“Time will tell.” “It depends on execution.” “There’s real upside and real risk.”

You’ve read these lines in memos and analyst notes, and probably written a few yourself. None of them is wrong, and that’s the problem. Each one fits every outcome. If the launch works, there was real upside. If it fails, there was real risk. A take that survives every outcome hasn’t said anything.

Ask a model whether a launch, a hire or a strategy will work, and these are the lines you’ll get back. That isn’t the model being lazy. It answers the question you gave it, and “will it work?” has nothing to be right about. When nothing can be right, the best answer available is the one that can’t be wrong: both sides, well written.

And you can’t grade it. A sharp answer and a lazy one read the same. Both are fluent, both sound balanced, and nothing on the page tells them apart. You end up keeping whichever sounds smartest, which is how a persuasive wrong answer gets through.

The fix is to give the question something it can lose. I call it a kill date: four lines you write before you ask anyone, a model included. One question keeps all four honest: could a stranger tell you were wrong, without asking you?

Verdict: one side. If you can add “but,” it’s two takes. Add a percent if you have one, so you say how sure you are, not just which way you lean.

Metric: one number you could look up. This is where the thinking happens. Choosing it forces you to say what “working” means, and plenty of takes quietly die here, when their owner can’t name the number. A number you can look up isn’t enough on its own, though. It has to be the number the argument is about, and one someone who disagrees with you would accept as the judge.

Wrong by: a real date. This is what turns commentary into a forecast. Without a date, “not yet” is always available.

Against it: the strongest fact against you today. Not a weak one you can beat.

Here’s one of mine

Earlier this month, Apple announced the iPhone Duo, starting at $1,999. Bold bet or strategic miscalculation? Ask a model and it’ll tell you it’s both.

My kill date:

- Verdict: Miscalculation, 70%.

- Metric: Apple’s iPhone revenue for the June 2027 quarter, against Visible Alpha’s consensus.

- Wrong by: Apple’s earnings report for that quarter, late July 2027.

- Against it: IDC expects Apple to ship roughly ten million Duos in the first 12 months.

The metric took some thinking. The obvious number was the Duo’s US delivery estimate. It was easy to look up, but a long wait can mean big demand or a small first batch, so it wouldn’t settle anything. What I needed was a reference someone on the other side would accept as fair. Revenue against consensus is that. Consensus is what analysts already expect, so beating it or missing it means something to both sides. It isn’t perfect. iPhone revenue isn’t Duo revenue, and that quarter carries other launches too. But it has a date, and it can go against me.

I committed to my side and my 70% before asking any model. Then I gave three models the same metric and date, and asked each for its own verdict, a percent and the strongest fact against it. None said “both.” Gemini said miscalculation, 60%. ChatGPT said miscalculation, 56%. Grok said bold bet, 58%.

One of the three facts against me held up when I checked: ChatGPT’s, IDC’s ten-million estimate. It’s now my Against it. The model found the case against me. It didn’t get to decide.

Three fluent models, each with a side and a percent, and they didn’t agree. Read cold, none of them sounds wrong. That’s why this matters more as models improve. A better model makes every take more persuasive. Only a number on a date tells you which one was right. The model makes the case. The date settles it.

Time will tell, and now it has a date: late July 2027. I’ll post the result here either way. I may be wrong. The point is that I can be.

Before you ask for the next take. Pick one you’re holding right now: a vendor, a hire, a launch, a market. Fill in four lines before you ask any model about it.

- Verdict:

- Metric:

- Wrong by:

- Against it:

Reply with a metric and a wrong-by date for a take you’re holding right now. One line is enough.