Research Practice

The record includes revisions.

This is the part of the site I care about most. A decision becomes useful when you can read the original reasoning, the tradeoffs, the strongest competing view, and the evidence that arrived later—all in one place, with no convenient memory edits.

Thesis before outcome Evidence before confidence Revision on the record

Decision Journal

Three live examples of how a view gets tested.

These entries point to the original work rather than rewrite history after the fact.

01Current

Position review · AMD

A credible second source, with no room for a slow quarter.

Decision: Hold as a core position, but treat execution evidence—not the stock move—as the test. The thesis rests on AMD becoming a meaningful alternative supplier for AI accelerators while server-CPU share gains continue.

Decision frame: Evaluate the enterprise-wide tradeoff: customer demand for supply diversity, product execution, competitive response, and the valuation that already prices in success.

Disconfirming evidence: Two quarters of Data Center growth below 30% year over year, or a major customer stepping back from MI-series adoption, requires a fresh underwriting.

Read the position Read the earnings review

02Revised

Risk review · SNDK

A winning position still has a sizing problem.

Decision: Acknowledge the 10% position cap rather than create an exception for a strong performer. Returns do not erase concentration risk.

What changed: The position drifted to 10.02% of the book. The case is not that the business became less attractive; it is that the portfolio became less aligned with its written rule.

Read the risk disclosure

03Guardrail

Portfolio construction · AI infrastructure

Correlation is a risk, not confirmation.

Decision: Treat overlapping exposure across semiconductors, cloud software, compute, and power as one macro question—not a collection of unrelated wins.

What would change the view: Evidence that AI capital expenditure is slowing or failing to translate into durable demand would affect multiple positions at once.

Read the full disclosure

Post-Mortems

The process is only credible if it includes misses.

July 2026 · Leveraged exposure

Writing down an action is not the same as taking it.

Original plan

Trim NBIL after it exceeded the 5% leveraged-sleeve cap.

What happened

The position fell below the cap because the rest of the portfolio grew faster—not because the trim occurred.

Process change

A cap breach now needs a stated action date, not only an acknowledgment.

Read the full post-mortem

Open Questions

Questions that remain open on purpose.

These are research prompts, not predictions. Each has an observable evidence trail.

Can AMD convert a credible product alternative into durable, measurable accelerator share?

Are AI-infrastructure returns broadening into earnings power, or only being pulled forward by capital expenditure?

When does a concentration that began as research conviction become an unpriced portfolio dependency?

Learning in Practice

Frameworks are useful only when they change the question.

Business, entrepreneurship, and writing coursework shape the way this work is approached: business models before ticker symbols, operating constraints before narratives, and claims written clearly enough to challenge. The About page traces those connections without turning them into a transcript.

Read the research standards Read the flagship decision memo