The AI Council, in One Page

A guide to the series on building and testing an AI layer for our systematic framework. Read it first, or keep it open beside any part.

What we are building

We run a systematic fund: rules-based strategies that manage real capital, including our own, and that read the market through a framework we call the Five Forces. We are also building an AI layer that reads the same five Forces, but through information a price rule was never designed to hold: the language of a policy statement, the meaning of a surprise, the context a number compresses away. We call that layer the Council. It runs in a paper environment, not with client capital, while we find out whether the design holds. The series is a working journal of that build: what we made, what we doubted, what the record showed, and what we changed.

The conviction underneath it belongs to our founder and is stated in the first piece: that artificial intelligence will become the dominant method for managing systematic investing. The discipline underneath it belongs to the whole firm: the AI earns a role only by meeting the same rigor we demand of every strategy we run.

The cast, in the order the read flows

The Five Forces. Our framework for how markets work, and it came before any of this: macro, technical and price, options and positioning, fundamentals, and sentiment. Everything the Council does is bounded by it.

The deterministic book. Our rules-based strategies, which already read all five Forces by distilling each into a clean signal. They manage the real capital. The Council is a second reader of the same world, not a replacement for them.

The Force analysts. One agent for each Force. Each is allowed to know only inputs that have earned a place, and it never sees a number that is not real, current, and from the source that produces it.

The Chair. The agent that turns the five analysts’ reads into one structured view: a regime, a direction, and how extreme conditions are. A reasoning read has to become structure to be usable, and the Chair is where that happens.

The wall. The number the system acts on is computed in code, from validated signals, before the model says a word. The model’s prose can propose; it can never set the dial. This is how we keep fluency from being mistaken for proof.

The Heretic. An adversary whose only assignment is to argue that the view is wrong. Behind it stands a battery of tests that doubts our methods the same way, and a scorecard we doubt last of all.

The independent manager. The role that forms an allocation of its own for books we run in paper, inside a mandate we fix in code. Where the Council’s judgment actually acts.

The guardian. A circuit breaker over the deterministic book. It may propose a defensive position from a short, pre-approved list, capped in size, and it is idle by default.

The leash. The mandate both roles hang from: the universe they may touch, the caps on exposure and turnover, and a person who can override any of it.

The record. Every read and every decision is frozen at the moment it is made, graded after the fact, and never rewritten. The Council reads its own graded record each night; it is not allowed to learn from its own trades.

The twin. The Null Council: the same model, the same data, the same rules, with everything we built removed. It runs beside the Council every night as a control group, so that what our apparatus adds becomes a number rather than a claim.

The series, in order

Part 1. AI and the Future of Systematic Investing. Where the AI fits: a second reader of the Five Forces, and why it cannot be backtested the way a rule can.

Part 2. Garbage In, Hallucination Out. What the Council is allowed to know, and the one rule we will not bend.

Part 3. From Votes to a View. Why a language model cannot vote the way our rules do, and what its read has to become instead.

Part 4. Fluent Is Not Proof. Making a fluent model trustworthy: the wall between its prose and the numbers that act.

Part 5. The Heretic. The adversary we build to lose to, and how far its doubt reaches.

Part 6. The Last Word. The two roles that turn a read into a position, and the leash that holds them.

Part 7. Hindsight at Machine Speed. Why the Council does not learn from its own trades.

Part 8. Fluent Everywhere, Valuable Somewhere. Where AI actually adds value in a fund like ours, and where it does not.

Part 9. The Control Group. How we know the process is real rather than plausible: the twin.

Part 10. Taking the Win Apart. What the control found in our best result, and why a model’s confidence is not a measurement.

Part 11. Obedient by Design. What the independent manager did with the slack we gave it, and where the line between structure and judgment now sits.

Blackworks Capital LLC manages funds through Blackworks Capital Management LLC, an Exempt Reporting Adviser. Nothing here is an offer or solicitation or investment advice. The systems described are in research and paper-testing and do not manage client capital. Past performance does not guarantee future results.