Mandelbrot
There is a line often attributed to Lenin: sometimes decades pass and nothing happens, and sometimes weeks pass and decades happen. I have been thinking about this all month — not as a historical observation but as a lived experience. War arrived in my region. The portfolio drew down nearly 16% from its peak. I found myself, one evening, toggling between a missile trajectory map and a brokerage screen, wondering which one deserved more of my attention. The absurdity of that moment stayed with me.
This was also the month I finished building the system. Five conversations with Claude, each maxed out, each producing another layer of what is now a comprehensive compliance and signal framework built from scratch. I started wanting a nice overview of my portfolio. I ended up with a mirror. And the reflection has been humbling. My five-year returns tell me I would have been better off in the S&P 500. That is not an easy sentence to write. The easy lesson would be to capitulate — index everything, stop pretending. The harder lesson is the one I am trying to learn: that the data is not an indictment of the attempt, only of the process that preceded the system. Whether that distinction holds up is something only the next few years can answer.
The ground beneath
The S&P 500 fell 5.13% in March. VIX averaged 25.6 and peaked at 31.05 — the kind of sustained elevation that makes every decision feel heavier than it is. Four of seventeen primary bear signals were active. The US-Israel-Iran conflict escalated through the month, close enough to my home in Dubai that I feared for my family’s safety. Tariff policy remained unresolved. Prediction markets priced recession probability at roughly 36%.
Against this backdrop, the portfolio lost 8.02%, underperforming the S&P by 2.89 percentage points — the tax you pay for running a concentrated book through a broad selloff. Year-to-date returns sit at negative 7.15%. The drawdown from the portfolio’s peak reached 15.64% within the month. Cash stood at negative 5.9%, meaning I was already on margin. These are the conditions in which I made my decisions. Whether those decisions were good ones is a separate question from the fact that the market punished them.
What the machine saw
Overall compliance averaged 0.883 across both scored weeks — a number that sounds respectable until you examine where the weight falls. The research commandments — Frameworks First, Valuation Discipline, Accounting Skepticism, Capital Cycle Awareness — all scored a perfect 1.00. The portfolio construction and risk commandments told a different story.
Build and Use the Machine scored 0.66 and 0.65 — the lowest marks on the entire card. This is the commandment that measures whether I actually follow the signals the system generates. The irony is not lost on me: I spent weeks building a framework and then, in my first real month of operating it, scored worst on the commandment that asks whether I listen to it. Six positions had overrides this month. Six. That is not a system operating with discipline. That is a person who built a compass and then walked in whatever direction felt right.
Percentage Gains Focus scored 0.30 both weeks — the absolute lowest. I know exactly what this reflects. There is a part of me that fixates on accumulating shares rather than thinking in terms of percentage returns on invested capital. As I wrote in my own notes: investing is not accumulating. A lower price is good, but it can always go lower. You need a framework for what your price should be and how much weight the stock should have. The score of 0.30 tells me I know this intellectually but have not yet internalized it behaviorally.
Concentration Risk improved from 0.70 to 1.00 across the two weeks, a correction that suggests the first week’s allocation was drifting. Geographic/Sector Diversity held steady at 0.70 — a structural issue with a portfolio weighted toward North American technology. Product Love Is Not a Thesis sat at 0.70 both weeks, a score that rhymes uncomfortably with what happened in my position overrides.
Six overrides
Let me start with the aligned positions, because they deserve acknowledgment even if discipline is less interesting to narrate. CSU at 7.3% allocation had an ACCUMULATE signal and I made no purchases — aligned by restraint given negative cash. EXO at 5.9% had ACCUMULATE and I executed four transactions — aligned by action. MELI at 3.6%, two purchases on an ACCUMULATE signal. ADBE at 4.3%, ACCUMULATE with no transactions — again, cash constraints enforcing what discipline might not have. The REBALANCE positions — CHG, CPRT, SCZM, SSRM, UBER, IGV — sat untouched, as they should. The GETTING INTERESTING names — thirteen of them from MSFT to PBR — remained on the watchlist. The system worked where I let it work.
Now the overrides. GOOGL is my largest position at 14.0% of the portfolio, carrying a lifetime return of 93.2%. The signal was TRIM. I bought. This is the clearest case of conviction overriding the system, and I have a formal exception on file — an intentional overweight I will revisit if quality drops below B or the P&L turns negative. The exception is bounded, with an expiration date in September. I can defend this intellectually. But I notice that the highest-conviction position is also the one where I most easily grant myself permission to ignore the machine.
RMS had a HOLD signal. I made eleven transactions. Eleven. RACE, also HOLD, saw two. These are the positions covered by my one permanent exception — the only unbounded override in the system. I wrote that I am “very confident in the overall investment thesis for Hermès and Ferrari” and that “these are going to be in my portfolio until I give up investing.” I believe this. I also believe that eleven trades against a HOLD signal in a single month is not the behavior of someone building a position methodically. It is the behavior of someone drawn to the falling price, trying to get the cost basis down while the world feels like it is ending. My own reflection captured it precisely:
I am trying to get my cost basis down and at the same time increase the overall position size in terms of number of shares.
That sentence describes accumulating, not investing. The system knows the difference even when I do not.
BRBR sits at negative 48.3% lifetime return with a REVIEW signal. I made two trades. IBIT, down 37.7%, had a REVIEW signal and saw one transaction. BABA had a REBALANCE signal and I bought. BRKB, at 0% allocation (a new entry or re-entry), had a REBALANCE signal and I made three purchases. Each of these individually might be defensible. Collectively, they paint a picture of an investor who, in a drawdown month on margin, found reasons to buy in six places the system did not endorse. I also fully exited VWRA early in the month — a global ETF, 291 shares liquidated — which may have funded some of this activity.
Eleven contracts, one obsession
Eleven options trades across the month: three puts sold, five calls sold, three positions closed. The underlyings — GOOGL, TSLA, WDO, SPY, OWL — span a range from core holdings to names not even in the portfolio. TSLA is notable: it does not appear anywhere in my position list, yet it warranted an options trade. And I caught myself fixating on the outcome of a TSLA bear put spread — enough that I flagged it in my own weekly notes:
The fact that I was obsessing on outcome of TSLA bear put spread trade is something I would not like to repeat in future.
This is outcome fixation in its purest form. A tactical trade designed around probabilities became a source of anxiety because the P&L was visible in real-time. The trade may have been good. The obsession was not. Options should be the most mechanical part of the portfolio — probability in, premium out. When they become emotional, they reveal something about the investor’s state of mind that has nothing to do with options.
Bounded and permanent
Two formal compliance exceptions are on file. The first is bounded: GOOGL’s concentration override, expiring September 2026, with clear exit criteria — quality below B or P&L turning negative. This is the kind of exception a system should accommodate. High-conviction positions exist. The exception has a clock and a tripwire. It can be evaluated.
The second is permanent: the Build and Use the Machine override for RMS and RACE. No expiration. No quantitative trigger for reassessment. The language — “until I give up investing” — is the language of identity, not analysis. I admire these companies. I believe in their durability. But a permanent exception to a discipline system is a contradiction in terms. It says: the system governs everything except where I feel most strongly. And where do investors most need governance? Precisely where they feel most strongly.
Eleven trades on a HOLD signal in RMS. Two in RACE. The permanent exception did not cause this behavior, but it permitted it. The Product Love Is Not a Thesis commandment scoring 0.70 both weeks is the system trying to tell me something about the relationship between admiration and analysis. Whether I hear it is another matter.
The question that remains
I wrote in my weekly notes that I am “not following the signals religiously” and need to “be more disciplined about it.” I also wrote that I am “confident that I will be able to make better decisions this time around.” Both statements were made the same month. Both feel true. The tension between them is the whole story.
Building the system was the easy part. Five chats, some clever architecture, a set of commandments that reflect real lessons from real losses. Using the system — actually deferring to it when the bombs are falling and the prices are dropping and every instinct says buy more, buy now, this is the chance — that is something else entirely. The system scored my compliance at 88.3%. Is that good enough? The portfolio’s 15.64% drawdown and negative cash balance suggest it might not be. But I do not know yet whether the overrides were mistakes or judgments. That is the problem with a single month: you cannot separate luck from skill, process from outcome, conviction from stubbornness. You can only document what happened and watch what happens next.
Can you build a system to protect you from yourself and then trust it more than you trust yourself?
