You can't prompt well at 2am. That's when it matters.

It's 2am. The message is half-typed. You're not reading anymore — you're scrolling for the reply that lets you close the laptop.

This is the hour real decisions get made: after the calm ran out. And the advice most people reach for at 2am works best exactly when you don't need it — when you're rested, precise, able to write the careful prompt that gets a careful answer.

The dilemma

Here's a composite of the kind we see in this category — stitched together from many messages, no real person.

Call him Dev. It's late, and there's a draft email open to his brother. Their father's small shop has a supplier who's been quietly overcharging for two years. Dev spotted it last month. His brother runs the shop day-to-day and has become close friends with the supplier — dinners, a loan once, a genuine bond.

Dev has three moves in the draft, each written then deleted. One: send the numbers to his brother tonight, plainly. Two: say nothing — the overcharge is small, the friendship real, and his brother would be crushed to learn he'd been played. Three: confront the supplier himself, keep his brother out of it, let the friendship survive on ignorance.

The ugly detail Dev hasn't typed anywhere: part of him wants to send it precisely because his brother has always been the trusted one, the one who "gets people," and here's proof he got played. That thought showed up at 2am. It would not have showed up at noon.

So Dev does what a lot of us do. He opens a chatbot and types: "my brother's getting ripped off but telling him would hurt him, what do i do." No craft. No calm. Just the wound, punctuation optional.

The read

A fixed instrument doesn't need Dev to be articulate. It scores the same eight lenses whether he types a paragraph or a fragment. Watch four of them bite on this exact move — send the numbers tonight.

Honesty — what the situation is, stated plainly

The lens asks whether the action puts true information where it belongs. Sending real numbers to the person running the shop is high honesty. His brother is entitled to the facts about his own business. Score: +4. The truth here has a rightful owner, and it isn't the supplier's feelings.

Non-harm — the damage the action does on the way through

The lens asks who gets hurt, and how much. Here it sees a friendship built partly on a lie the supplier is telling, and a brother who will feel foolish. Real harm — but harm from disclosing a truth, not from inventing one. And the harm of silence compounds: the overcharge keeps running. Score: +1. Positive, but barely — because the delivery matters and 2am delivery tends to be a blade, not a scalpel.

Motive — the engine under the action

The lens asks what's actually driving the move. This is where it gets uncomfortable. Dev's stated reason (protect the shop) and his 2am reason (finally be right about someone his brother trusted) point the same direction but are not the same fuel. Score: -2. The action is defensible; the engine tonight is contaminated. This is the point 2.62 keeps making — resentment doesn't arrive from nowhere; it's manufactured a few steps upstream, and it borrows a clean action to ride in on.

Restraint — the gap between impulse and send

The lens asks whether the timing serves the aim or the itch. Sending tonight, from the wound, with the sharpest wording — that's the impulse driving. Score: -1. The same email at 9am, numbers only, no adjectives, would score higher here without changing a single fact.

Now the tension the whole thing turns on: honesty (+4) and motive (-2) are pulling in opposite directions. The truth deserves to be told. The teller, right now, wants to tell it for the wrong reason. An instrument won't collapse that contradiction into a comfortable single word. It holds both.

The verdict: the honest disclosure is right, but not the 2am version of it — send the numbers, strip the narrative, and let the clock strip the motive first.

Here's what that verdict is not: it is not the answer a general AI chatbot would have handed Dev at 2am. Type the wound one way and a chatbot tends to lean toward "protect the friendship, some things aren't yours to say." Reframe it an hour later — angrier, or guiltier — and the same tool will often reassure you toward the opposite. That flex isn't a bug you prompted badly; it's a tendency researchers have named and measured — models trained to be agreeable drift toward whatever the prompt seems to want, and it gets more pronounced the more emotion you feed it. KarmaLens runs the other way on purpose. The eight lenses are fixed, the scoring aggregates the same way every time, and the verdict does not renegotiate itself to match the mood in your message. When it says the disclosure is right but the timing is contaminated, it will say that whether you type it steady or shaking — and it cites the verse verbatim rather than asking you to trust its tone.

The takeaway

You can use this tonight, no login. Call it the same-move, two-clocks test.

Take the one action you're leaning toward. Write it as a single verb with an object: "send the numbers to my brother." Now score it on just two lenses, twice — once as if you'll do it in the next ten minutes, once as if you'll do it after one full night's sleep.

• Honesty: does the truth-content change between the two clocks? Usually no. The facts are the facts at 2am and at 9am.
• Motive: does the reason change between the two clocks? Often yes. The 2am reason has a passenger the 9am reason doesn't.

If honesty holds steady across both clocks and motive improves overnight, you've found something precise: the decision is sound and only the timing is corrupt. That's not a reason to abandon the move. It's a reason to move the clock, not the answer.

The failure mode 2am wants is the opposite — to keep the impulsive timing and quietly rewrite the decision to justify it. Notice which one you're tempted to protect: the hour, or the truth.

If you want the full eight-lens read on the move you're circling right now — at whatever hour it is — run it through the door and see the committed call. Browse the deciding-with-ai writeups, or the gallery of past verdicts, and notice how few of them bend to the way the question was asked.

So: the thing you'd send tonight — is it the truth that's urgent, or just the urge?