An investment process is auditable when someone who is not you, holding only your trades and your stated reasons, could determine whether your results came from your reasons or arrived despite them. Almost no retail process survives this definition, not because retail investors are stupid but because the definition requires stated reasons, written down before the outcome, and nearly nobody writes their reasons down before the outcome. I certainly did not. For five years I ran money on instinct, a trusted feed, and conviction sizing, and the result was about 4.84x per dollar against the index’s 1.85x over the same period, which is the kind of number that sounds like a defense of the process and is actually the strongest force working against ever examining it. A losing streak interrogates you for free. A winning record charges you for the privilege, and the price goes up every year it compounds.
This is the story of paying that price late, after a small drawdown rather than a catastrophic one, and of what the payment bought: two verdicts about myself that refuse to resolve into one. The machinery that came out of the audit (falsifiers, exit rules, a monitor that watches my theses while I sleep) has its own post, because some readers will want the system without the confession. This post is the confession, and I think it is the half that generalizes, because the obstacle it describes is not a gap in anyone’s spreadsheet. It is the record itself, sitting in the account of everyone who has ever been lucky in a bull market and quietly concluded they were good at this.
Five years of vibes
The record is real, which I can say now with a confidence I could not have claimed before, because a model rebuilt it from raw transaction exports rather than from my memory. Every dollar I moved into the market since 2021, simulated into the index on the same dates, becomes 1.85x; the same dollars, run through whatever it is I actually do, became 4.84x. The honest decomposition of that gap was the first thing the audit produced and the first thing my sense of myself had to absorb: roughly 94 percent of the lifetime gains came from five conviction decisions, while the dozens of other trades, the rotations and satellites and clever little swaps, netted out to approximately nothing. I was not a good trader who also made big calls. I was a maker of big calls who also traded, and the trading was a hobby the calls were paying for.
The calls had a shape. A leveraged energy fund bought in 2021 on conviction, held through a drawdown of nearly half its value, and exited at a 170 percent gain that funded a property purchase the index could not have funded; the simulation is blunt about this, showing the shadow portfolio short by a meaningful fraction of the withdrawal at the moment I needed the money. Two chipmakers bought in early 2026 on a thesis about agentic AI needing CPUs, both up well over 100 percent within three months. A memory manufacturer bought that spring on the conviction that the market was mispricing a shortage, which became the largest position I have ever held and, for a while, the best trade I have ever made. Each call was sourced the same way: a small circle of pseudonymous analysts whose previous calls had earned trust, a verification layer of my own on top (live module prices, reseller lead times, the kind of checking that feels rigorous while you are doing it), and then size, real size, the kind that makes the position the portfolio. The method worked. That sentence is true. The method was also never written down anywhere it could be defended, and that sentence is true at the same time, and this post lives in the space where both stand.
The flinch
The clearest evidence that nothing was written down arrived in April 2025, when a tariff panic that the financial press briefly treated as the end of the global trading order met a portfolio whose entire thesis existed as a feeling in my chest. I read one prestigious magazine’s coverage, became convinced the consequences were unprecedented, and sold into the bottom of a 36 percent drawdown, which is to say I sold a thesis I still believed because an eloquent narrative outcompeted it inside my own head. The thesis turned out to be fine. The narrative turned out to be noise. What failed was neither my conviction nor my information but the absence of any procedure for adjudicating between them, because a thesis that lives nowhere cannot be defended against an article that lives on good paper with good typography. The market recovered, my relative performance gave back a year of ground, and I filed the episode under bad luck, which is where unexamined processes file everything.
There is a version of this story where the flinch teaches the lesson immediately. That version did not occur, and the reason it did not occur is worth stating plainly: the very next conviction call paid off so violently that examination became unaffordable again. By mid 2026 the memory position was 63 percent of my portfolio, sitting on top of a line of credit, in a stock whose own history contains, by the audit’s cycle count, sixteen separate drawdowns of more than 40 percent, and I had labeled it “the safe bet” in my internal taxonomy because my confidence in the thesis was the highest I had ever had in anything financial. Confidence was the only axis the taxonomy measured. I was over my skis and accelerating, and some part of me knew it, because some part of me is the one that finally asked.
The ask
Here is the thing I most want to communicate, the reason this is a post rather than a private document: handing five years of trades to an intelligence and saying “check me” was genuinely hard, and the hardness had almost nothing to do with the trades. The record looked impressive. I had, on some level, enjoyed that it looked impressive, enjoyed it in the specific way you enjoy something you suspect you could not justify if pressed. Submitting it for audit meant accepting in advance that the impressiveness might dissolve under examination into luck and survivorship, that the theses I remembered holding might not match the theses I actually held, and that I might turn out to be the kind of investor my own analysis would warn a friend about. There is a reason almost nobody does this voluntarily, and it is not laziness. It is that the act is a confession, and confessions have an audience problem.
Which is where the model changed the economics, and where I think something genuinely new is happening in personal finance. Confessing a flimsy process to a human advisor costs face with a person who has opinions about you, incentives around you, and a memory that persists into every future interaction. Confessing it to the feed is impossible, because the feed is a performance and so is your presence in it. Confessing it to Fable cost nothing socially and could not be performed for, because whatever came back was going to be arithmetic before it was anything else, and arithmetic neither flatters nor judges. The disclosure barrier, the actual psychological gate that keeps winning processes unexamined for decades, dropped low enough to step over. I do not think this point is small, and I will make it as directly as I can: the largest advantage AI currently offers an individual investor is not analysis. It is that you will tell it the truth, including the truths you have not admitted to yourself, because it is the first audience in the history of money that can listen like a person and respond like a ledger.
What came back was a double verdict, delivered without ceremony in either direction. The record is real: the outperformance survives every methodological correction the model could throw at it, the conviction engine is statistically the entire business, and the property that the energy trade bought was a purchase the index could never have made. And the process could not defend itself: the audit found theses blurred in transit, labels inverted, exits absent, and a five year old habit of mistaking confidence for safety. Both verdicts arrived in the same report, neither softening the other. No human relationship I have would have delivered both halves undiluted, and I include the relationships I pay for.
What an audit finds
The findings that mattered were not numerical. They were about provenance, the question of where a thesis came from and whether the version in your head still matches the version at the source, and in five years I had never once asked it of my own positions.
I held a silicon carbide company on what I remembered as a thesis about indium phosphide substrates, which is a different material made by a different company. The audit caught the conflation in an afternoon, and the full story turned out gentler and stranger than the headline: the analyst I trusted really does hold the silicon carbide name, inside a broader thesis about subsidized Western supply chains, so the circle was right and the ticker was right, but the mechanism I remembered buying belonged to a neighboring stock in the same constellation, one which subsequently went up roughly fifty fold while I held its sibling. Somewhere between the feed and the buy button, two theses fused into one remembered reason. Nothing in my process existed to notice, because the process was my memory, and my memory, it turns out, is a lossy medium that recompresses everything toward whatever I already believe.
I held a grid storage company partly for the pleasure of being early to what I believed was a private rumor about hyperscaler contracts. The audit checked and found the rumor was a weeks old disclosure from the company’s own earnings call, public the whole time, surfaced quickly by my circle and mistaken by me for something whispered. My edge was not information asymmetry. It was attention speed, which is a real edge with real value, but a different one, with different decay and different risks, and I had been holding the position under a flattering misunderstanding of why I deserved to profit from it. The same week, a story I had absorbed about institutions accumulating my largest position near its highs dissolved under checking into two unrelated data points that headlines had fused, stale filings from far lower prices and a flow report about a different sector, glued together exactly the way my own memory had glued the substrate theses. The lesson was never that my sources were bad; mostly they were right. The lesson was that transmission corrupts, that a thesis passing from source to feed to memory to portfolio arrives mutated, and that without a written record at buy time there is no original left to check the mutation against.
The structural finding explained the position sizes, and it was the one that actually frightened me. My internal risk ladder was ordered by how confident I felt in each thesis, and risk is not that; risk is the width of the distribution of outcomes, and in my portfolio the two axes were inverted, because the position I called safest was the one whose historical busts have a median depth around 54 percent and arrive about every two years. The bet was real and well founded. The label “safe” was doing structural work it had no right to do, sizing the next decision, justifying the leverage, soothing the part of me that would otherwise have counted the exposure honestly. What do you do when the audit shows you still believe the thesis and the thesis still cannot carry the label? You stop asking the position to be safe and start asking the process to be survivable. That answer became a system, and the system became its own post.
What it took, and what it gave
I want to name the emotion precisely, because “embarrassing” is the wrong word and I spent a while thinking it was the right one. Exposing the gap between how my investing looked and how it actually worked required courage of a specific, unglamorous kind: not the courage to risk money, which I evidently have to excess, but the courage to risk a self-description that had been compounding alongside the account. The model made that courage cheaper, not because it was gentle (it was not; the phrase “the label was doing structural work” is its observation, not mine) but because it was structurally incapable of the thing I actually feared, which was an audience storing my confession for later. Whatever judgment exists in this story, I supplied it myself, reading the findings. The model just kept producing arithmetic until the arithmetic was impossible to argue with, and then, this is the part I did not expect, it kept extending the same honesty to my wins, defending the record’s reality as rigorously as it dismantled the process. Bitter medicine, but the bottle was labeled accurately, and it was taken prophylactically: after a 17 percent pullback rather than a 54 percent bust, with the leverage still comfortable, while the choices all still existed.
The growth I am claiming is narrow and I want to claim it carefully. I am not a better stock picker than I was a month ago, and the audit suggests I never needed to be. What changed is that my conviction now operates inside a structure that can survive my conviction being wrong, and that structure exists because a model made it affordable, psychologically and literally, to look at what was actually there. Whether the machine works is not yet knowable; a system built in June has survived nothing by August, and I am writing this with my largest position down 17 percent and the falsifiers mostly quiet. But the alternative had five years of data and exactly one stress test, and it failed the stress test in April 2025 for the most human reason there is: nobody can defend a thesis that was never written down. The uncomfortable conclusion is that the scarcest resource in retail investing was never information, and was never even discipline. It is a context in which you can afford to be honest about how you actually decide, and the first technology that made that context cheap has no face, which turns out to be the feature, because everything with a face was part of the problem.
Comments