The Contradiction Scanner: Blunt Feedback With Receipts

The Contradiction Scanner points out where your stated values and your actual behavior don't line up. It's short, and it pairs well right after the Self-Discovery Profile, once the model has a picture of what you say you care about.

The "roast me, find my contradictions" idea has been going around in plenty of versions. The ones I've seen mostly reward the model for sounding harsh. This is the grounded version I've kept: it can only flag a contradiction it can actually point to, so it's honest instead of just performing bluntness.

The prompt

Point out the contradictions you can actually see in what I've told you. Where do my words not match my actions?

Rules: every contradiction has to cite the specific things that clash. "You said X here, but then did Y there." If you can't point to both sides of it, don't raise it. Be direct and don't cushion it, but don't manufacture tension for effect either. A real contradiction with receipts is worth ten dramatic ones you made up. If you don't see many, say that.

How to use it

Run it in the same chat as the Self-Discovery Profile, or in any chat with real history behind it. It needs material. A fresh conversation with two messages in it doesn't have enough of you in it to contradict.

When the list comes back, go through it one item at a time. For each one, check the two things it cites. Did you really say that? Did you really do that? Is there context the model doesn't have?

Tip

If one item stings, ask a follow-up about just that one: "What would it look like if I actually lived the first half of this?" The broad scan finds things. The follow-up is where you figure out what to do about them.

What it's good at, and where it isn't

The whole prompt hangs on one rule: "If you can't point to both sides of it, don't raise it." Without that, asking a model to find your contradictions invites it to manufacture them. Language models are good at producing text that sounds insightful, and a dramatic "you claim to value rest but you never stop working" reads as sharp whether or not you ever said either half. Requiring the receipts turns each item into something you can verify in about ten seconds.

"If you don't see many, say that" matters just as much. Models tend to give you the shape of answer you asked for, so "find my contradictions" leans toward finding some. Explicitly allowing a short list, or none, takes that pressure off. If it comes back with two items, that's not the prompt failing. That might just be the answer.

A few honest limits:

• It only knows what's in the chat. Your actual life has a lot more in it, and some things that look like contradictions have perfectly good explanations the model never heard.
• Changing your mind isn't a contradiction. If you said one thing months ago and something different last week, that might be growth. Worth telling it so.
• Direct doesn't mean right. The prompt says don't cushion it, and it won't. That makes the wrong items land just as hard as the right ones, so check them before you believe them.

Heads up

This is a thinking tool, not a verdict on you. If a pattern turns up that's bigger than a mismatched habit, Emotional Pattern Mapping is a slower, more careful way to work through it.

From here, the thread moves on to looking ahead with the Five-Year Prediction and The Flip.