What Aqueduct Is, and Where It Is Going
9 min read · also published as a field note
Aqueduct is the newest thing on our desk and the easiest to misunderstand, so here is the plain version. It is not a second strategy. It is Conduit, run a particular way, with a forward record kept in the open. This note is where it stands as of this week and where it goes next. No figures, by design; the record itself is gated and this is the method.
Where it came from
After Conduit survived the twin test, the obvious question was whether every one of its signals deserved to be taken. The doctrine did not change: higher-timeframe bias first, a short-timeframe stretch in that same direction second, a resting order beyond the stretch so price has to continue through it before there is a position. Aqueduct is that same continuation model. What it adds is a stricter opinion about which continuations are worth committing to.
We went back through Conduit's historical fills and looked at the three candles either side of each one, the way a trader flips through a chart the morning after. A pattern showed up that a human could see and say out loud: the candle the order filled on was itself telling us something. A fill that landed on a candle already carrying the move behaved very differently from a fill that landed on one still fighting it, and the candle just before it mattered too. That is an observation about whether the continuation is real yet, which is exactly the question a continuation model should be asking. The observation became a rule, the rule became mechanical, and the machine wrote the code. Human led, mechanically derived, AI built, the same order as always.
Confirmation, twice
Conduit already makes price prove the continuation once, by resting the order beyond the stretch instead of entering at it. Aqueduct asks a second time, at the fill itself: is this candle carrying the move, or is it still arguing with it? That means fewer trades and some good ones missed, in exchange for skipping a class of entries that history said were the poor ones. Whether that trade is worth making is exactly what the forward record exists to answer, and we are not going to pre-empt it here.
The rest of the refinements are less glamorous and matter more. The order rests for a set number of bars and then cancels itself, so a stretch that never continues is never a trade. Once a position has travelled far enough, its stop moves to break even, because on a continuation model the trades that hurt are the ones that went your way and came all the way back. Position size is mechanical rather than discretionary, a fixed fraction of the account, with a hard ceiling no single position may exceed. And the logic takes one position at a time in an instrument: no averaging in, no second bite while the first is unresolved. We tested adding to winners hard, and killed it.
What we changed this month
Four things, all of them the kind of change that makes a record more honest rather than more flattering. First, the tape: Aqueduct is now tested and run on a five-minute timeline rather than a fifteen-minute one, because the coarser tape was hiding the order in which things happened inside a bar, and that order decides whether a stop or a target was hit first. Second, the fills: every resting order now fills at the open of the bar that gaps through it, never at the level it was resting at. Third, the cost of a winning exit was being credited a spread it should have paid, which flattered every profitable trade by a tiny amount and every strategy by a large one once compounded. Fourth, the leak: the higher-timeframe bias filter Aqueduct inherits from Conduit was corrected, which reset every number we had and is why the record restarted.
There is a fifth that is really a lesson about nulls. Our random twin was not as random as we thought: a naive random walk drifts upward over long horizons for a purely mathematical reason, which was quietly handicapping the comparison. Correcting that changed what "no skill" looks like, and therefore changed what beating it means. If your null is wrong, everything measured against it is wrong in the same direction and you will never notice.
We also tested a handful of refinements that looked promising and killed every one of them, including a two-ticket variation of Aqueduct itself that we built, measured and withdrew. That story is its own note, and it matters more than the changes we kept.
Why there is more than one of these
Conduit and Aqueduct are not competitors and the second did not replace the first. They are two expressions of one doctrine, and the reason we keep both is that they are meant to run *in parallel*. A single logic on a single instrument is one opinion, expressed once. The same doctrine expressed across several instruments, and in time across more than one of the proprietary and demonstration accounts Kingdom Portfolios owns and trades for itself, is the same opinion expressed at a size those accounts can actually carry.
Those expressions are legs: one instrument each, the same doctrine on every one, each leg measured on its own history first and then assembled into the group we call the cabinet. How Conduit was built covers what we learned the hard way about tuning a leg in isolation, which is that the tuning tends to describe the past rather than the market. Aqueduct inherits that caution: the refinement it adds is applied to the doctrine, once, on every leg, rather than hand-fitted per instrument.
What the cabinet adds is a choice the individual legs cannot make. At any moment only a few positions may be open, which means the cabinet is constantly deciding which legs get the slots. It does not hand them to whatever has been winning; in our own sample recent winnings anti-predicted the next stretch while drawdown behaviour was the more persistent of the two, so the scoring is built on how a leg behaves when it is wrong. A leg that is quiet but well-behaved outranks a leg that has been hot. Instruments are also kept for coverage rather than for their curve, because a cabinet that can only act in one session spends most of the week idle, and a leg that looks mediocre alone can be the one that keeps the group awake.
That is what we mean by better-qualified selection over time rather than more of it. The slot count stays where the capital tolerance put it, and nothing here says what any of those positions will do. What changes, season by season, is the strength of the evidence behind whatever is standing in those slots.
That word *carry* is the whole governor. How many logics run, on how many instruments, in how many accounts, is not a matter of ambition. It is set by the principal capital we are willing to have exposed at once, and by three mechanical limits that sit above every individual trade: a fixed fraction of the account per position, a hard ceiling no single position may cross however good it looks, and a cap on how much of the account's margin may be committed in total. Under those limits, adding a logic, or an instrument, or another of our own accounts, does not make any single bet larger; it spreads the same tolerance across more expressions of it. That is the only honest reason we scale anything, and it is the same reasoning behind scale by accounts, not by size on the prop side.
It is also the point at which most people go wrong. Two legs that agree with each other are one bet placed twice, not two bets, so stacking only means anything when the pieces genuinely disagree about when to be in the market. We tested that directly by running a second, separate rule alongside Aqueduct and found they agreed too often to help. Knowing that is worth more than adding it would have been.
The forward record
Aqueduct now runs on a demonstration account, server-side, on its own, with a fixed starting balance and a written floor and goal. It reports once a day, at the same time, whether the floor or the goal was hit and nothing else. It does not run on anyone's laptop, it does not get nudged, and nobody presses anything. The record exists so we can be honest with ourselves about the method, not so it can be shown to anyone: no results, rates or outcomes from it are published here, and none will be while our registration with the NFA is pending. What we can publish is the method, which is what this note is. The honest state of the thing is a rule we can explain, a null we have not been able to beat it with yet, a correction already on the record, and time.
What is next
Three questions we have not answered. Whether the confirmation rule survives the forward record the way it survived the backtest. Whether the sizing that history suggested holds up with a real margin rule in front of it. And what a second logic would have to look like, in terms of when it wants to be in the market, to be worth standing beside this one rather than doubling it. These are questions for the record, not for another backtest. Everything about how Aqueduct was built is available to any steward who wants to build their own the same way: one idea, stated in a sentence, made mechanical, tested against a fair null, and watched in the open. Built, Bought, or Blackboxed is the doctrine; the Dominion Tools page is where the building happens.
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.
Where this note refers to a forward record, it means an automated demonstration account using simulated funds. Simulated trading has inherent limitations of its own and does not represent actual trading. No results, figures, rates or outcomes from that record are published on this site, and none will be published while registration with the NFA is pending.
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 Aqueduct better than Conduit?
That is the question the forward record is running to answer. In the lab, on historical data, the confirmation rule changed which trades were taken and how the losses were distributed. Whether that holds forward is unknown, which is why the record exists.
Can I follow the Aqueduct signals?
No. Aqueduct trades only our own demonstration and proprietary accounts, and its board is gated. We publish how it was built so you can build your own, not so you can copy ours.
Why do you keep more than one of these?
On our own desk, running one doctrine across several instruments lets us express the same capital tolerance more than once instead of making any single position bigger. We also found that it only did anything for us when the pieces genuinely disagreed about when to be in the market: when we ran two rules that agreed, we were placing one bet twice. That is a description of what we do and what we measured, not a recommendation for anyone else's account.