Every evening, after the market closes, someone on television explains with complete confidence why stocks moved the way they did. The explanation is fluent, plausible, often even correct. And yet the same person, handed tomorrow's headlines in advance, could not trade their way out of a paper bag. That failure is not an accident of punditry, and self-deception only explains part of it. The rest is a fact about causality itself, and it hides not in the definition we all carry around but in how we use it.

I want to take that definition apart carefully, because the crack runs deeper than it looks.

A common definition of causality combines two statements: if X happens, Y happens, and if X had not happened, Y would not have happened. The second clause, the counterfactual, is what earns its keep. Without it, we would be forced to relate any two events that reliably co-occur. The sun rises, and somewhere on Earth, rain falls. Both have happened every single day throughout recorded history; as correlations go, this one is flawless. But two events do not become related by showing up together. To call sunrise the cause of rain, we would also need the counterfactual: had the sun not risen, nowhere on Earth would it rain. And no observation can verify this statement: no one has ever run the experiment of canceling a sunrise. Counterfactuals are checked against a model we trust, which is to say, against science. Our models find no immediate causal road from sunrise to rainfall, so the flawless correlation is dismissed. (Ask over a long enough horizon and the verdict flips: solar heating drives the whole hydrological cycle. Whether sunrise causes rain depends on the timescale you ask at, a small oddity now, a load-bearing one later in this series.) The definition works.

But there is a silent assumption underneath, and it sits in the way we use the definition rather than in the definition itself. Look at the two clauses again. If X, then Y measures whether X is enough to bring Y about. If not-X, then not-Y measures whether X was needed for the Y that actually happened. These are two different quantities. Backward attribution asks the second question; forward prediction asks the first; and the answers travel together only in a special kind of world: a machine in which each state has exactly one successor and exactly one predecessor, one past and one future threaded through every moment like beads on a string. The definition never promised us that world. We assume it silently, every time a narrative verified backward gets read forward at no extra charge. And for most of the systems we actually care about, the assumption is false.

The simplest way to see what breaks is to look at a tree.

Start from a leaf and trace it back toward the trunk. You get a single possible path. At every fork you encounter from below, there is no ambiguity about which larger branch you came from; the geometry decides for you. And the backward counterfactual holds perfectly: if this larger branch had not existed, this leaf would not exist either. Standing at the leaf, causality is exactly what the textbook promised.

Now reverse the journey. Start at the trunk. Immediately the wood divides, and divides again. There is no statement of the form if this trunk exists, then this particular leaf must exist. The trunk does not know its leaves. There were a thousand other ways the branching could have unfolded, a thousand other canopies this same trunk could have carried.

So let me name the two operations, because the rest of this series will lean on the names. Retrospective causality is the causal story constructed by tracing backward from an outcome that already happened. Prospective causality is the forward claim that given these conditions, that outcome will follow. The textbook definition treats them as one relation read in two directions. The tree says otherwise: they are different operations, with different guarantees, and only in a special kind of world, one that maps states one-to-one with no open doors, do they collapse into the same thing.

When does the real world branch like the tree? Whenever the system is not deterministic, or, far more commonly, whenever it is open: influenced by things outside whatever boundary we drew around it. I will make that precise later in the series, because the question of where we draw the boundary turns out to carry more weight than it seems. For now, the tree is enough.

Return to the market. Suppose the evening explanation is: Apple jumped five percent because earnings beat expectations. Let us be generous: suppose this retrospective narrative is one hundred percent correct. It traced the leaf back to its branch, and the path it found is real.

Now run the test. Next quarter, imagine you somehow know in advance that earnings will beat expectations. You take the narrative, flip it forward as prospective causality, and invest accordingly. You will quickly be humbled. The stock may fall on the same news, shrug at it, or move for reasons that were never on your map. And here is the revealing part: when the thesis fails, you will not abandon the method. You will go looking for a new retrospective narrative to explain why it failed and you will find one, because a path from leaf to trunk always exists. The explanation machine never comes back empty-handed.

Notice what that last fact quietly concedes. The backward path is unique, but your story about it is not the path. We reconstruct the corridor from records, and records are incomplete; several incompatible narratives can fit the same surviving evidence, which is exactly why a fresh one is always available after each failure. (Why records must be incomplete has a beautiful answer of its own; it gets its own essay.) So the pundit's confidence is doubly unearned. The forward direction was never guaranteed and even the backward direction, where uniqueness holds in principle, does not guarantee that this narrative found it.

A method like this explains everything and forbids nothing. It is not describing the future's structure; it is describing the past's geometry. The pundit is not lying. The pundit is walking down the corridor and calling it a map of the hall of doors.

Nearly every historical causal narrative is retrospective causality. This is why “history repeats itself” is a sentence uttered only while facing backward, where the patterns sit still and let themselves be admired. Some people do try to predict from historical patterns. Most fail; we simply never hear from them, because the microphone is handed to the successful few. Survivorship does the editing for us.

I am not claiming history holds no wisdom for the present. A retrospective narrative, carried forward with care, is still a better prior than an arbitrary guess: it can prune the space of futures, rule out branches, narrow the hall. But pruning the possible is not the same as picking the actual. That distance between narrowing the doors and knowing the door is exactly the distance between the two causalities. Confuse them, and the error does not stay constant; it compounds, narrative by narrative, each failed prediction patched with a fresh explanation that will fail in its own turn.

So here is the single question I want to leave in your pocket: When you meet a causal claim (in a market recap, a history book, a post-mortem meeting, your own head, etc.) ask: in which direction was this claim built?

Backward-built claims are cheap. There is always a path from the leaf to the trunk, so finding one proves very little; the corridor guarantees its own existence. Every engineering incident review finds a root cause and when the fix does not hold, the next review finds a different root cause, delivered with the same confidence. Forward-built claims are expensive, because the trunk genuinely does not contain its leaves. Nothing is wrong with retrospective causality. We will see later in the series that it is, in a precise sense, the only kind of story the past permits as long as it is labeled as what it is, and never smuggled across the direction line with the price tag still reading “prediction.”

One question, though, is now hanging in the air, and it is the strangest one. Why is the backward direction unique at all? Why does the past reach us as a single fixed corridor while the future fans out into doors. What, physically, makes the two directions so different? That answer is not logic. It is physics: the past and the future are not even written in the same language. The past is classical. The future is quantum mechanical.

That is the next essay.