Built, Bought, or Blackboxed? Where an Algo Comes From
6 min read · also published as a field note
Every algorithm that has ever traded a live account came from one of three places. It was built, it was bought, or it was blackboxed. Two of those three are how most people meet automation, and they are also how most people leave it, usually poorer and convinced that "algos don't work." Only the first is the one we teach, and it is the only one we run on our own desk. This note is the whole doctrine in one place.
Built
A built algo starts with a trader, not a dataset. The trader has a way of reading the market that they can say out loud: this is the condition I wait for, this is what confirms it, this is where I am wrong, this is where I take profit. It is mechanical in the sense that it can be written down as rules a machine could follow without judgment. Only then does AI enter, and it enters as the builder's hands, not the builder's brain. It writes the code, runs the harness, checks the edge cases, and comes back with questions. The developer leads; the AI is led. Human-led, mechanically derived, AI-built, cohesively, in that order.
Cohesive is the word that matters. The algo has to be of a piece with the trader's actual strategy, because the trader is the one who will have to watch it through a losing stretch and decide whether to keep faith with it. You can only keep faith with something you understand. That is the whole reason our Dominion Tools are taught as something you build rather than something you get, and why the Algo Armory exists as a workshop rather than a shop.
Bought
A bought algo is the one on the marketplace with the smooth equity curve and the five-star reviews. You do not know how it decides, you cannot see its stops, and the person who sold it to you does not know how you trade. That last part is the problem, not the first two. An algorithm is a set of decisions made in advance, and those decisions have to agree with the trader who is going to sit behind them: the same instruments, the same session, the same tolerance for a losing week. A bought algo agrees with nobody. It is a stranger's discipline running on your money, and the day it does something you would never have done, you will not know whether that was the design or the failure. In our own experience, the bought algos we have looked at have ended badly for their owners, and rarely because the code was bad. They ended badly because the owner could not tell the difference between "this is the drawdown the design accepts" and "this has stopped working." We wrote about that gap in every algo has a decay rate: decay is normal, and only the builder can see it coming.
Blackboxed
The blackboxed algo is newer and it feels smarter. You point an AI at a pile of price data and tell it to find the edge. It will find one. It always finds one, because a large enough search over enough parameters will fit any noise you give it, and the result will look wonderful right up until the first day it meets data it has not seen. In our own lab we have watched this exact thing happen more times than we can count: a strategy that looks wonderful on a recent stretch of history and is indistinguishable from a coin flip once it is run over the decades before it. The tell is always the same. Nobody can explain, in a sentence, what the strategy is *doing* in the market. When the logic came from a search rather than from a thesis, there is nothing to explain. An edge you cannot explain is not an edge, and an AI that was told to go find one will hand you exactly that.
There is a second, quieter cost. When the machine leads, the human stops learning. The developer becomes an operator of something they do not understand, which is the bought-algo problem again wearing a lab coat.
How it looks on our desk
Conduit, the first algorithm we run for ourselves, began as a rule the founder had traded by hand for years: when the larger timeframe is already leaning one way and the shorter timeframe stretches hard in that same direction, that stretch is usually the trend continuing rather than the trend ending, so you go with it and make price prove it before you are in. It was written down as a mechanical rule before any code existed. The code came after, and the AI wrote most of it, under direction. Then the rule went through the same gauntlet we would put anyone's idea through: it had to beat a random-entry twin with the same exits, on more than two decades of data, net of costs, and the twin has to be given every fair advantage. How Conduit was built tells that story properly, and the twin test explains the gauntlet.
Aqueduct, the newer one, is Conduit run a particular way, and it came from the same place: a human observation about which of Conduit's trades were worth taking and which were not. Again the thesis came first and the measurement came second. What Aqueduct is covers where it stands.
What we tell every steward
Do not buy an algo. Do not let an AI go looking for an edge on your behalf. Learn to trade one idea by hand until you can state it as rules, then build the rules into a tool, with AI doing the labour and you doing the leading. It is slower. It is also the only version of automation where the trader ends up wiser than they started, and it is the only version we have ever seen survive contact with a market that changed its mind.
Kingdom Portfolios LLC is an independent education publisher. It is not registered with the CFTC or the NFA, is not a Commodity Trading Advisor, does not offer or manage investments, and does not trade anyone else's capital. Nothing here is investment advice, a signal, an offer, or a solicitation.
No performance results of any kind are presented in this note and none should be inferred. Where this note refers to our lab, our harness or our measurement, it means simulated research on historical data. Hypothetical performance results have many inherent limitations: they are prepared with the benefit of hindsight, they do not involve financial risk, no hypothetical record can completely account for the impact of financial risk in actual trading, and no representation is made that any account will or is likely to achieve results similar to anything described. Past performance is not indicative of future results.
The algorithms described trade only Kingdom Portfolios' own accounts. They are not for sale, not licensed, and not run for anyone else. Trading forex and futures carries substantial risk of loss, including loss of the whole account. Education only.
Common Questions
Is all AI in trading a black box?
No. The distinction is who leads. AI that writes and tests code for a rule a human designed is a tool. AI that searches data to invent the rule is a black box, because nobody can explain what it found or why it should keep working.
What if I cannot code?
That is exactly the gap AI closes well. If you can describe your rule precisely enough that a machine could follow it, an AI can write the code under your direction. What it cannot do is supply the rule for you.
Do you sell Conduit or Aqueduct?
No. They trade only Kingdom Portfolios' own accounts. We teach how they came to exist so that stewards can build their own, cohesive with their own strategy.