The Barbell Architecture — Why the Middle Is Decay

This is Part 1 of two. Part 1 below lays out the architecture — what the two poles are, why each exists, and why the middle is structurally a drag. Part 2 covers sizing the two legs and the rebalancing discipline that keeps the barbell from quietly collapsing back into a balanced portfolio over time.

The wrong object

Most of the portfolio architectures the asset-management industry still sells were built for a return distribution that does not exist. Markowitz mean-variance optimization, the Sharpe Ratio, every "efficient frontier" the private bank shows you, Risk Parity, Black-Litterman, 60/40, Dividend Aristocrats, Defensive Equity tilts — every one of them assumes the relevant statistics of an equity portfolio are adequately described by its mean and variance. In a symmetric world that assumption is reasonable. In the world the data actually describes, it is empirically false.

The empirical foundation has been hardening for a decade. Bessembinder's 2018 paper showed that, of the US common stocks in the CRSP database from 1926 to 2016, the best-performing 4% accounted for the entire net wealth ever created — the other 96% collectively matched one-month Treasury bills. The 2023 follow-up extended the analysis to 64,738 stocks across 43 markets over 31 years: globally, 2.39% of firms account for 100% of the $75.66 trillion in net wealth creation; outside the US, the figure tightens to 1.41%. The full math sits in Heresy I and Heresy II. The Closelooknet reading sits in Heresy III. This piece is the operational distillation: given the evidence, what shape does a portfolio take?

One caveat worth flagging up front

The Bessembinder evidence assumes that public markets capture the bulk of equity wealth creation. The current trend points the other way. Databricks, OpenAI, Anthropic, SpaceX — all have enough private liquidity to stay private indefinitely, and their private valuation marks now exceed what the public market would currently pay. This pattern has happened before: between 2000 and 2015, post-dot-com, the IPO window narrowed sharply and the headline names of the cycle arrived late — Google in 2004, Meta in 2012, both roughly a decade after they were already category leaders. If the next decade's structural compounders stay private until they are already trillion-dollar businesses, the Bessembinder concentration story shifts: the public 1.41% will still concentrate public-market wealth, but they will not be where the marginal new wealth is being created. The argument for owning the public-market wealth distribution stays valid. The assumption that all of equity wealth creation flows through public markets needs watching, and is part of why the architecture below pairs the strategic pole with a tactical pole driven by signal rather than passive cap-weighting alone.

Why "balanced" is structurally worse than barbell

The intuition every traditional portfolio architecture leans on is that diversification across the middle of the distribution reduces risk. In a symmetric-return world that is true — a 60/40 portfolio dampens drawdowns by giving up upside in roughly equal measure, and the geometric mean of the resulting return series benefits from the variance reduction. The math is clean.

In a positively-skewed world the math inverts. The reason the strategic pole compounds is that it owns the few extreme winners by construction — Apple before the iPhone was a footnote, Nvidia before 2022 was a sub-3% position in QQQ, ASML was barely visible in MSCI Europe before the EUV cycle. The strategic pole's job is not to be smart. It is to be present when the next 1.4% of firms creates the next $30 trillion of equity wealth. Owning it requires nothing except patience. The cap-weighted machinery handles the recycling — losers shrink, winners get bigger weights, failed companies are removed by index policy without the investor having to make a single decision.

The reason the tactical pole compounds is the opposite. It captures specific structural waves while they are running — Rubin Build-Out 100 for the current AI-infrastructure cycle, HALO Growth 100 for the physical-world compounders, Euro-AI Sovereign 50 for the European sovereignty thesis, AW25 for tactical agentic exposure — and exits when the wave matures. The tactical pole's job is to add concentrated exposure to a structural rotation that the cap-weighted index will only catch years after the fact (because cap-weighting only assigns large weight to a wave's beneficiaries once they have already become large).

These two functions are structurally orthogonal. The strategic pole works through patience and breadth; the tactical pole works through signal-driven concentration on a 1-to-3-year horizon. Trying to average them — buying a mid-cap value tilt that is neither broad nor concentrated, or a balanced 60/40 that dampens both — gives the investor neither.

The strategic pole — 66 to 75%

The job of the strategic pole is to own the worldwide Wealth-Creator distribution at minimum cost. Two cap-weighted index vehicles may even be enough:

• QQQ (Invesco NASDAQ-100) — the US growth-and-tech tail, where the historical skewness has concentrated most heavily. Every US wealth creator of the last twenty years — Apple, Microsoft, Amazon, Alphabet, Nvidia, Tesla, Meta — sat inside this index long before any of them was big enough to dominate broad MSCI ACWI weights. QQQ is, in passive form, the cheapest way to be concentrated where the skewness has been.
• VEU (Vanguard FTSE All-World ex-US) — cap-weighted, ~3,700 names, the global rest. This is where Tencent, ASML, TSMC, Novo Nordisk, LVMH, Saudi Aramco sit. The non-US 1.41% of firms that captured all $30.7 trillion of non-US net wealth creation per the Bessembinder 2023 sample.

Together they cover the worldwide Wealth-Creator distribution at minimum cost. Buy and hold. Twenty-year horizon. Don't touch.

That last clause matters more than any other. The strategic pole is not where defensive tactics belong. No "the market is overvalued so we de-risk." No rebalance into bonds when the VIX spikes. No 60/40-style rebound-to-balance when winners have run hard. The point of holding the strategic pole is to remain present through every regime — the drawdowns, the corrections, the bear markets, the slow grinds — so that when the next Apple is still a $200 billion footnote inside QQQ at year 3 of its compounding cycle, the position is still there at year 8 when it has become a $2 trillion anchor.

The tactical pole — 25 to 34%

The job of the tactical pole is to capture specific structural waves while they are running. Explicit 1-to-3-year holding period. Not buy-and-hold. Driven by signal. Exited when the wave matures.

The Closelooknet reference portfolios are designed for this slot:

• Rubin Build-Out 100 — one hundred stocks across eighteen sectors of AI infrastructure equities: optical and photonics, memory and HBM, advanced packaging, power, storage, the network and platform layer. The current AI-infrastructure cycle expressed as a tradable index.
• HALO Growth 100 — twelve sectors of physical-world compounders. The structural opposite of the AI buildout — companies whose long-run economics survive because they own physical reality.
• Euro-AI Sovereign 50 — Europe's domestic AI infrastructure thesis. Fifty European-listed companies that benefit if the continent builds, hosts, trains and governs its AI capacity within its own borders.
• AW25 Agentic Winners 25 — seven sectors of agentic exposure. A tactical seismograph designed to read the rotation between cybersecurity, SaaS exposure, vertical agents and infrastructure as the agentic stack matures.

Each of these is a thesis-driven allocation onto a specific structural wave. Entered when the wave is in Dawn or Early-Ramp phase under the Money Temperature framework, exited when the wave matures into Sunset.

Within the tactical pole, individual story stocks can also be held — but only when guided by a method that survives Bessembinder. The worked example is Pattern 03 — NVIDIA Strategic Authority Investment: admit a public-equity ticker to the cohort when NVIDIA discloses a corporate equity stake or warrants (Tier 1 — Capital) or admits the company to a named NVIDIA architecture layer (Tier 3 — Architecture integration); equal weight at entry; exit when the position disappears from NVIDIA's 13F. Live cohort XIRR was +109% at the dossier's most recent print, with the headline INTC trade running +437% in seven months. The framework runs against the 54% directional-accuracy baseline of sell-side analysts and the ~6% twenty-year alpha-survival rate of active managers — single-name story exposure is admissible only where a comparable operator-authority signal is identifiable.

The middle — zero, by design

The Closelooknet reading is that the entire middle of the distribution is structurally excluded:

• No Mid-Cap Value tilt.
• No Dividend Aristocrats sleeve.
• No Risk Parity allocation.
• No Markowitz-optimized mean-variance mix.
• No Defensive Equity tilt.
• No "bond ladder ballast" on a multi-decade equity portfolio.
• No 60/40 carve-out for "diversification" sake.

Each of these is built on the symmetric-return assumption Bessembinder's data refutes. Value Investing in particular has a sharper problem — every classical Value screen (low P/E, low P/B, high dividend yield, high free-cash-flow yield at stable margins) is by construction backward-looking, and the firms that produced the world's equity wealth during the phases when they produced it violated every Value screen at once. Apple in 2009 was "expensive" by P/E. Amazon for fifteen straight years had no Value buyer in the room. The structural Value Trap is not a market anomaly. It is the design output of using a screening rule that systematically selects from the part of the distribution that produces no wealth.

How to look at your current portfolio

The operational starting point is to look at the existing holdings and classify each line by which pole it belongs to:

1. Strategic pole. Anything that is broad, cap-weighted, low-fee, and held with a 20-year horizon. QQQ, VOO, VTI, VEU, VXUS, ACWI, MSCI World total-market ETFs all qualify. If the position has been held more than three years with no intention of trimming, it lives here.
2. Tactical pole. Anything thesis-driven with an explicit exit signal. The Closelooknet reference portfolios. Sector ETFs held with a thesis (e.g. a semiconductor ETF position entered on the AI buildout call). Individual story stocks held on a structural-authority basis (Pattern 03 cohort names, or any analogous methodology with a defined exit rule). Holding period: 1-3 years.
3. Middle. Everything else. Balanced funds. Risk-parity allocations. Mid-cap value sleeves. Dividend Aristocrat baskets. Defensive equity tilts. Single-name holdings with no thesis and no exit rule. Bond positions held "for ballast" in a portfolio with a 20-year-plus horizon. Long-duration treasuries from the 2021 rotation. The 60/40 portfolio inherited from the prior advisor.

The middle is usually larger than the holder expects. The Closelooknet reading is that re-deploying middle capital into the two poles — keeping the strategic and tactical weights inside the 66-75% / 25-34% bands — is the single highest-impact structural change available to most multi-decade equity portfolios.

What this isn't

Two things worth being explicit about:

• This is published as a market diary, not investment advice. What is set out here is the architecture Closelooknet runs against — the framework, the rationale, the math. Whether, when and how to act on it depends on personal tax position, cost-basis, holding-period optimisation, family circumstances and risk tolerance, all of which call for a licensed advisor.
• This is not a claim that no other architecture has ever worked. Markowitz mean-variance was a Nobel Prize for good reason — it solved an important problem under specific assumptions. The Closelooknet reading is narrower: the assumptions Markowitz needs do not hold in the data we now have, and the architecture that follows is the one published in the Heresy III piece and operationalised here.

What's in Part 2

Part 2 takes the same architecture and works on the sizing math: how to split QQQ vs VEU within the 66-75%, how to weight the four Closelooknet indices within the tactical pole, when to add single-name Pattern 03 exposure on top of the index allocations, what the rebalancing cadence is on each pole, and the five failure modes that cause well-designed barbells to quietly collapse back into balanced portfolios over time. The architecture is the easy part. The sizing discipline is where most real-world implementations come apart.

One question worth sitting with

The Bessembinder evidence has been in the public domain for nearly a decade. The 2023 global extension is two years old. The asset-management industry has had ample time to absorb it. The fact that the dominant architectures sold to retail and institutional clients alike are still built on the symmetric-return assumption the data refutes is not, on the Closelooknet reading, an analytical failure. It is a commercial one. The fee structures, regulatory frameworks, and benchmarking conventions of the industry are all built around the centre of the distribution. A barbell with zero in the middle does not generate the trading volume, the rebalancing fees, the active-management mandates, or the benchmark-tracking discipline the industry's economics depend on. The question worth sitting with, for any reader of this piece, is: whose economics is my current portfolio architecture actually optimising for?