The Conviction Play
Underwriting, not promotion.
A journal about individual companies, written the way an underwriter would write it: state the question the market is actually arguing about, put the bear case at full strength, and name in advance the evidence that would prove the whole thing wrong.
Each piece is an assessment of what is known on the day it is written. Company reports bring genuinely new information, and a view that holds this quarter may not hold the next. Nothing here is a prediction, and nothing here updates itself.
What makes it different
The bear case gets stronger where the conviction is highest.
The single finding from reviewing the first twelve pieces: writing that read as promotion had a weak bear case, and the bear case got weakest exactly where the author was most bullish. So the rule inverted. If the case against can't be built, the risk isn't understood yet — and the piece isn't finished.
Every falsifier carries a date.
"What would change my mind" is not an honest section until it ends with "…and here is when we'll know." Each piece names the conditions that would break the thesis and the scheduled event that will test them.
One company, one open question.
Five underwriting questions get asked; four usually settle quickly. The article is the fifth — the one that is genuinely still open. A piece where every answer comes back clean has lost its nerve and turned into a pitch.
The record is reviewed in public, including the bad ones.
Every finished piece is scored against a fixed ten-criterion rubric and the scores are published — the 2.3 alongside the 4.7. It is a review of the writing, not a scoreboard of returns.
Recent
Read it
New pieces go out by email. No trading alerts, no model portfolio, no urgency.
Subscribe on Substack →The evidence
What we said would happen.
Every piece ends by naming the conditions that would prove it wrong, and the scheduled event that will test them. This is what happened next — kept up to date, including when the answer is that we got it wrong.
Nothing on this page is ever deleted or quietly revised. A test can only gain an outcome, and changing our mind is recorded as a dated event with a reason, not as an edit. That is deliberate: a record with nothing broken in it would not be a record.
Reading a test
Pending — the event hasn't happened yet.
Confirmed — the event happened and the thesis held.
Broken — the condition came true. Evidence against us.
Each line is written as the thing that would damage the argument, so a test passing means the damage did not occur. Dates are when the evidence is scheduled to arrive, not when we expect a share price to do anything.
Nothing tracked yet.
The record
Reviewed on the writing.
Most newsletters ask to be judged on their winners. This one is judged on whether the reasoning was any good — because that is the part the reader is actually being asked to trust.
The scores below rate the quality of each article against a fixed rubric: was the central question clear, was the evidence checkable, was the bear case real, were the falsifiers dated. They are not investment returns and they say nothing about how any share price has since moved.
Editorial scoring
| Date | Company | Score | Biggest weakness found |
|---|---|---|---|
| 2026-07-07 | ORCL | 4.7 | Swing factor restated across three sections |
| 2026-06-19 | MTCH | 4.7 | No dated catalyst tying the test to a schedule |
| 2026-06-18 | MSFT | 4.7 | Central idea restated four times; mid-sections loose |
| 2026-06-17 | TMO | 4.7 | Durability drivers repeated across closing sections |
| 2026-06-30 | UZNF | 4.5 | "Can I actually buy this" section read brochure-ish |
| 2026-06-18 | SPCX | 4.5 | A "do nothing" conclusion risks feeling anticlimactic |
| 2026-06-19 | ACN | 4.3 | Thin on checkable facts; ran on reasoning not filings |
| 2026-06-16 | MSCI | 4.3 | No dated falsifier; "wait for a panic" is untestable |
| 2026-07-05 | NFLX | 3.8 | Soft rounding; bear case pre-rebutted before it was made |
| 2026-03-06 | GRND | 3.5 | No real bear case; earnings section read as a pitch |
| 2026-04-07 | AVAV | 2.8 | No bear case; leaned on statistics instead of argument |
| 2026-01-16 | CBRL | 2.3 | No thesis, no bear case, no falsifiers, thin financials |
The four weakest pieces shared one fault, and it is the fault this journal now exists to avoid: the case against was missing. The editorial principles were rewritten around that finding, and the review pass tests for it on every draft.
The archive
The method
Seven questions, every time.
Every piece answers the same seven questions in the same order. That is what makes two articles about completely different businesses comparable — and what makes it obvious when one of them is dodging something.
- What does the market believe?
Stated as the market would state it, not as a strawman. - Why does it believe that?
The bear narrative's own logic, taken on its merits. - What evidence supports it?
Figures traced to public filings, not to a summary of them. - What evidence contradicts it?
The gate. If this answer is weaker than the one above, the piece goes back. - What would invalidate the thesis?
Named conditions, each with the date the evidence arrives. - What remains genuinely uncertain?
Written as open questions, never as instructions. - What follows from all of it?
A personal stance, with the position disclosed.
How a piece gets made
Research runs through Legend and produces an evidence pack: the market narrative, the bull and bear cases, the verified figures, the dated falsifiers, the unknowns. Only then does anything get written.
The draft is written against a fixed structure, then put through a second, adversarial pass that compresses it, strips decorative metrics, calibrates every claim, and re-audits it line by line against the rules below. Two automated checks run before publication. Neither of them can approve a piece — every discretionary call is signed off by a person.
What this journal will never print
- A price to buy at, or a price to sell at.
- A rating. No buy, hold or sell — on any company, ever.
- A manufactured reason that today is the day.
- A claim that cannot be traced back to a public filing.
- A position held but not disclosed.
These are editorial lines, not legal ones. They exist so the writing stays analysis rather than a recommendation, and they are checked on every piece before it goes out.
Built on Legend
The research runs on a system.
Legend is a rules-based investment research system, built over several years and used on every company in this journal. It is not a product, it is not for sale, and it does not issue signals to anyone. It is the reason the analysis is consistent rather than a matter of mood.
What it does
It values a business by what kind of business it is.
A bank, a cyclical, a compounder and a holding company do not become worth something for the same reasons, so they are not valued the same way. Legend picks the method that fits the business and applies it identically to every company of that type — which is what stops the method being chosen to suit the answer.
It separates value from quality from timing.
Three different questions that get conflated constantly: is it worth more than it costs, is the business actually good, and is now a sensible moment. Kept apart, they disagree usefully. A cheap price on a deteriorating business looks nothing like a fair price on an improving one.
It writes down what it thought, when it thought it.
Every assessment is recorded at the time it is made, in a record that is only ever added to. That is what makes an honest review possible later — including of the calls that aged badly.
It is the same system running the author's own money.
Legend was not built to produce a newsletter. It was built to make decisions about a real portfolio, and it still does. The writing is a by-product of that work, not a business it serves.
The system's internals stay private: the formulas, the thresholds, the levels at which it changes its mind, and whatever it currently concludes about any named company. What appears in an article is the reasoning and the public evidence behind it. The tool shows through the discipline, not through screenshots of a dashboard.
How I make this
I research every company through Legend, my rules-based investment research system. It values businesses using the right method for their type, then tests that value against business quality and timing. AI helps me research, challenge the thesis, and write the piece. The judgement is mine, and so is the money.
This statement appears on every piece published. It names the AI assistance outright rather than describing everything except the drafting — an under-disclosure costs more trust than the disclosure does.