Research · June 5, 2026 · 6 min read
How our machines read a supply chain shock
When something breaks in the semiconductor supply chain — an export control, a fab outage, a surprise capex cut — the price reaction is nearly instant. The explanation usually arrives days later, in sell-side notes assembled by analysts working through the same filings everyone has had all along. The information was public; the bottleneck was reading speed.
That bottleneck is exactly what we built our research stack to remove. Our models don't watch tickers. They read primary sources — SEC EDGAR for US issuers, DART for Korea, TWSE/MOPS for Taiwan — and maintain a live, structured thesis on every vendor in the chain: what they make, who supplies them, who buys from them, and what has to be true for their earnings to hold up.
Tracing the shock, hop by hop
Consider a tightening in high-bandwidth memory. A screen tells you memory makers are up. Our system instead asks the question a careful analyst would: who sits one hop away? Packaging and test capacity gets scarcer; GPU vendors face a new bill-of-materials constraint; the power and cooling vendors serving datacenter buildouts see their backlog quality improve. Each hop is a claim about a real commercial relationship, and each claim is written down with the evidence that supports it.
The output is not a sentiment score. It is a chain of reasoning — supplier to customer, with direction, magnitude, and confidence at every link — that a human can audit and challenge. When the model is wrong, we can see exactly which hop failed, fix the context, and re-run it.
Why auditability is the product
Most machine-generated research fails in the same place: it produces conclusions without showing its work, so when markets disagree, there is nothing to learn from. We took the opposite constraint. Every thesis in our dashboards traces back to filings and disclosed relationships, every revision is logged, and subscribers can edit the context the model reasons over.
That discipline is what lets the same system run our fund and serve our research clients. The dashboards you can subscribe to are not a marketing window — they are the working surface of the process we trade on.