Rigorous, repeatable, and audited in public.
The same process runs behind every authored note on this site — here it is end to end, including where AI tools help and where they are deliberately kept out.
Sourcing — follow the capex, not the headlines
Ideas start from physical and financial trails: hyperscaler capex disclosures, grid interconnection queues, supplier order books, filings, and transcripts. The technology background matters here — knowing what a data center actually needs (power, cooling, networking, silicon) points at companies before they become consensus.
Business first — segment builds before price targets
Every name gets broken into its operating segments: revenue drivers, unit economics, capital intensity, and competitive position (EIC and SWOT where they earn their keep). The goal is to understand what actually moves cash flows before any valuation math runs.
Valuation — three lenses, assumptions shown
DCF, trading comps, and — where the business is a conglomerate — sum-of-the-parts with segment-level WACC and capex assumptions. The full model deck ships with every note as a downloadable PDF, because a number you can’t audit is just an opinion with decimals.
The write-up — a view you can hold me to
Each note states the thesis plainly, carries a conviction rating, and names what would change my mind. First person, no hedging into mush. TLCapital.AI is the masthead; the byline — and the accountability — is mine.
Post-mortems — the track record is public
When a call moves against me, I publish the review: what I got right, what I got wrong, and whether the thesis survives the new facts. Error-correction in the open is the most honest signal of process quality I can offer.
What the conviction ratings mean
Every authored note carries one. They are personal-judgment labels, not price targets — and they get revisited in post-mortems.
I would put meaningful personal capital behind the view; the thesis survives my strongest bear case.
The direction is right but timing, catalysts, or key assumptions carry real uncertainty.
Early or speculative work — published for the reasoning, flagged so nobody mistakes it for a table-pound.
Two lanes, one hard wall.
In authored research, AI is a research assistant, never the author. I use LLM tooling to search filings, parse transcripts, and automate data pulls — the same way I use Excel or a screener. The theses, models, judgments, and every published word are mine, and each note carries my byline.
The AI briefings are the opposite lane, and say so. A multi-agent pipeline I designed drafts short daily market briefings from primary sources; I review them before they publish. They carry a persistent "AI-generated / AI-assisted" label, no personal byline, and they never appear in the research archive. If it isn't labeled, I wrote it.
Building and operating that pipeline is itself part of the coverage: it is hard to value AI infrastructure well without ever having shipped anything on top of it.