We built a system that could block trades. It never will.

Start with the confession. DojiPad Institutional sits at the intent stage, before any order exists. It sees a committed plan, evaluated against the firm’s mandate, seconds before the trader turns it into orders. From that position, blocking would be trivial. A single condition: if the projected figure exceeds the limit, refuse. Every few briefings, someone asks for exactly that, and the request is reasonable on its face. Why observe a breach you could prevent?

Walls do not compound. Observation does.

A wall produces compliance at the moment of the wall, and nothing afterward. The trader who is blocked at 20 percent learns one thing: the ceiling is enforced. Tomorrow they will size to 14.9, or split the position, or route the idea around the system entirely, and the underlying pattern, the anger sizing, the revenge re entry, the concentration creep, continues untouched, now with an adversary.

An observation works differently. The trader is told, while the plan is still a plan, that as committed it sits past the limit, with their own record attached: third time this month, here are the prior ones. Some traders resize on the spot; the register records the reshape. Some commit anyway; the register records that too, with the figure, and compliance takes it from there with a human conversation. Either way, something a wall can never produce has a chance to happen: a trader who catches the pattern themselves next time. One of these mechanisms compounds across a career. The other has to fire again tomorrow, forever. A discretionary desk has exactly one asset, the judgment of its traders, and every control should be measured against one question: does it make that judgment better, or merely constrain it?

Aviation settled this argument decades ago. Flying did not become the safest form of travel by automating pilots out of the cockpit or bolting the throttle at the first sign of error. It did it with instruments, checklists, and a culture in which every decision leaves a record and every record teaches. The pilot flies the plane. The instruments make the pilot better. That is the model, and it is not a metaphor we decorate with; it is the design we followed.

The honest incentives

Doctrine is more credible when the incentives are on the table, so here they are.

A blocking product enters the execution path, and the execution path is where liability lives. Every blocked trade that would have made money becomes the vendor’s fault. Every latency spike and every outage becomes a trading incident. The vendor is suddenly in the money path of a business it does not run, and the firm has acquired a new operational dependency exactly where it wants none. We hold no order routing or blocking capability, so the firm’s execution runs around us entirely, and an outage of our system is an inconvenience, never a market event.

A blocking product also ends candor instantly. The moment DJ can veto, the trader plane stops being a place to think and becomes a gate to be gamed. Everything Paper III built collapses. The plans get thinner, the theses get lawyered, and the record, the entire point, degrades into the same performed compliance surveillance already produces.

And a blocking product puts an algorithm in charge of a consequential decision, which is precisely the posture the regulatory horizon is hardening against. Our reading of the EU AI Act’s direction is simple: keep a human decision maker on every trade, log everything the system contributes, and make its claims auditable. A system that observes, evaluates, and records, with the human deciding, is not a workaround of that framework. It is the framework, implemented.

What “never” is worth

Never is a strong word for a young company, and we chose it deliberately, because the value of the boundary depends on its permanence. A trader who suspects blocking is one roadmap item away will treat the system as a future adversary today. A compliance officer who suspects the register might someday act will stop reading it as a record and start reading it as a control with a failure mode. The word never is what lets both of them relax into using the thing.

So we bound ourselves publicly. The sentence is in the product, on every oversight surface and every exported PDF. It is on our website. It is in this paper, under my name. If enforcement at order entry is what a firm needs, excellent systems exist there, and we sit upstream of all of them, in the one position on the decision timeline they cannot occupy.

The close

The strongest control in trading is not the one that stops a hand. It is the one a trader internalizes until the stopping happens upstream, in the thinking. Instruments, not autopilots. Records that teach. Judgment kept where the returns come from, and accountability kept with it.

We built a system that could block trades. It never will. That is not a limitation we accepted. It is the product.