The most expensive trade I have ever watched someone hold was one that had already told them it was wrong, three separate times, in plain language. Each time, the story got a little longer. The catalyst was delayed, not cancelled. The market was being irrational, not correct. The whales were accumulating, they just could not see it yet. By the fourth reinterpretation the thesis had mutated into something that could survive any outcome, which is another way of saying it no longer meant anything. That is the failure mode I want to talk about, because I have done it too, and the only reliable defense I have found is deciding what would make me exit before I have a position to defend.
Why the story always survives
Psychologists call it belief perseverance. Once you commit to a view, especially in public or with money behind it, your brain stops treating new evidence neutrally. Confirming data gets waved through. Disconfirming data gets cross-examined until it confesses to being noise, or a fakeout, or a shakeout designed specifically to knock you off the trade you were right about. The mechanism is not stupidity. It is that reinterpreting one data point is cheap and admitting the whole thesis is dead is expensive, so you keep paying the cheap price over and over until you have paid more than the expensive one.
Markets make this worse than almost any other domain, because prices are noisy and thesis timelines are fuzzy. If you believe a coin is going to run because supply is getting locked up, there is no bell that rings when you are wrong. The price can drift against you for weeks and you can always say it has not played out yet. The absence of a hard deadline is exactly the gap belief perseverance grows in. So the job is to install the deadline yourself, in advance, in a form your future self cannot negotiate with.
What a real kill criterion looks like
A kill criterion is a condition you write down before entry that, if it happens, means the thesis is falsified and you are out. Not out because you are scared. Out because the specific thing you said would prove you wrong actually happened. The whole point is that it has to be falsifiable. If you cannot describe an outcome that would make you exit, you do not have a thesis, you have a mood.
The good ones tend to come in three flavors, and a strong setup usually has one of each.
- Price levels. A close, not a wick, below or above a level that your thesis says should hold. Use a level that means something structurally, like the invalidation point of the pattern you are trading or the price where your reason for entering stops being true. Wicks lie, closes lie less. Pick the timeframe in advance too, because otherwise you will quietly drop to a lower one to avoid the signal.
- Scheduled data prints. If your thesis rests on an event, name the event and name the result that kills it. An earnings number, an inflation print, a protocol upgrade shipping or slipping, an unlock date passing without the price impact you expected. These are the cleanest kills because the calendar does the work and there is a specific moment where reality answers the question.
- On-chain or flow metrics. If the reason you are in a trade is that supply is leaving exchanges, or a cohort of wallets is accumulating, or funding is flipping, then the reversal of that exact metric is your exit. If the flow you built the thesis on turns around and you are still holding, you are no longer trading your thesis, you are trading hope.
Separating thesis failure from noise
Here is the part people get wrong even when they do write kill criteria. They set the trigger too tight, get stopped out on ordinary chop, decide kill criteria do not work, and go back to holding bags on vibes. A kill criterion is not a stop loss. A stop loss protects your capital from a single bad move. A kill criterion answers a different question, which is whether the reason you entered still exists. Those are related but they are not the same, and conflating them is why people abandon the practice.
The way I keep the two apart is a review cadence tied to the thesis, not to the price ticking. If the thesis is a multi-week supply story, I am not checking it every four hours, because at that resolution everything looks like it is failing half the time. I decide up front how often the thesis gets reviewed, maybe daily on the close, maybe weekly, and the kill criteria only get evaluated on that schedule. Noise happens between reviews. Signal is what is still true when the review comes around. That single rule, only judging the thesis at its natural cadence, filters out most of the false kills.
A rough workflow I would give anyone starting this:
- Before entry, write one sentence that states the thesis and the mechanism behind it. If you cannot fit it in a sentence you do not understand it well enough to size it.
- Under it, write two or three kill criteria, at least one price-based and one event or flow-based, each specific enough that a stranger could check it without asking you what you meant.
- Write the review cadence next to them. Daily close, weekly, whatever matches the thesis timeframe.
- Write it somewhere you cannot silently edit after the fact. A dated note, a message to yourself, anything with a timestamp beats memory, because memory is exactly what belief perseverance rewrites.
- When a criterion fires on a scheduled review, you exit first and analyze later. The order matters. If you analyze first you will talk yourself out of it.
The discipline is in writing it down
The reason this works is almost entirely about timing. Before you have a position, you can think about the trade honestly, because there is no ego attached to a thesis you have not acted on yet. Ten minutes after entry that neutrality is gone and every incoming data point is either a friend or an enemy. So the kill criteria have to be written in the honest window and enforced in the compromised one. You are basically leaving instructions for a version of yourself you know will be biased, and trusting the earlier, clearer version over the later, invested one.
I keep a running log of theses and their kill criteria, and I check the fired ones against on-chain flows and scheduled catalysts at a fixed cadence rather than staring at the tape. On Blockcircle we ended up building a lot of that review scaffolding into how the tooling works, precisely because the wiring to watch a level or a wallet is easy and the discipline to act on it is the hard part. But you do not need any of that to start. A dated note with a thesis, three falsifiable exits, and a review schedule will already put you ahead of most people, who are still out there patiently reinterpreting the same rejection for the fourth time.