This website uses cookies

Read our Privacy policy and Terms of use for more information.

Systematic Portfolio & Signals · Signals before every US open

Four proprietary models read the market every morning. The portfolio does what they say.

Every name is read by the model line-up that earned its place on it. A name reaches the portfolio only when it is long and three of the four agree.

Every trading day you see where each name stands and what changed overnight. Each week those readings become a portfolio of ten names or fewer, with an exact weight on each and cash for the rest.

It runs the same way on a good week and a bad one, and nothing in it is overruled by hand.

TWO UNIVERSES · MACRO & MEGACAPS · COMMODITIES, EACH WITH ITS OWN PORTFOLIO
01What a morning looks like

Both halves of the same morning, on one page.

On the left, where the models stand on each name. On the right, the portfolio those signals produced. Every holding traces back to a line in the table, and the code will not let the two disagree.

THE SIGNALSPUBLISHED 2026-08-04
TickerNamePositionTierConsensusIn the portfolio
HYGHigh-yield creditLONGBroad · Long15.0%
LMTLockheed MartinLONGBroad · Long15.0%
NVDANvidiaLONGBroad · Long10.4%
OKTAOktaLONGBroad · Long5.7%
QQQNasdaq 100FLATMixed · FlatGate, not eligible
CDNSCadenceFLATflipped ▼Mixed · FlatTurned flat, not held
MACRO & MEGACAPS · DELIVERED BEFORE THE US OPEN · MEMBERS SEE EVERY NAME, WITH WRITTEN COMMENTARY

Why QQQ is absent. Only two of the four back it. Being long on the signals page is not enough, three of four have to agree. Why the 2× badges do not matter here. HYG and OKTA carry the leveraged tier on the signals page and both sit in the portfolio at 1×. The portfolio is never levered.

THE SYSTEMATIC PORTFOLIOTARGET PORTFOLIO · 2026-08-03
Macro & Megacaps, construction10 holdings · 79.1% invested
79.1%INVESTED
10 holdings79.1%Cash20.9%

Ten wedges, shaded by weight. Cash sits outside them, because it is a position here.

EVERY HOLDING, AS A SHARE OF THE INVESTED PORTFOLIODASHED LINE = 15% CAP
HYG15.0%
LMT15.0%
XLF15.0%
EEM11.7%
NVDA10.4%
IGV10.0%
ANET6.2%
OKTA5.7%
LRCX5.6%
SNOW5.4%

Three names sit exactly on the cap, and that is the cap working. Sizing wanted to give HYG, LMT and XLF more, and the cap said no.

The cash is the interesting number. The 20.9% is deliberate. When the names get jumpy, the portfolio holds less of them. Across the simulated decade it sat in cash 32% of the time and averaged 8.4 holdings.

POSITION AND TIER ARE SEPARATE

Position is long or flat, never short. Tier says how hard, 1× or 2×, and it lives on the signals page only. The portfolio is long-only and unlevered, so the tier never reaches it.

FOUR CONSENSUS BANDS, NOT A SCORE

How many of the four back the call. Broad is three or more, Mixed is fewer. Three models agreeing a name should be flat reads just as strongly as three agreeing it should be long.

WHY A POSITION CAN OUTLAST ITS CONSENSUS

The badge shows where the models stand today, which can differ from what is held. A short minimum-hold keeps the portfolio from flipping on one day’s wobble. We publish both.

THE COLUMN THAT MATTERS

In the portfolio is the column nobody else publishes. It links a signal to a weight, every day. Both panels are real: signals from 4 August 2026, weights from 3 August.

02Why four models, and what it changes

Every model is built for one kind of market. It still has to trade through all the others.

Most systematic portfolios run off one thing, usually momentum. Every approach has conditions it loves and conditions it hates. In the good ones it looks brilliant, and in the bad ones it bleeds slowly.

The turn arrives without a warning
One model can print for two years, then hand back eighteen months of gains in six weeks. The market underneath it changed, and nothing in the output says so.
A rough patch and a broken model look the same
Both are losses in a row. Whether you stay with it becomes a question of nerve, on the days nerve is shortest.
Most of the damage is human
The system was usually fine. It got abandoned in the third bad month, or left running through a market it was never built for.

You cannot diagnose a model from inside it. A second opinion only helps if it is built to be wrong about different things. That is what the other three are for.

SO THE ANSWER IS FOUR, BUILT TO FAIL DIFFERENTLY
MMachine learning
WHAT IT READS
The name’s own price, plus what is happening around it: bonds, the dollar, credit.
WHAT IT IS GOOD AT
Connections across markets that nobody would think to look for.
WHERE IT FAILS
It learns from history, so a move that runs further than history allows makes it cautious too early. In steady megacap uptrends it has sat out for long stretches.
NNeural network
WHAT IT READS
The same evidence, learned a completely different way: recurring shapes, short and long.
WHAT IT IS GOOD AT
Catching turns the first model reads as noise.
WHERE IT FAILS
It reads the same raw evidence as the machine-learning model. When that evidence misleads, both get misled together.
TTrend
WHAT IT READS
Price, and nothing else.
WHAT IT IS GOOD AT
Staying with a real move for as long as it lasts, with no cleverness about the top.
WHERE IT FAILS
It is always late at the turn, and in a choppy range it gets whipsawed. It can never tell you why.
RMarket regime
WHAT IT READS
The market’s underlying state: risk-on, risk-off, or neither. It reads the season, not the weather.
WHAT IT IS GOOD AT
Knowing the weather has turned before individual names show it.
WHERE IT FAILS
It is about the market, not the name. It can be right about the season and wrong about one tree.
THE PART THAT MATTERS

No pilot flies on one instrument.

An altimeter, an airspeed indicator, an artificial horizon. Any of them can fail quietly, and each is convincing while it does. Pilots cross-check, and when the instruments disagree they slow down.

Same here. A name reaches the portfolio only when it is long and at least three of the four agree. When agreement thins below three, the position closes and the money waits in cash. Disagreement is worth seeing, so we publish all four every day.

AND THEY ARE NOT ALL ASKED THE SAME QUESTION

Letting all four vote equally on everything would be tidy, and lazy. Every asset gets its own line-up, tested alone and in every combination on data the models never saw. On one name that line-up is a single model, on another it is three of them. The daily page still shows all four, even when only some are driving the call.

AND WHAT THAT CHANGES FOR YOU

We never enjoy closing a position we have defended out loud.

WHEN THE VIEW IS YOURS
  • A losing position gets held too long, because selling it concedes the point.
  • A position that is working gets trimmed early, because banking it settles the argument.
  • The review question has to be survived.
WHEN THE VIEW BELONGS TO THE MODELS
  • The exit fires the day the models fall out of agreement. It asks nobody.
  • A position runs for as long as the agreement holds.
  • The review question has a dated answer, written before the outcome was known.

You go from forecaster to observer. What you watch is which models agree this morning and how broadly, and that question has an answer, published daily and checkable later.

03How the portfolio is built

Four stages, and none of them need a view on the day.

It means one week can be compared to the last, and it means you can check the work.

01
BEFORE ANY SIGNAL
Set up once

What goes in, and which models each name gets

  • The list of names is fixed and published. No late additions, no survivor bias.
  • Every name gets the model line-up that won on ten years of data it had never seen, so the line-up differs from one name to the next.
WHAT COMES OUT →A published universe, and a model line-up chosen on evidence.
02
EVERY MORNING
Before the US open

The four models read every name

  • Machine learning, neural, trend and regime.
  • Where they disagree, we print the disagreement instead of smoothing it away.
WHAT COMES OUT →One locked signal per name, frozen the moment it sends.
03
EVERY MONDAY
The portfolio

Signals become weights

  • A name is eligible only if the published record already has it as long, and only if three of the four models back it.
  • Up to ten names, each sized so no single one decides the week, with the rest left in cash. Cash is a position here, not a leftover.
WHAT COMES OUT →Tickers and target weights, published in full.
04
AFTERWARDS
Every week and every day

The part that decides the outcome

  • The portfolio is rebuilt weekly. Weight drifts under 5% in between are left alone.
  • The sell check runs every day. A holding goes the same day the record turns it flat, or the day agreement drops below three of four.
WHAT COMES OUT →A dated record that cannot be rewritten.

Long only, never levered. Every rule and every number is set in advance. If you want the full construction, the caps, the sizing maths and the six things that can go wrong, it is all in the inception guide, free to read.

04The two portfolios

Two universes, built independently, and one of them read at two depths.

Each ships both halves: the daily signals, and the portfolio built from them. Same construction, different universes, so their bad days rarely line up. The lists only grow, and new tickers are added every month.

THE BROAD ONE · 70+ NAMES

Macro & Megacaps

MACRO · RATES · MEGACAPS · CRYPTO · RAILS
US indices, rates, the dollar, credit and the sector ETFs, the Mag-7 and 40 megacaps behind AI, plus the crypto majors and the payment networks between them. Crypto is read inside the universe, not sold separately.
TWO DEPTHS, SAME PORTFOLIO
Signals$3,000
Signals & Research$5,000

Research adds the macro note twice a week and written deep dives on the names the portfolio holds.

Explore →
THE STEADY ONE · 24 NAMES

Commodities

METALS · ENERGY · AGRICULTURE · URANIUM
The complex end to end, plus the producers, read as one board. It is the portfolio that most often moves the other way when equities fall, which is the case for owning it. The point of it is the difference, not the headline return.
ONE DEPTH, TWO PRICES
Bought on its own$3,000
Added to either MM tier$2,000

A third off for anyone already on a Macro & Megacaps tier. The code is provided privately.

Explore →

Every subscription includes its portfolio at no extra cost. There has never been a separate fee for it, and every coverage addition is included for as long as you stay. The price is locked, the universe is not. Prices exclude VAT and any other applicable taxes.

05The record

A decade of walk-forward testing stands behind every position.

The testing sits on each portfolio’s own page, at whichever horizon you want, on the same locked construction that runs today. The signals have published since late June 2026 and the portfolios started trading on 3 August 2026, so the live series is young and is labelled that way wherever it appears.

LATE JUNE 2026
The signals went live
Publish-locked from day one, each day’s calls frozen the moment they are sent, and readable ever after.
3 AUGUST 2026
The portfolios started trading
One per service. The P&L series begins here, at zero, and is added to every Monday. Month one will be described as one month of data.
2016 to 2026
Everything before that line
A decade of walk-forward simulation on the same locked construction, net of costs, labelled as simulation, everywhere it appears.
EACH PORTFOLIO’S WALK-FORWARD BACKTEST, ON ITS OWN PAGE

Walk-forward means every point used only the data available on that day, with no re-tuning afterwards. Each portfolio page carries its own panel, so the figures live in one place and cannot drift.

SIMULATED AND LIVE CAN NEVER BLUR

The platform holds them in separate tagged series and the P&L endpoint will not merge them. A backtest cannot be dressed up as a track record here, because the data layer will not produce one.

The full record →
06What to watch out for

The five things I would want to know before holding this.

A backtest is a backtest. These are the awkward parts, said plainly.

Cash is a deliberate lever, and it can be a large one.
The portfolio is long-only and unlevered, so cash is how it dials risk. The invested share is scaled to hold volatility near target, and in a downturn fewer names qualify for selection, so the cash balance rises by itself. Expect stretches where a large part of the portfolio sits in cash by design.
It runs more concentrated than you may be used to.
Ten holdings, and often fewer, because three of four cannot always agree on ten names. Macro & Megacaps averaged 8.4 in the backtest. The cap stops any one name taking over: 15% of the invested portfolio at ten holdings, 25% ever.
The turnover is not tax-friendly.
The portfolios rebuild weekly and sell mid-week, which in a taxable account costs real money. Trading costs are modelled at 5bp per side and already deducted here, roughly 0.9% a year. Tax is yours and is not modelled anywhere.
Do I place the trades myself? Could you simply be wrong?
Yes, and yes. The portfolio is published as a reference: tickers, weights, reasoning. It is not a managed account and not advice. Four models agreeing is a higher bar than one, and it is not a guarantee. There will be months when all four are wrong together.
None of this is investment advice.
Everything published here is for information and education. You are getting access to the work I have built and read myself, in the form I read it. It is not personalised advice, not a managed account, and not a recommendation to buy or sell any security. I do not know your circumstances, your tax position or your tolerance for risk, and I take no responsibility for the trades you place or the outcomes you get. Capital is at risk, and past and simulated performance tell you nothing certain about the future.
07Read the methodology

Read the methodology.

This page summarises two notes published in full, to everyone, before you got here. Both are open, no paywall, no card. If you are going to disagree with the construction, disagree with the actual thing.

The order matters. The signals note is step one, the inception guide is step two. Read them that way and everything on this page should be checkable.

08Who this is not for

The method suits some people badly, and it is better to say so here.

None of these is a caveat tacked on at the bottom. Each one is a design decision, and to the wrong reader it will look like a flaw.

01
If you need an entry, a stop and a target.
We publish target weights, not fills. The exit is a condition, so a holding leaves when the record turns it flat or when agreement drops below three of four.
02
If a cash balance reads as indecision.
These portfolios sit in cash whenever three of four cannot agree, which was about a third of the time across the simulated decade.
03
If you want the models overruled when they look wrong.
They are not overruled, by anyone, including me. That is the only reason the record means anything, and it means there will be positions here you would not have taken.
04
If you need a long live record before you start.
Fair enough, and it does not exist yet. The portfolios began trading on 3 August 2026, and everything before that line is simulation. Waiting is a perfectly good answer.
AND WHO IT IS FOR

Someone who runs capital to a professional standard and wants the same work on the hard mornings as the easy ones. The record is open.

09  ·  BEFORE YOU DECIDE

Read a month of it before you decide anything.

The portfolio publishes every trading day before the US open. It holds ten names at most, each sized so no single one decides the year, and the rules are enforced in code. Annual plans carry a free first month, so you can run a month of it against a live market and cancel in-month having paid nothing.

The work is closer to observation than prediction. None of it asks you to be attached to a particular trade, mine included. The market is the thing being read, and we are watching it from the same side of the glass.

Published for information and education. It is not personalised advice, not a managed account, and not a recommendation to buy or sell any security. Capital is at risk, and past and simulated performance tell you nothing certain about the future.