The Mosaic and the Long Tail

Why we at Hatchworks, helped build an internal, automated fund on public 13F filings — trading only our own capital, never the public’s — and what mosaic theory, portfolio theory, fat-tailed statistics, and behavioral psychology have to do with it.

Every quarter, the world’s most sophisticated capital allocators are legally required to tell the public exactly what they own. Form 13F is not a leak, and it isn’t an edge obtained through relationships or expensive data feeds — it is public record, filed with the SEC within 45 days of quarter-end, downloadable by anyone with an internet connection. And yet almost nobody reads it systematically. Retail investors, if they look at all, glance at a headline position or two and move on. Even professional allocators mostly skim. The information asymmetry left in modern markets is rarely about who has access to data anymore — it’s about who has the discipline to actually process it, cross-reference it, and act on it without emotion. That gap, not some hidden signal, is where LongTail lives.

What LongTail is

LongTail is an automated, research-driven internal fund — it does not raise, manage, or accept capital from the public; it trades only its own proprietary capital. It ingests 13F filings from ten institutional managers we’ve selected for a demonstrated record of concentrated, high-conviction investing — funds like Elliott, Duquesne, Whale Rock, and Lone Pine, whose public disclosures reveal genuine, differentiated views rather than closet-indexing. Every quarter, as filings land, they’re detected and parsed within seconds via **Zensider’s institutional filing engine. From there, LongTail’s research pipeline screens new and changed positions against a conviction threshold, and looks for consensus — instances where two or more of these managers have independently arrived at the same idea. Positions that clear every filter are opened systematically, in LongTail’s** own live brokerage account, sized by a fixed rules engine, monitored continuously, and exited according to a ratchet mechanism that locks in gains as they develop. Each position also gets a full AI-generated investment thesis — built on Claude — laying out the entry rationale, catalyst timeline, and risk factors, so the reasoning behind every trade is documented and visible, not just the trade itself.

Every open position, every entry price, every running profit-and-loss figure is visible in real time to anyone who logs in. LongTail is not a signal service and does not give advice: it reports, in real time, on decisions it has already made in its own account.

Why we built it

LongTail did not begin as a hedge-fund experiment. It began as an attempt to answer a narrower question: could a fully systematic process — no discretion, no emotion, hard rules for entry and exit — find a persistent, small edge in a market that was mostly noise? The first version of that process was built for prediction markets: low-priced, thinly-traded contracts where the market’s probability estimate often lagged the real-world facts. The specific market turned out to be the wrong one. But the machinery built to trade it — a scoring engine, position-sizing rules, a ratchet that protects gains without capping upside, and above all the discipline to let a system make decisions a human would normally second-guess — turned out to be the right machinery for a much deeper and more durable opportunity: the public disclosures of the world’s best investors.

The name LongTail is a nod to what both experiments were actually hunting for. Statistically, a “long tail” describes a distribution where extreme, low-probability outcomes matter disproportionately — most of the return in almost every real-world portfolio comes from a small number of very large winners, not from the average trade. That is the shape of the opportunity this was built to find.

Theory I

Mosaic theory

There is a formal name for what serious fundamental analysts have always done, and it is the closest thing to a legal doctrine that fundamental research has: mosaic theory. The idea, developed largely through decades of SEC and industry guidance on insider trading, is that an analyst may lawfully combine many individually immaterial, public or non-material pieces of information — a filing here, a supplier’s earnings call there, a hiring pattern, a patent filing — into a mosaic that produces a genuinely differentiated, non-public conclusion, even though no single input was itself material or non-public. It’s the difference between having an edge and having inside information.

13F aggregation is mosaic theory in its purest, most literal form. No individual filing is a secret; the SEC publishes all of them. But cross-referencing ten managers’ filings simultaneously, quarter after quarter, and surfacing the handful of names where independent conviction is converging — that is a mosaic no single filing reveals on its own, and it is a mosaic that no human analyst can build exhaustively, quarter after quarter, across a full universe of tickers, without eventually cutting corners. LongTail’s contribution isn’t the underlying idea; mosaic-building is as old as fundamental research itself. Its contribution is doing the mosaic exhaustively, mechanically, and continuously, at a scale and speed no analyst desk can sustain by hand.

Theory II

Portfolio theory, and where we diverge

Modern portfolio theory, in its classical Markowitz form, treats diversification as close to a free lunch: for any level of expected return, there is a portfolio that minimizes variance by holding many imperfectly-correlated assets. It’s a powerful and largely correct insight, and it underlies almost every index fund and target-date product sold today. It is also, deliberately, not what LongTail does.

LongTail is closer to the opposite instinct: rather than diversifying away conviction, it concentrates on it. A position only qualifies for entry when a genuinely skilled investor — or better, multiple independent skilled investors — has expressed real conviction through a real capital commitment. That is a bet that concentrated information, not statistical diversification, is the primary source of edge here. The tradeoff is real: a concentrated, conviction-following portfolio will be more volatile than a diversified index, and it will occasionally be spectacularly wrong alongside the manager it is following.

Rather than manage that volatility with a covariance matrix, LongTail manages it with a simpler, rules-based mechanism: a ratchet. As a position’s gain crosses defined thresholds, its stop-loss floor rises with it — never capping the upside of a genuine long-tail winner, but mechanically locking in profit as conviction is validated by price. It is a cruder tool than mean-variance optimization, but it requires no assumptions about correlations that tend to break down exactly when you need them most — in a crisis, when everything correlates to one.

There is a second, quieter lever here beyond simply detecting consensus: how much weight to put behind different levels of conviction. Positions are sized into tiers — multi-fund consensus, single-manager high conviction, and a baseline standard tier — and our own testing shows that weighting genuinely matters, in a way that isn’t obvious in advance. Counterintuitively, the tier built around a single manager’s outsized conviction has performed the weakest per position in testing; multi-fund consensus and the baseline tier have performed comparably well individually, but consensus produces roughly five times more qualifying opportunities. Most of the strategy’s return comes from deploying capital across that wider set of consensus opportunities, not from any one tier being a dramatically better stock-picker. Exactly how much relative weight to place behind each tier is itself a proprietary, continuously-tested decision — a second, distinct source of edge, separate from the act of finding consensus in the first place.

Theory III

Statistics, fat tails, and the problem of noise

Classical finance leans heavily on the assumption that returns are approximately normally distributed — a bell curve, thin tails, extreme moves vanishingly rare. Benoit Mandelbrot showed, as early as the 1960s, that real market returns don’t behave this way; they exhibit fat tails, where extreme moves happen far more often than a normal distribution would predict. Nassim Taleb later popularized the practical consequence: in a fat-tailed world, most of the outcome — good or bad — is concentrated in a small number of extreme events, not the average day.

Our own backtesting bears this out uncomfortably well: across every historical window we’ve tested, a small handful of trades account for the overwhelming majority of simulated profit, while the median trade is far more modest — sometimes barely positive. That is not a flaw in the strategy; it is the signature of the phenomenon the strategy is trying to harvest. But it raises a real statistical problem: with a limited number of managers, a limited number of positions, and a strategy whose returns are dominated by outliers, how do you distinguish genuine skill from a handful of lucky trades — especially once you’ve searched across many possible rule variations to find the best-looking one? That is the multiple-comparisons problem, and it is precisely why LongTail requires cross-fund consensus rather than following any single manager: agreement between independent, differently-motivated investors is a cheap and effective way to filter noise, the same way ensembling several weak models reduces the variance of any one of them.

What the backtests actually show — and why we don’t fully trust them

We ran the consensus configuration’s exact rule-set against eleven years of real filing and price data, with every position sized as a percentage of the current portfolio value rather than a flat starting balance — so gains compound the way they actually would in a live account.

Compounding, not luck on any single year, drives most of the gap — the strategy outgrew a flat-sizing simulation of the same trades by several hundred points simply by letting winners increase what the next position could be sized at.

Read plainly: over the full eleven-year window, properly compounded, the strategy beat the index by a wide margin on both absolute return and, on this dataset, without a deeper drawdown than the market’s own worst declines. That is a genuinely strong result — and also exactly the kind of number that should make a careful reader skeptical, not more confident. A large share of it is concentrated in a small number of extreme winners: Shopify, a Whale Rock conviction held for roughly six and a half years, returned over 1,800% on its own. AppLovin and Nvidia, also Whale Rock positions, each compounded past 500% in under two years. Microsoft, a Duquesne holding, added over 400% across four and a half years. Those four positions alone account for a disproportionate share of the entire eleven-year result.

None of these were selected because the system somehow knew in advance they’d be the outliers — the consensus and conviction filters only required that a real manager had committed real capital to the idea at entry. What let the returns actually compound to these levels afterward was the exit rule, not the entry rule. An earlier version of the system used a hard take-profit target, and testing it against the same historical data showed it quietly destroying returns by capping exactly this kind of position right as it started to run. The ratchet-only exit that replaced it — raise the floor, never cap the ceiling — exists specifically so that when a long-tail winner shows up, the selection process doesn’t get in its own way. That is the fat-tail signature again, showing up directly in our own numbers, for better and for worse.

We’re not going to pretend this settles anything, because it doesn’t. Backtests are simulations against the past, built and refined using the same historical data they are then evaluated on. Every choice along the way — which funds to track, which conviction threshold to use, whether to filter out positions that already ran up before entry — was tuned, at least partly, with knowledge of how it would have performed historically. That is a well-known trap; the technical term is overfitting, and no honest backtest is fully immune to it, including ours. A backtest can tell you a strategy isn’t obviously broken. It cannot tell you it will work going forward, and treating it as if it can is one of the more common and costly mistakes in systematic investing. (Hypothetical and backtested results have inherent limitations, do not represent actual trading, and are not indicative of future performance.)

This is exactly why LongTail exists as a live experiment rather than a backtest report. Every position discussed above is being traded, right now, with real capital, in a live brokerage account you can watch directly — no cherry-picked window, no rule tuned after the fact.

Watch the live account in real time

The efficient-market question, honestly addressed

The most serious objection to everything above is also the oldest one in finance: the semi-strong form of the efficient market hypothesis holds that all publicly available information — including 13F filings — should already be reflected in prices the moment it becomes public. If that is fully true, there should be nothing left to harvest here at all.

We think the honest answer is: mostly true, with real cracks at the margins. Filings are public, but they arrive with a built-in 45-day lag, they are filed in a genuinely tedious legal format, and the base rate of anyone actually cross-referencing ten managers simultaneously, in real time, quarter after quarter, is low. Institutional capital also carries behavioral drag that has nothing to do with information: mandate constraints, career risk around concentrated bets, window-dressing near quarter-end, and plain organizational inertia. None of these frictions are large. None of them will make anyone rich overnight. But a market being efficient does not mean it is perfectly, instantly efficient at every margin — and the academic literature on 13F-based “copycat” strategies has, with real inconsistency across studies and time periods, found small residual effects at exactly these margins for decades. LongTail’s bet is that systematic, unemotional, exhaustive processing of exactly this kind of information is a legitimate place to look for a small, real edge — not a loophole in the theory, but a byproduct of how expensive genuine diligence still is, even now.

Psychology: why systematic beats discretionary here

Even a genuinely sound strategy is easy to sabotage with human judgment. Behavioral finance has a long list of well-documented ways investors damage their own returns: overconfidence in one’s own analysis, the disposition effect (selling winners too early and holding losers too long — exactly backwards from what a ratchet does), herding into whatever is already popular, and narrative bias — falling in love with a story about why a position “has to” work, past the point where the evidence supports it.

The entire point of LongTail’s rules engine is to remove the moments where those instincts would normally intervene. There is no discretionary override on entry: if a position clears the conviction and consensus thresholds, it is opened. There is no discretionary override on exit: the ratchet fires mechanically, without asking whether the investor “feels good” about the position. This is not a claim that systems are smarter than people — often they are not. It is a narrower, more defensible claim: a consistent, pre-committed set of rules, applied without exception, removes a specific and well-documented source of return-destroying error. That is a psychological firewall as much as it is an investment process.

The goal

LongTail’s goal is not to replace human judgment in investing, and it is not to sell signals. It is to make institutional-grade mosaic research — the kind that has historically lived almost exclusively inside hedge funds and family offices — visible, in real time, to anyone who wants to watch it happen. Every position, every entry, every exit, and every piece of AI-generated reasoning behind it sits in the open, on a live account, with real capital, reported as it happens rather than after the fact.

We are not a registered investment adviser and nothing here is investment advice; LongTail’s decisions are its own, made systematically in its own account, and nobody places a trade through this site. What we are offering instead is transparency about a specific, testable question: can a disciplined, unemotional, exhaustively-mechanized reading of what the smartest institutional money in the world is already required to disclose produce something real, over time, in public view.

Close

Ten managers file quarterly. The rules are public. The math is public. The only genuinely scarce resource left is the discipline to read all of it, all the time, without flinching. That is the long tail we built LongTail to find.

LongTail is not a registered investment adviser, broker-dealer, or financial planner. Nothing in this article constitutes investment advice or a recommendation to buy or sell any security. All trading decisions are made systematically by LongTail in its own account; no reader ever places a trade through this content. Hypothetical and backtested performance results have inherent limitations, do not represent actual trading, and are not indicative of future results. Past performance is not a guarantee of future returns. Investing involves risk, including the possible loss of principal.