How the analysis works
Methodology
A single pipe runs everything: reads, positions, and post-mortems. Five steps, written down so the conclusions can be argued with directly.
1. Research ingest
A curated author set surfaces structural reads. Filings and primary supply-chain disclosures land in parallel. Convergence is the bar: a structural claim only counts when a primary source confirms the same thing on the same day. Anchor posts that just react to a print get marked soft, not skipped.
2. Landscape synthesis
A rolling state-of-the-build-out file is rewritten daily. Each layer of the AI supply chain (memory, packaging, optical, neocloud, power, defense) carries a conviction level: very high, high, medium-high, medium. Conviction is the call. Position weight follows conviction, not the other way around.
The current state is on the Build-Out. The structural map is the Constraint Stack.
3. Thesis types
Every position is classified by what it actually is. Five categories do the work:
- Bottleneck: captures a specific supply constraint.
- Thematic: broader exposure to a working theme.
- Barbell: paired against a contrasting expression in the same theme.
- Reversion: a bet against a stretched move.
- Momentum: pure trend continuation, sized small.
4. Sell triggers, written in advance
Every position carries pre-declared exit conditions. When a trigger fires, the position exits, regardless of price. The five triggers, in order of frequency:
- Theme breakdown: the underlying thesis broke.
- Peer underperformance: same theme, a better expression of it appeared.
- Capital migration: capital is flowing into a better layer of the stack.
- Competitive shift: a structural competitor changed the unit economics.
- Discipline break: the operator broke a rule, not the thesis. The position exits anyway.
5. Graded debriefs
Every closed position gets a letter grade A through F on the call, not the P&L. The dollar outcome is irrelevant. The question is whether the entry, the sizing, and the exit were the right decisions given what was known. A great trade that printed by luck grades worse than a flat trade with disciplined execution. The record stands.
Receipts policy
The site publishes return percentages and position weightings. It never publishes dollar balances, share counts, account totals, or net worth. The line is bright and it does not move.
Discovery patterns
Names surface through a handful of reusable patterns. A few of the load-bearing ones:
- Customer-specific chokepoint: reverse-engineer from a hyperscaler spec to the smallest publicly-tradeable supplier of the critical component.
- “Actively working with X”: small- supplier language about qualification or co-development is a leading indicator for design wins six to nine months out.
- Stack-shift naming: when a new architectural layer is named in a filing, trace the supplier chain at every depth.
- Backlog-collateralized debt: when customer- contract-backed financing is bankable at multi-billion-per-facility scale, the supply side is the constraint, not the demand side.
- Customer-as-strategic-investor: a direct equity stake from a customer outweighs a supply agreement. The customer is internalizing supply risk.
- Macro / fundamental-confirmation divergence: when a rate-driven selloff hits a held name on the same day a demand-confirming disclosure lands, the divergence is the signal. The price is transient. The disclosure is durable.
That is the pipe. Everything published here was made by it.