Kill criteria: deciding in advance what would make you sell

Most people who hold individual stocks for years can describe, in detail, why they bought. Far fewer can say what would make them stop. That asymmetry is where a surprising amount of damage happens: not in the buying decision, which usually gets real thought, but in the years afterwards, when the reasoning is never revisited and the position quietly becomes a habit.

The failure is rarely that someone was wrong about the company. It is that they stopped checking whether they were still right, and had no particular moment at which checking was required. A thesis with no exit condition cannot be falsified, and a position you cannot be wrong about is not an investment — it is an attachment.

A stop-loss triggers on price. A kill criterion triggers on the thing you actually believed. Those come apart constantly, and in both directions.

A stock can fall 30% while every reason you bought it remains intact — a sector de-rating, a rate move, an unrelated scare. Selling there converts a paper move into a realised loss and hands the position to someone who did the same work you did. A stock can also rise 40% while the specific thing you were betting on quietly stops being true, which is the more expensive case, because nothing about the price tells you to look.

Price is a measure of what other people currently think. Your thesis is a claim about the business. If you want a rule that protects the thesis, it has to be written in the same units the thesis was.

A useful kill criterion has three properties. Miss any of them and you have written down a worry rather than a rule.

It is observable. Someone other than you, reading a filing, should be able to say whether it happened. “Competition intensifies” is not observable. “Gross margin below 55% for two consecutive quarters” is. If checking it requires a judgement call, you will make that call in the mood you happen to be in, which defeats the purpose of writing it down early.

It is tied to the reason you bought. The test is simple: if this became true, would the argument I made when I bought still stand? If the answer is “it would be bad news, but my case survives”, it is a risk to monitor, not a kill criterion. Keeping those separate is what stops the list growing until it is never read.

It is pre-committed. Written before you own the position, or at least before the thing you are worried about starts happening. Criteria invented during a drawdown are usually rationalisations of a decision already made emotionally, and they tend to be conveniently just below wherever the metric currently sits.

Suppose the thesis on a semiconductor company is that data-centre demand is structural rather than a build-out spike, that pricing power shows up in gross margin, and that revenue concentration among a handful of hyperscalers is the main fragility. Criteria that follow from that reasoning might be:

“Gross margin below 65% for two consecutive quarters.” — the pricing power claim, made checkable. One quarter is noise; two is a trend.

“Combined capex guidance from the top four hyperscalers declines year over year.” — the demand claim, tested against the customers rather than the company’s own commentary about them.

“A single customer exceeds 30% of revenue.” — the fragility, given a number so it becomes a fact rather than a feeling.

Notice that none of them mention the share price, and that each one maps to a specific sentence in the original argument. That mapping is what makes them useful two years later, when you have forgotten the detail and only the conclusion remains.

Start from the thesis, not from a list of risks. Take each load-bearing claim — the sentences that, if removed, collapse the argument — and ask what number or event would show that claim failing. Then ask what threshold would make you act rather than merely wince, and where that number is reported so you can check it.

Three to five criteria is usually right. Fewer and you have not covered the argument; more and you have written a monitoring project you will abandon. It helps to include at least one criterion that would genuinely hurt to act on. A set you would be comfortable triggering is usually a set calibrated to never trigger.

Write the consequence next to each one. “Reduce to half position and re-write the thesis” is a decision. “Reconsider” is not, and you will be reconsidering under exactly the conditions in which you reason worst.

A triggered criterion is not an instruction to sell. It is an instruction to stop and re-derive the argument from where things now stand — and to do it in writing, because the failure mode here is a fluent explanation of why this particular breach does not count.

The honest version has three possible outcomes. The thesis is broken and the position should go. The thesis has changed and needs rewriting, with the new reasoning recorded and the old version kept so you can tell later which one you were acting on. Or the criterion was badly chosen — it measured something that turned out not to be load-bearing — in which case the criterion changes and you write down why.

All three are legitimate. What is not legitimate is quietly deciding it does not apply and leaving no record that it fired, because the next time you look, the reasoning will read as though nothing happened.

The hard part is not writing kill criteria. It is checking them, every quarter, on every position, against filings that arrive whenever they arrive — which is a chore precisely when it matters most, because a thesis under pressure is the one you least want to examine.

That is the job Okidor was built to do: hold the criteria you wrote, read each new filing, earnings call and material news item against them, and tell you when one has been met. You can see the output on a worked example without signing up.

Nothing here is investment advice, and the examples are illustrative rather than recommendations. See the disclaimer.