The switch behind the screen

Elm Wealth ran the experiment everyone imagines winning. They handed four leading AIs tomorrow's front page, a day before it printed, with the market moves blacked out, and let them trade the S&P and long bonds on it. A working crystal ball. The one thing every trader who has ever stared at a screen has wished for.

The machines read it well. On direction they matched a bench of professionals you could not assemble twice: the head of trading at a top-five bank, the founder of a top-ten macro fund. They knew which way the market would break. Handed the future, most of them still went broke.

They went broke on size. Given tomorrow's news and asked only how much to stake on it, they levered seven to twelve times and ran daily swings of twenty to forty percent, when the arithmetic of survival called for a third of that. Every one of them could recite the Kelly criterion on a coin. None could apply it once the coin became a market. They knew the rule. They could not size the bet. The ball worked. The hand on it did not.

Every trade is two decisions and only two: what to own, and how much of it. Prediction answers the first, and prediction is precisely the half that frontier AI has made cheap, because it now reads the direction of the world about as well as a room of experts and no better. Bloomberg's Matt Levine set the ceiling on it: if a model knew which stocks would rise, it would be running Renaissance, not answering prompts. Foresight is the crowded half, the near-zero-sum half, the half arbitraged away the moment it is real. The market has never paid for knowing what happens next. It pays for reading what is already true.

And what is already true is that there is a hidden switch behind every market, and it decides which of those two decisions is the one that matters today. Most of the time it sits in one position, and the winning move is to size and to ride. Then it flips, and the same moves are ruin. The whole of investing skill, once prediction is cheap, is knowing which way the switch is set before you choose how to play. Not what will happen. What state you are standing in.

That switch has a name, and it is permission. Before growth or value or news can matter, permission determines how the market responds to them. It is the state that governs whether risk is currently allowed at all. When permission is on, capital is willing, and money flows into whatever it is already flowing into. When permission is withdrawn, the willingness goes, and everything that depended on it falls together. Permission is the governing state, and the reason it governs is that it is upstream of everything else the market does.

Permission is observable because it is priced. My preferred reading of it is the spread on BBB corporate debt, and one level on it that works as a tripwire, because credit is where risk appetite is priced first and most honestly. Others will reach for the VIX, for the MOVE index, for dollar funding or repo, and those are not wrong; they are lenses on the same thing. The argument here is not that one scalar is the market. It is that permission is the state that matters, and that a credit spread reads it more directly than most. Why that particular spread carries the regime is the argument of Equity positioning in an age of permissioned levitation. What builds behind the wall while the spread stays calm, the stress pooling in private credit that the public market has not yet been made to print, is the argument of Private credit dislocation. Read the spread and you know which way the switch is set. That is all it is for.

Read it, and it routes the entire approach.

With permission on, return does not come from being right about value. It comes from flow: money moving into what money is already moving into. The cheap thing stays cheap and the rising thing keeps rising, because permission is intact and flow is self-feeding. So the regime carries two things side by side, and how you weight them is a matter of your own appetite for risk. There is low-risk money to be made simply riding the beta, standing where the flow concentrates mechanically, in the largest and most liquid names, the mega-cap index, a Global 100 or whatever tracks the same gravity. You do not pick. You own the concentration. And there is more alpha to be taken alongside it, for those who want to reach, by hunting the momentum the regime rewards. That is Flowsurfer: an engine that looks for return where return is driven by momentum rather than mispricing, because when permission is on, momentum is what pays. Beta for the base, Flowsurfer for the edge.

With permission off, the ground reverses. The flow that lifted everything runs out, momentum breaks in the hand, and the names that rose because they were rising fall for the same reason. Now being cheap matters again, because dispersion returns and price has to answer to what a business can actually deliver. Here you no longer ride; you read. The method is impliedby: take the price the market is quoting and run the discounted cash flow backwards, not to ask what the company is worth, but to extract the growth and the margins that price is silently assuming. Every quote encodes a forecast, whether the buyer knows it or not. Set that implied forecast against what the business can plausibly deliver, and the space between the two is the delivery gap. Screen for it and it resolves two ways. To the handful of names where the market's implied story is one the company can beat. Or, when nothing clears the bar, to nothing. Cash is a position. Standing aside is a trade.

Which is where the whole thing stops being an argument. Run that screen today and it returns almost the entire equity market as untouchable. Nothing clears the bar, because at today's prices the implied forecasts are already richer than the businesses can meet. That is not a shrug. It is what permission-on looks like read from the value side: nothing is cheap, because risk is still allowed and flow is still lifting everything above what delivery can justify. The switch is set to ride, and the tools are turning against current data to say so.

Which returns us to the four machines and their crystal ball. They had tomorrow's paper and went broke anyway, because they mistook the ball for the instrument. It never was. The ball shows you the future. It does not tell you which way the switch is set, and so it does not tell you whether to size, to ride, to hunt, or to stand aside. That is the one thing the machine handed the future could not supply. AI has made prediction cheap. It has not made judgment cheap. If anything, by making prediction abundant, it has made judgment the only scarce asset left. The full Elm result is here.