I Said I Wasn’t Sure This Was Going Anywhere. Here’s Where It Went.

It has been thirty-five days since I told you I’d tell you what happened next, and in that time I have: moved my oldest into her first apartment, survived the crowds at the Wisconsin State Fair for a cream puff that took twenty-two minutes longer to acquire than it took to eat, and let a $200 automated trading experiment quietly compound in the background of my life like a crockpot I forgot was plugged in. None of those things are related. All of them are more honest an explanation for the gap than “I was busy,” which is what I’d normally say and what you’d normally not believe.

So: last time, I left you with a cliffhanger I did not mean to leave you with. I’d written, more or less verbatim, “I’m not sure this is going anywhere,” followed by a tidy little menu of four options for what to do about it: hold the rules still, simplify on purpose, accept it’s a slow story, or just quit. I promised the next update would tell you which one I picked, in real time, no smoothing it over after the fact.

Here’s the real-time part: I didn’t pick any of them. Here’s what happened instead.

The blocked candidate got un-blocked

If you read the last post, you might remember KMX, CarMax, a five-insider cluster with the CEO in it, blocked three sessions running because two of the five insiders’ purchases could only be confirmed through a news aggregator’s summary, not the actual SEC filing. The rule requires the real filing. Aggregators paraphrase; the aggregator’s paraphrase of a Form 4 is not the Form 4. I said as much, called it “the rule doing the thing I asked it to do,” and left it sitting there, blocked, mildly annoying, unresolved.

The very next session, the day after that post went up, the agent found a working path to the actual filings, verified all five insiders directly against sec.gov, and bought it. Twenty dollars, same as always. I did not plan that timing. I’d love to tell you I orchestrated a dramatic reveal for narrative purposes, but the truth is a websites-and-XML-parsers story, not a suspense story: the tool that fetches Form 4s from SEC EDGAR worked one day and hadn’t worked a few days earlier, for reasons involving a browser extension’s permission layer that I still don’t fully understand and don’t feel great about not fully understanding.

Since then, two more trades. BSX, Boston Scientific, a CEO-led cluster with a genuinely enormous dollar commitment behind it, at forty dollars instead of the usual twenty, a size I approved on purpose specifically because I wanted to see what a deliberately oversized position looked like in this format. (I want to be very clear that “deliberately oversized for blog content” is a sentence I said out loud to an AI agent managing real money, and if that doesn’t tell you something about where this project sits on the spectrum between “serious financial research” and “guy with a hobby,” nothing will.) And then, this week: AMRZ, Amrize, a building-materials company I had never heard of and would not have found on my own, where five C-suite executives bought stock in the open market five days after the company missed earnings, cut guidance, and got downgraded. Buying into the bad news, not ahead of good news. First time that’s happened in this experiment. The stock kept falling after they bought, too, hitting a fresh 52-week low the same morning I placed the order, which is either a great entry point or the first ten minutes of a cautionary tale, and I genuinely don’t know which yet. Neither does the agent. That’s sort of the point.

Right direction, wrong reason, again

Here’s a pattern that’s shown up often enough now that I think it’s actually a pattern and not just me pretending three data points is a trend line: the insider cluster keeps calling the direction correctly and getting the mechanism wrong.

UUUU, Energy Fuels, went up double digits on a quarter it missed badly, because the real driver turned out to be an acquisition closing, not the earnings print the thesis was ostensibly about. KMX went up on a strong earnings beat that had nothing to do with the CEO-led board-drama storyline I bought it for. And now AMRZ, still too fresh to score, is either about to be the third instance or the first counterexample. Three insiders with real money on the line pointed at three stocks that all moved the way they pointed. None of them moved for the reason I would have written down at the time. I don’t have a tidy conclusion about what that means yet, beyond: the signal might be less “these people know something specific” and more “these people are willing to bet on their own company when it’s genuinely cheap,” which is a real thing and a slightly less exciting thing to have discovered.

The number, honestly

Portfolio’s up about 3% against the $200 actually put in. The shadow QQQ benchmark, same dollars, same dates, just parked in an index fund instead, is down about 0.4% over the same stretch, mostly because the Nasdaq had a rough week and a half right as I’m writing this. So there’s a real gap now, for the first time. I want to undersell that appropriately: four trades, five weeks, a $200 account. This is not a track record. This is barely a trend. If you squint at the chart below you can see the exact week the gap opened up, and it says a lot more about the market’s mood than it does about my rulebook.

Answering my own question, sideways

So, the four options from last time. Did I hold the rules still? No, I kept adding to them. There’s a whole logging system now that lives in Google Drive instead of a file I kept forgetting to re-upload, a rule about excluding congressional over-traders that got formally written down instead of just applied ad hoc, a running calibration tally, a weekly postmortem field. That’s not “hold still.”

Did I simplify? Also no. If anything the bar got higher: every insider now needs a direct SEC filing, not an aggregator’s word for it, and that bar has caused real stuck weeks.

Did I quit? Obviously not, you’re reading this.

Did I let it be the slow thing it’s turned out to be? This is the closest one, except it hasn’t actually been that slow, four trades in five and a half weeks against a one-trade-per-week ceiling I set myself isn’t fast, but it isn’t the “watched paint dry” pace I was bracing for when I wrote the last post either.

So the honest answer is a fifth option I didn’t list, because I hadn’t lived it yet: keep tightening the rules and keep trading, and let those two things happen at the same rate they were always going to happen at, instead of forcing a decision about which one mattered more. Which, in retrospect, is probably how most systems that survive actually work, and I only had to build a small AI-managed brokerage account to relearn it.

What’s next

There’s a jobs report coming September 4th, and for the first time since I cut the Fed-meeting leg out of the third signal for not having a real anchor to measure against, there’s a genuine, measurable gap between what the prediction markets are pricing and what economists expect: Kalshi‘s leaning toward a softer number than the consensus estimate, and it’s a real, explainable gap, not noise. What I don’t have is a clean rule for which direction that should point a trade, because weak jobs data is the kind of thing that can mean “more rate cuts, buy stocks” or “the economy’s actually slowing, be careful” depending on which week you ask, and I built this whole rulebook specifically so I wouldn’t be the one making that call by vibes at 9pm on a Tuesday. So it’s flagged, not traded, and I’ll tell you what I decide.

There will be losing days. I said that in the first post and I meant it, and AMRZ, sitting at a fresh 52-week low as I bought it, might be the one that collects on that promise. I’m not sure this is going anywhere.

I said that last time too. Turns out it was already going somewhere. I just hadn’t looked up from the rulebook long enough to notice.