Prices fell. The evidence didn't.
In late July 2026, AI stocks sold off hard. My screen was red for days. And when your screen is red for days, one question shows up whether you invite it or not: do I need to do something?
Most people answer that question by reading more news. The problem is that most of what's in the market is noise. You can have a thesis that's fully intact, hear enough noisy signal, and sell anyway. I've done it. The times I've lost money were almost never because my thinking was wrong — they were because I stopped trusting my thinking at exactly the wrong moment.
So I built a system for myself. For the past year I've been building Portfolio OS. It records the reasoning behind every position I own — the actual beliefs, ranked, and the specific conditions that would make me sell. Every morning it reads my sources and checks what happened in the world against what I believe. During the eight days of the selloff, it read 227 developments.
Here's what the evidence actually said. Alphabet raised capex guidance to $195–205 billion. Google Cloud grew 82%. Supermicro reported roughly $60 billion in orders. Celestica beat expectations, raised guidance — and its stock fell anyway. Think about that one for a second: a company told the market that business is better than expected, and the market paid less for it. The companies serving AI infrastructure weren't reporting weaker demand. They were reporting more.
Prices and evidence are two different things. Prices tell you what other people will pay today. Evidence tells you whether your understanding of the world has changed. Usually they move together. That week they didn't.
The honest read isn't "nothing changed," though. One thing did change — and it wasn't the thing everyone was afraid of. The new information was about how the buildout is being financed. Nvidia discussing guarantees on roughly $250 billion of OpenAI financing. Meta moving $30 billion of data-center financing off its balance sheet. Oracle downgraded as investors questioned how aggressively it's funding AI infrastructure. Demand held. The question that actually moved is whether the money behind the demand is getting fragile. Different belief. Different conversation.
Why could I see this? Not because the system predicts anything — it doesn't. It's because my reasons were written down before the drop. When your thesis lives only in your head, a red screen edits it in real time. Fear rewrites memory. When your reasons are written down, you get to ask a precise question instead of a vague one. Not "should I be worried?" — "which of my stated beliefs did this week actually touch?"
That week the answer was: prices moved a lot, my evidence moved a little, and where it moved, nobody was panicking yet. So I did the hardest thing in investing, and for once I did it with confidence instead of anxiety.
Nothing.
Sometimes the most important thing you do is hold.