Bear Case vs Bull Case: Break Your Own Thesis First

A bear case is not pessimism and a bull case is not optimism. They are the same list of sentences about a business, read by someone who wants one of them to be false. So label every load-bearing claim in your notes as sourced, inferred or assumed, then write down what would have to be true for the whole thesis to be a mistake, and the date you will check it. A thesis you have never tried to break is a hypothesis wearing a verdict's badge.

What most investing journals record#

The standard advice is sensible and about half right. Keep a fixed format: instrument, date, price, size, costs, reason for entering, outcome, note after the fact. Fill it in as you go, because an entry written a week later is already a memory, and memory flatters the person doing the remembering. After a year you have material that shows a repeated mistake instead of a run of individual bad luck.

That half genuinely works. Without a written record, memory rewrites the reason for a decision into one that fits the result, and it does so silently.

The trouble is that such a journal records the decision, not the reasoning you could break.

A decision has one outcome. Reasoning has several assumptions, and each can break separately, in a different quarter, for a different reason. If the "reason" column says "the company has a strong brand", there is nothing to check three months later, because a strong brand takes no truth value.

Both records answer real questions, which is why both deserve to exist. The trade journal answers whether you follow your own rules. The thesis journal answers whether the reason you look at this company at all is still true. You audit the first one against yourself. The second is audited for you by the filing calendar, and only the second can be lost quietly. The rhythm of coming back to that page, meaning what to read after a quarter and what to ignore on purpose, is following a company on the filing calendar instead of the news feed.

Three labels instead of one certainty#

Every sentence in your notes has one of three origins, though they all look identical on the page. Sourced: you can point at a page in a filing or a line in an earnings-call transcript. Inferred: you connected facts into a chain the company never stated outright. Assumed: you believe it and have not checked.

The label describes where a sentence came from, not how sure you feel about it.

Say you hold a company because "its customers are not going anywhere". Break that into three sentences. The top three customers account for 30% of revenue: sourced, because the number sits in the annual report. Switching suppliers is expensive because the tool lives inside the customer's accounting process: inferred, because it sounds right and the company has never costed it. Contracts renew at higher prices: assumed, until you find a renewal figure in a document.

Which means that of the three sentences holding up your thesis, one has backing, one has a sketch of backing, and one has none. Round numbers here, because they are invented, but that proportion is real and looks much the same for almost any company you know at second hand.

The problem is the calendar. After a week, "inferred" reads exactly like "sourced": the sentence is the same one, and what has gone is your memory of where it came from. The label is a cheap barrier between an assumption and a fact it has no right to become.

Why bother sorting them, if the sentence reads the same either way? Because each label calls for a different next move.

A sourced sentence is invalidated by one new number in the same place in the document, so checking it takes two minutes a quarter. An inferred sentence breaks along with any of its links, and it reads as the most persuasive of the three. So it is the last column, not the first, that tells you what each sentence costs you in work and which one will break without warning.

Which sentences carry the load#

Not every claim deserves this rigour, because at thirty labels you stop reading them. Pick three or four load-bearing sentences: the ones holding up your whole picture of the company the way a load-bearing wall holds up a floor. Customer concentration, regulatory exposure, the direction of demand, the ability to raise prices, one specific promise from management. The rest are partition walls and can be moved without touching the structure.

The load test fits in one question: if this sentence turned out to be false, would you have to rebuild your picture of the company from scratch?

How many should there be? Three is the floor at which a description of a business still holds together, five the ceiling at which you still come back to the list. At two you lose sight of half the company; at six you have mixed what carries weight with what is merely interesting, the review starts to look like work, and you quietly stop doing it.

Tidy sentences are the most suspect, because they read like conclusions and are really bags of unexamined claims. "The brand is strong" smuggles in five at once: customers know it, prefer it, pay more for it, come back, and would not switch quickly to a substitute. Each of those five has a different origin and a different document behind it. Unpacking a phrase like that into smaller, settleable sentences is the cheapest thing you can do to your notes, and the ready-made set of questions to put to any company sits in the piece on 25 checks a good analysis passes.

Bear case vs bull case, plainly#

With three labelled load-bearing sentences in front of you, each one gets a breaking condition. Not after the fact, when the price has already fallen and you are looking for an explanation. Earlier, while nothing hurts and you have no stake in the answer.

A bear case and a bull case differ only in which sentence on your list they treat as false. Without such a sentence, a price falling stays just a price falling. A bear case written as "what if the market crashes" is useless, because it names no sentence. The same case written as "if the top three customers exceed 40% of revenue for two consecutive quarters" can be settled from a table, on one day of the year, by a person with no opinion on the matter.

The price of unchallenged certainty was measured a quarter of a century ago. Brad Barber and Terrance Odean traced the accounts of 66,465 households between 1991 and 1996 and found the most actively trading fifth earned 11.4% a year against a market return of 17.9%. Which means they handed back more than a third of what somebody sitting on their hands collected, and the authors put the gap down to overconfidence.

The paper says nothing about whether any of those households wrote breaking conditions, because nobody collects that. My own view is that certainty nobody has challenged shows up eventually as an invoice.

The simplest way to force the sentence out of yourself is uncomfortable and takes ten minutes. Move a year forward and assume the thesis failed. Not "could have failed". It failed, case closed. Then write three sentences about what happened, or rather about what was already visible today and you did not want to look at. The exercise works because explaining a finished outcome is an easier question than estimating the odds of an outcome that has not happened.

I think the label that matters most of the three is "assumed", not "sourced". Sourced sentences get checked anyway, because they come with a document and a date. Assumed ones are invisible until somebody makes you name them, and then it usually turns out that they are the ones holding up the floor.

How to write one assumption#

Five steps, for one sentence:

1. Write a sentence about the business, not about the share price.
2. Put a label on it: sourced, inferred or assumed.
3. For sourced, give the document and the place; for inferred, the facts it follows from; for assumed, leave the space blank.
4. Add the breaking condition, with a number and a threshold.
5. Add the date on which you will check that condition.

Step three hurts most, which is why it matters most. A blank space in your notes is honest; the same space filled with a link to somebody else's article looks like backing and is not.

When two good sources disagree#

Sometimes you go to check an assumption and come back with two answers, both from documents. Say a maker of data-centre components. Case one: high-speed copper cables are cheap and power-efficient over short links between chips, so they stay in those racks for years. Case two: co-packaged optics matures faster than expected, runs fibre straight to the silicon, and cuts out the copper growth engine. Both are sourced and both sound reasonable.

This is not noise to average away.

The reflex is to split the difference: one side says this, the other says that, the truth is somewhere in between, move on. Except there is no in-between here. Either copper stays in those racks for another five years or it does not, and a portfolio does not get the intermediate version. A compromise at that point is not caution, it is a way of avoiding writing down that you do not know.

A split like that points at the single sentence the whole thesis rests on, and it hands you a deadline, because a deployment schedule will settle it rather than another article. That is exactly why a Full report pro is produced by Three models and a judge: three models write the same analysis independently, a fourth compares the versions, marks the contested places and keeps what a source can carry. It does not end in a vote or a view on a transaction, only in a list of the places where your work begins.

You can run the same move in conversation: ask a model to build the case against your thesis instead of asking it to confirm one. A request for confirmation always comes back with confirmation, and the difference between the two is worked through in the piece on using AI as a sparring partner.

Where this method fails#

A breaking condition can be written so that it never triggers, and most people write it that way on the first attempt. "If the fundamentals deteriorate" is not a condition, it is a promise to think about it some day. A threshold with no number can always be moved one quarter to the right.

The second failure is quieter. The labels turn into ritual and everything gets "sourced", because a document exists somewhere in which something similar was said. Rereading the same paragraph hoping it still says what you remember is not a check.

The third is the most human. A page that keeps holding becomes an alibi: no condition broke, so everything must be fine, even if the company has changed in a way your three sentences never covered. So once a year the question is not "did the conditions hold" but "are these still the right conditions".

There is one more caveat, without which all of this sounds too tidy. A broken thesis tells you the reason has expired, not what to do about it, and the rest of your process sits between those two things. A price falling with no sentence settled is, for now, just a price falling.

Where the page lives#

The method has one flaw: somebody has to keep it. In Taufolio that part happens without you. An Investment thesis comes out of a Full report already broken into assumptions, each with a measurable condition, and those conditions get revisited after results and after the events that could settle them. You can see what a thesis split into conditions looks like in a finished report on the product page.

Start with one company you believe you know. Write the three sentences your view of it rests on, put a label on each, and give one of them a breaking condition with a number and a date. If a year from now you cannot point at the document that settled at least one of those sentences, you do not have a thesis. You have a habit with a good explanation attached.