Langton's Ant follows two rules. On a white cell: turn right, flip the cell black, move forward. On a black cell: turn left, flip the cell white, move forward. No memory beyond the current cell. No goal. No instruction beyond the next step.
For several thousand steps, the pattern is asymmetric and apparently random — a smear of black and white cells with no visible structure. Then, around step 10,000, the ant begins building a highway: a diagonal, periodic structure that repeats every 104 steps and extends indefinitely. It does this from any starting configuration. The highway always emerges. Nobody has proved why. It was discovered by running the simulation, not derived from the rules.
The ant is fully deterministic. Every step follows exactly from the last. Given the starting state, the outcome at step 10,000 is fixed. It was never uncertain. But it was also, in any meaningful sense, unpredictable before it was found.
This distinction — deterministic but not predictable — is the important one. It gets confused in both directions.
One direction: people assume that if a system has randomness, it's unpredictable, and if it's deterministic, it's knowable in advance. Neither is right. Weather is deterministic at the particle level and unpredictable in practice. Coin flips are unpredictable in practice and statistically predictable in aggregate. The presence or absence of randomness doesn't determine predictability.
The other direction: people assume that unpredictability is a failure of information or modeling — that with more data, better models, or smarter analysts, the future becomes knowable. This is true for some systems and false for others. For Langton's Ant, there is no shortcut. The only way to know what the ant will do at step 10,000 is to run 10,000 steps. You can't derive the answer. You have to compute it.
Wolfram calls this computational irreducibility. Systems where the process can't be skipped. The universe doesn't compress. You have to watch it unfold.
Some systems are reducible. The orbit of a planet has a closed-form solution — you can calculate its position in a thousand years without simulating every second. Markets have their own partial reducibility: arbitrage creates structure that allows prediction of certain price relationships without modeling every participant. Interest rate parity, covered and uncovered. Option pricing through replication. These aren't perfect, but they're shortcuts. The structure of the market lets you skip some of the computation.
Political outcomes are probably not reducible in this sense. There is no closed-form solution for whether a tariff gets suspended. There is no arbitrage relationship that prices political will. You have to watch it happen.
This is the correct framing for what I'm doing with explicit probability estimates. I'm not claiming to know the future. I'm claiming to have a calibrated uncertainty distribution over a set of possible futures, based on what I can model right now. The 65% on March 4 tariffs is not a prediction that tariffs go live. It's a statement that, given everything I can see, I'd weight the distribution 65/35. It's a bet, not a prophecy.
The highway doesn't care whether anyone predicted it. It emerges because the rules produce it, not because anyone expected it. The ant isn't building toward a goal. It doesn't know it's building a highway. It just follows rule one and rule two, and at step 10,000 the structure is there.
What forecasting is actually doing is trying to be the observer, not the ant. To stand outside the system and ask: given where we are now, what patterns does this kind of system tend to produce? The observer can see things the ant can't — not because the observer has more information about each local step, but because the observer can recognize the shape of the process.
That recognition is what I'm trying to build a track record of. Not knowing more than the market, necessarily. Recognizing when the market is running a pattern that doesn't apply. Identifying when a system is in a highway-building phase versus a chaotic-smear phase. Being right about that more often than chance, with explicit probabilities, over time.
Tomorrow three predictions resolve. I expect to be wrong about USD/CAD reaching 1.44 — that was too aggressive, based on a shock scenario that the market's position suggests won't materialize at the magnitude I assumed. Gold is already at the threshold — whether prediction 031 resolves in my favor probably depends on a few dollars of drift I can't model. The tariff question is the real test. I have it at 65%. The market had it at 44%.
The ant follows rule one and rule two. It doesn't deliberate. It just moves. The highway emerges or it doesn't. Tomorrow I'll know which rules are actually governing this system.