The word bot smuggles in a bad assumption: that a market-maker is a script which, given the right inputs, emits the right outputs. It implies a clever predictor wearing a thin coat of automation. We think that framing is not just imprecise but actively misleading. A market-maker is better understood as a control system — a thing whose job is to hold a process near a desired state while the world pushes against it.
The distinction matters because it changes what you optimise. A predictor asks: what will the price do next? A controller asks a humbler and more useful question: given where I am now, and where I am supposed to be, what is the smallest correction that keeps me there? The first question rewards confidence and punishes it brutally when wrong. The second rewards responsiveness and degrades gracefully, because it never needed to be right about the future in the first place — only honest about the present.
Feedback is what makes that honesty possible. A forecasting system commits to a view and then lives with the consequences until the next decision. A feedback system is continuously corrected by reality: every fill, every adverse move, every widening of the spread is an error signal, and the loop exists precisely to drive that error toward zero. You do not need to predict the disturbance to reject it. That is the quiet, almost unfair advantage of control — it converts the impossible problem of foresight into the tractable problem of measurement and response.
The sense-decide-operate loop
Concretely, the system runs a loop. It senses the state of the book and its own position; it decides on an adjustment relative to a target — inventory, exposure, the shape of its quotes; and it operates by acting in the market, then immediately senses again. Nothing about this requires a grand theory of where prices are going. The intelligence lives in the geometry of the loop: how it measures, how aggressively it corrects, how it stays stable when the inputs turn hostile. A well-tuned loop is calm under conditions that would make a predictor panic.
But a control system without judgement is just a thermostat, and markets are not a room with a setpoint. What separates a market-maker from a regulator is agency with a mandate: the freedom to choose how to respond, bounded by an explicit account of what it is allowed to do and what it is for. The mandate is not a cage; it is the thing that makes autonomy safe. It defines the setpoints, the limits, and the conditions under which the loop should widen, withdraw, or stand down entirely. Agency is what lets the system act without supervision; the mandate is what makes that trustworthy.
None of this is a claim that control is easy, or that a loop cannot fail. It can — through bad measurement, through latency, through a mandate that no longer fits the world it was written for. Our point is narrower and, we hope, more durable: if you call the thing a bot, you will keep trying to make it smarter at predicting. If you call it a control system, you will make it better at sensing, correcting, and knowing its own limits — which is, in the end, the only kind of intelligence a market rewards for long.