You asked three chatbots and got three answers

You opened three tabs and asked the same question three ways. One assistant was warm and told you love is worth the leap. One was cautious and asked whether you'd resent it. One printed a tidy table of pros and cons and let you keep both columns. Four hours later the offer letter for the sublet is still open, and you are exactly where you started — with more words and no verdict.

Here's the part nobody names: you didn't collect three opinions. You collected your own three moods, read back to you in a confident voice.

The dilemma

This is a composite of the kind we see all the time in the deciding-with-ai category — stitched from many real submissions, no real person.

Nisha, thirty-one, has a stable remote job and a lease that ends in six weeks. Her partner, Arjun, has been offered a role 700km away that he has wanted for years — the kind of role that opens once. He asked her to come. She said yes in the moment and has been quietly unsaying it ever since.

She loves him. That part is not in question. What is in question is the sentence she hasn't typed into any chatbot: he didn't ask whether I wanted to move — he asked me to move to where he's going. There's a difference, and she can feel it, and she keeps sanding it off her prompts.

So she asked three AI tools. To the warm one she wrote "we're so excited about this new chapter." It said yes. To the cautious one she wrote "I'm not sure I'm ready to give up my whole life." It asked gentle questions and leaned no. To the third she wrote a neutral summary, and it handed her a balanced list she already owned. Three tools. Three answers. The thing she is actually deciding never appeared in any of them.

The read

Run the same dilemma through the eight lenses, and notice they don't move when the wording softens or hardens. That's the whole point.

Honesty — what's true between them (and inside her)

Honesty asks whether the account you're giving matches the account you'd give at 1am. Here it sees a deletion. The excited prompt is true; the resentment is also true; only one made it into the box. The move-to-support story is honest. The he never asked what I wanted story is honest too — and it's the one being withheld from every tool and from Arjun. A curated omission scores mildly negative here — not a lie, but honesty's most common disguise.

Non-harm — who absorbs the cost

Non-harm asks who quietly pays. Moving harms no one on the surface; Arjun gets his role, Nisha keeps her remote job, the lease was ending anyway. But an unspoken resentment carried 700km doesn't evaporate on arrival — it compounds, and it lands on the person she moved for. The move itself reads faintly positive; the silence riding shotgun drags it back toward the line. The gentle act and the withheld truth are stapled together.

Motive — the engine under the verb

Motive asks what's actually driving the action once you strip the reasons you already believed. Under "support him" and "fresh start," there's a quieter engine: if she goes without saying the true sentence, she gets to be the good partner AND keep a grievance in her back pocket for later. Support with a receipt attached. This is the lens that scores worst — the generosity is real, and it's also collateral for a debt she hasn't named.

Discernment — is the question even the question

Discernment asks whether you're deciding the thing in front of you. Nisha thinks the decision is move or don't. It isn't. The decision is say the sentence or don't — and the move only becomes decidable after that. Asking three chatbots "should I move" is asking a question the real question is hiding behind. The lens marks negative because the frame itself is wrong.

Now the tension worth staying with: honesty and non-harm are pulling against her instinct to keep the peace. Staying silent feels like non-harm — no conflict, no hard conversation. But honesty scores it as a withheld truth, and non-harm scores the withheld truth as the actual injury, delivered on a delay. The kind move and the honest move are not the same move here, and she's been shopping for a tool that will tell her they are. Three lenses dragging negative under one leaning-positive act is not a close call — it's a decision aimed at the wrong target.

The verdict: this isn't a move-or-stay decision, and no amount of re-asking will make it one — say the unsaid sentence to Arjun before either of you touches the lease.

Here's what a general AI chatbot structurally can't do, and it's not a knock on the machine — it's the shape of the machine. Ask it warmly, it warms up. Ask it defeated, it commiserates. Sycophancy — agreeing with the framing it's handed — is a documented behavior of these models, which is why three tabs gave you three moods back. KarmaLens doesn't renegotiate. The eight lenses are fixed, the scoring aggregates the same way whether your prompt is hopeful or bitter, and the verdict does not soften because you leaned on it. When a verse is attached, it's cited verbatim from the corpus, not paraphrased to flatter your angle. You don't get to pick the answer you like best from a spread of three. You get one committed call — and one call you can argue with is worth more than three you can shop between.

The takeaway you can run tonight

Free, no login. Call it the three-prompt collapse test.

1. Open a note. Write the three prompts you'd give three different tools: the hopeful version, the defeated version, and the "neutral" version.
2. Read them side by side. Circle every concrete fact that appears in one prompt but vanishes from the others.
3. Those vanishing facts are not noise. They're the load. Finish this line for each: "I phrased it that way so the answer would be ___."
4. Now write the ONE sentence that appears in none of the three prompts — the thing you're steering every version away from.

If that sentence exists, you already have your finding. The chatbots weren't disagreeing with each other; they were each faithfully mirroring a different mask you handed them. Three verdicts you get to pick from is zero verdicts, because the choosing was always yours and you're using the tools to launder it.

Say the collapsed sentence out loud once. Then say it to the person it's about — not to a fourth tab.

When you're ready to stop collecting answers and receive one you can't talk down, run your dilemma through the eight lenses. If you'd rather watch a few committed calls get made before you make yours, the gallery is right there. So — which sentence have you kept out of every prompt so far?