Paper I · The series
The supervisory record stops where the decision begins
Every control a trading firm runs fires at order entry or after the fact. The decision that costs or earns the money forms before either. This paper measures the hole that leaves in the supervisory record, shows the rules and the industry moving toward it from two sides, and sets out the five properties any record at the intent stage would need.
Where this started
I am a discretionary order flow trader. Years in footprint charts, CVD, open interest, volume profile. And like every discretionary trader I know, I learned the hard way that the market was never my biggest problem. My own patterns were. The revenge entry after a red trade. The size creep after three greens. The plan that quietly changed between writing it and executing it.
So I built DojiPad, a behavioral coaching platform for traders, around a coach that learns a trader’s specific failure patterns, predicts them before a session, and grades its own predictions against real fills. Not vibes. Falsifiable predictions, scored. Then I started talking to people who trade professionally, inside funds and prop firms, and noticed something strange: the most controlled traders in the world have the same blind spot.
Three controls, one silence
A professional desk is wrapped in controls, all competent. Pre trade compliance evaluates coded limits the moment an order is entered. Surveillance scans executions and communications for abuse after the fact. Analytics tell a PM what past decisions cost, months later, from holdings data. I mapped the vendors, the regulatory drivers, what risk officers buy and what they refuse to buy. Across the entire stack, every control fires at order entry or after the trade. Nothing operates in the window before the order exists.
Which is strange, because that window is where the trade actually happens. The decision to take the position, the reasoning, the sizing logic, the conditions for abandoning it: all of it forms before any order touches the OMS. By the time compliance sees it, the thinking is done. By the time surveillance sees it, the money has moved. The intent stage, the richest and most consequential part of the process, produces no record at all. Ask a fund how trading decisions are supervised and the honest answer is limit reports, surveillance alerts, and committee minutes. All downstream. All silent about intent.
What the silence costs
Essentia Analytics, the firm that pioneered behavioral analytics for professional investors, reported in 2020 that it had uncovered an average of 94 basis points a year for its clients, previously lost to behavioral biases, across roughly 150 billion dollars under daily analysis at the time. On a one billion dollar book that is roughly nine million dollars a year, leaking out at a stage of the process nothing in the control stack can see. Their number, their research, and it should keep risk committees up at night.
The finding underneath the number matters more. Decision behavior is measurable, trader by trader, and it persists. Failure patterns are not noise; they repeat. Whatever repeats can be observed, and whatever can be observed while it is happening can, in principle, be supervised while it is happening.
Why now
Two currents made this worth building now instead of later.
The rules are converging on the gap. The SEC modernized electronic recordkeeping under Rule 17a-4 in October 2022 and has spent the years since widening what counts as a record of advice, with the off channel sweep producing billions in penalties from September 2022 onward under Advisers Act Rule 204-2(a)(7). The EU AI Act, in force since August 2024, phases in logging, human oversight, and record keeping duties for AI that touches consequential decisions, on a published timeline that a proposed delay would shift but not cancel. And AI governance now sits on most institutional operational due diligence checklists: allocators ask managers how AI touches the decision process, what controls exist, and what record remains. An intent stage record, generated inside the firm’s own walls, is a new kind of answer to all three.
The industry is moving the same way from the other side. It keeps concentrating into structures where individual trader behavior is the whole business. In a pod, the behavioral layer is not a nice to have. It is the product.
Five properties of a record at the intent stage
If a firm wanted the missing record, it would need five things, and each constrains the design hard.
Contemporaneous. A decision record reconstructed later is testimony. Evidence is written while the plan is still a plan.
Evaluated, not merely captured. A transcript of intent is a pile. A plan measured against the firm’s mandate, clause by clause, in the mandate’s own words, with the figure, is a supervisory record.
Inside the firm’s walls. A record of unexecuted intentions is among the most sensitive data a firm produces. It cannot transit vendor infrastructure, and the guarantee must be architectural, not contractual.
Honest about its limits. A record that silently omits what it could not measure is worse than none, because it reads as clean. Every gap belongs on the face of the document.
Non blocking. The moment an intent record gains veto power over orders, traders route their thinking around it and the record dies. Evidence and influence coexist at the intent stage only if the system observes everything and blocks nothing.
The empty position
Read the timeline left to right: intent, order entry, execution, review. By order entry the thinking is done. By execution the money has moved. By review the quarter is over. The intent stage is the only point where evidence and influence coexist, and until now nothing operated there. That is where DojiPad Institutional runs: traders plan positions in conversation with a coach that knows their patterns, every committed plan is evaluated against the firm’s actual mandate before capital moves, and every plan, check, and observation forms an evidence chain the risk function can read while the conversation is still about a plan instead of a position.
Two design decisions define it, and both came from listening to what desks refuse to buy. The system runs inside the firm’s own tenant with zero data egress, and it runs on two planes: the firm sees process, the trader keeps the conversation. The next two papers take each in turn, and the fourth explains the decision I expect to be argued about most: this system observes, and it never blocks.
The further right a control sits on the decision timeline, the more it can prove and the less it can change.
- Essentia Analytics client research, stated publicly by CEO Clare Flynn Levy, Markets Media interview and company publications, August 2020.
- U.S. SEC, Electronic Recordkeeping Requirements adopting release, October 12, 2022.
- Advisers Act Rule 204-2(a)(7); SEC recordkeeping enforcement orders, September 2022 onward.
- Regulation (EU) 2024/1689 and the European Commission's published implementation timeline.
- AIMA Illustrative Questionnaire for the Due Diligence of Investment Managers; industry ODD analyses, 2025.