluent answer feels the same whether it's right or wrong, so nothing on the inside ever tells you you've hit the wall. This week you'll catch one real answer you were about to build on, name exactly why you can't check it, and run the exit, on a question of yours that matters: health, training, money, a rule. After that, the difference between "sounds right" and "checked" stops being a mood and becomes a decision you make.

Last lesson's move, running a claim down to a source that can carry it, has a failure mode the lesson didn't cover: you find the source and can't read it, or two good sources disagree, or the check would cost an hour the answer isn't worth.

You found the study the AI cited but can't follow its statistics. Is the claim verified?

No. That the study exists and that it supports the claim are two separate checks, and you could only run the first.

Two sources you trust give different answers and you can't rank them. What do you know now?

That the question is contested and your check has hit its ceiling. Deciding between experts takes the expertise you were trying to borrow.

What does using an answer "as a lead" mean?

Acting on it only in ways that survive it being wrong: a question to ask someone, a search to run, never a fact you build on.

When is skipping a check honest?

When the check costs more than the answer is worth, and you drop the certainty along with the check. The claim travels labeled as unverified.

How much skepticism does an unverified answer deserve?

Proportional to stakes: what being wrong would cost and who would pay it. Fine for group chat trivia, disqualifying next to a prescription.

The answer still "sounds right" after a careful reread. What did the reread check?

Fluency only. A wrong answer written well produces the same feeling as a right one, so the reread can't find the boundary.

Why is "I can't check this," said out loud, useful to you?

It forces a decision about how much weight the claim gets. Unsaid, "sounds right" quietly becomes "verified."

A claim arrives wrapped in jargon and study citations. Does that make it hard to check?

Not by itself. "The study is from 1994 by this author" is a two-minute search. The topic's difficulty and the check's difficulty are separate.

When does "learn enough to check it myself" beat asking a person?

When the question will keep coming back, in a field you'll be in for years. Learning is the slowest move, so it needs repeat use to pay off.

1. You read that niacinamide and vitamin C cancel each other out, ask your AI, and it says that claim is outdated, citing a study. You search: the study exists, the title matches. You open it and meet formulation chemistry you can't follow. Where does the claim stand?: Unverified by you, since the study existing and the study supporting the claim are two different checks, and you could only run the first
2. You ask how much caffeine is safe at your age before training. A health authority page gives one number, a sports medicine site you've used before gives a higher one. Two sources you'd normally trust, two numbers. What's the honest reading?: The question is contested past your ability to judge, so act inside the lower number while you find someone who can read the evidence behind both
3. Your AI says magnesium before bed improves sleep. You look into it: the readable sources conflict and the studies are past you. In which of these does the unverified answer disqualify itself?: Passing it to your grandmother, who asks whether she should take magnesium alongside her blood pressure medication
4. Among the four moves, when is "learn enough to run the check myself" the right pick?: When the question belongs to something you'll be in for years, so the learning gets reused
5. You suspect an answer, so you reply "are you sure? give me sources." The AI apologizes for any confusion and restates the claim with three citations. What changed about the claim's status?: Nothing yet; the same system restated it, and the citations are unchecked until you run last lesson's move on them, if you can
6. You bought headphones in a store, changed your mind, and your AI says you have a fourteen-day legal right to return them. You can't judge consumer law. Which of these treats the answer as a lead?: Calling the consumer advice line and asking whether an in-store purchase has a legal return right, letting the claim stand or fall there
7. The group chat is arguing about whether Helsinki had electric trams before Stockholm. Your AI answers instantly. Verifying would mean an hour in city history pages. The honest handling?: Repeat it with the label, "my AI says, I haven't checked," and let the chat price it
8. A friend defends a claim: "I did check it, I read the AI's explanation twice and it holds together." What did the reread measure?: The answer's fluency, which reads the same whether the claim is true or false