Opinion · July 2026
The Book's Four Largest Semiconductor Positions Are All “Medium” Conviction
Building the Coverage Book meant writing the same eight fields for every position: business description, thesis, bull case, base case, bear case, key risks, catalysts, valuation, and a conviction rating. Laid out side by side like that, a pattern showed up that was easy to miss reading the positions one at a time in separate letters: SNDK, AMD, MU, and INTC — the largest position in the book and two of the next seven largest — are all rated Medium conviction, not High. Together with ARM, that's 29.78% of the entire portfolio sitting in Medium-conviction semiconductor names. High conviction is real weight in this book too — 43.46% of it — but it isn't concentrated in the single largest position. That's worth sitting with rather than smoothing over.
What "Medium" is actually doing in these five ratings
Conviction here isn't a computed score; it's my own subjective read of how much I'd bet on the thesis being right, stated plainly on every Coverage Book entry rather than hidden behind a single "buy" label. Medium, on the semiconductor names specifically, means the same thing five separate times: the demand side of the thesis is real and I believe it, but the supply side is out of my control and historically brutal. SNDK's own entry says it directly — sizing "reflects a high-conviction cyclical call," but the rating stays Medium because NAND pricing has crashed this hard before, most recently in 2022–23, on nothing more than competitors adding capacity at the wrong time. AMD, MU, ARM, and INTC carry versions of the same caveat: a real, working thesis with a ceiling on how sure anyone can be about a commodity-priced, capital-intensive industry three competitors also get a vote in.
Sizing answered a different question than conviction did
SNDK is the largest position in the book at 9.64% not because it's the highest-conviction idea — META, SPY, VOO, and PANW all outrank it on conviction while sitting smaller or comparably sized — but because it's up 230.97% on cost and the IPS doesn't force a trim until a position actually breaches its cap. That's a mechanical, price-driven answer to "how big is this position," sitting next to a deliberately qualified answer to "how sure am I about it." Those are two different questions, and this book currently answers them differently for its single largest holding. The Compliance Ledger's own account of this — the position drifting back under its 10% cap on a price pullback rather than a trim I said was owed since June — is the same tension from a different angle: sizing in this book has been more responsive to price than to conviction, and the Coverage Book is what made that legible across five names at once instead of one letter at a time.
What I'm actually going to do about this
Nothing dramatic, and that's worth stating honestly rather than promising a rebalance I might not execute. I'm not selling a Medium-conviction position just because it's large — the thesis hasn't broken, and the IPS is explicit that a working thesis outrunning its cap gets trimmed back to policy, not exited. What changes is smaller: any future addition to SNDK, AMD, MU, INTC, or ARM now has to clear a higher bar than "the sector is still working," because the book is already carrying 29.78% of itself in names I've rated one notch below full conviction. Growing that further on momentum alone would be sizing answering the conviction question by default, which is exactly the pattern this piece is naming.