Kill tests: how to find out you are wrong early, with a one-page worksheet

To find out early that you are wrong, write down before you act the specific result that would prove your main claim false, find the cheapest safe way to look for that result, and then run five challenge checks: is the frame wrong, is there another explanation, who gains from misleading you, do your sources all trace back to one place, and what would flip the decision.

The Of One method calls the first part a kill test. Its rule is short: every strong claim carries one. This guide explains kill tests, the related idea of a reversal condition, and the five challenge passes, and ends with a one-page worksheet and a filled-in example.

Why a claim needs a way to lose

The philosopher Karl Popper argued that what makes a theory scientific is that some possible observation could contradict it. The Stanford Encyclopedia of Philosophy puts his criterion this way: “if a theory is incompatible with possible empirical observations it is scientific.” A theory that fits every possible result is, on that criterion, unscientific.

The same holds for business and personal claims. “This hire will work out” fits almost any outcome if you never say what failure looks like. “If the next release slips by more than two weeks with the hire on board, the claim that the hire fixes our speed is wrong” can lose, and so it can teach you something.

A kill test names the claim it targets and the result that would kill it. Add a third part, when you will look:

The claim it targets. One sentence.
The killing result. An observation that would show the claim is false. Specific enough that two people would agree whether it happened.
The deadline or trigger. When you will look.

Write it before you act. After you act, it is too easy to decide the result you got was not really the killing one.

Kill test versus reversal condition

They are close cousins.

A kill test attacks a claim: “this belief is false if we see X.”
A reversal condition attacks a decision: “we would choose differently if X were true.”

A decision can survive a dead claim (other reasons may still hold), and a decision can reverse while every claim survives (the stakes change, or a better option appears). Of One tracks both. In the repo’s example map, the tradeoff surface lists exactly which unknown and which trigger would reverse the choice of option. And one of the required challenge passes is a “stopping counterexample”: before you stop, name the reversal condition and the cheapest safe test for it.

Where kill tests come from: the pre-mortem

The hardest part is thinking of how you could be wrong while you are excited to be right. A useful tool here is the pre-mortem, described by Gary Klein in the Harvard Business Review in 2007. The team imagines the project has already failed and writes down plausible reasons why. Klein’s point is that this makes it safe for worried, knowledgeable people to speak up.

Each reason from a pre-mortem is a kill test waiting to be written. “It failed because customers would not pay upfront” becomes “if fewer than X of the next Y offers are accepted, the claim that customers will pay upfront is dead.”

The five challenge passes

Of One does not let a decision be released until a set of challenge passes has been run and answered. In plain words:

Pass	The repo’s name	The question	What counts as done
Wrong frame	frame challenge	Are we even asking the right question?	The assumption tested, the other frame tried, and the result
Another explanation	model expansion	What else could explain what we see?	A surprise test, the candidate model class, and the result
An adversary	adversarial	Who gains if we believe this, and how could they bend the evidence?	The actor or attack surface, the distortion test, and the result
Dependent sources	source independence	Do our “separate” sources really trace back to one?	The sources examined, how they depend on each other, and the result
What would flip it	stopping counterexample	What result would reverse the decision?	The reversal condition, the cheapest safe test, and the result

A sixth check, the causal descent, is covered in Why until bedrock.

The repo is strict about what counts. A pass is not done because someone mentioned it or ticked a box. It counts only with an actual answer that names its source, how reliable that source is, its custody, its evidence, and what the check actually tested. You can use the same standard on paper: a pass with no evidence behind it is not a pass.

A note on dependent sources, because it is the one people skip most. Three news stories quoting one press release are one source. Two colleagues who both heard it from the same customer are one source. The repo puts it plainly: repeated observations from one upstream process do not count as independent confirmation.

The kill-test worksheet

Copy this onto one page. Fill it in before you commit to anything you cannot easily undo.

Part 1: The claim and its kill test

Line	Prompt	Your answer
1	The decision, in one sentence, with a date
2	The claim the decision rests on most
3	How sure are you: low, medium or high, and why
4	Kill test: the result that would prove line 2 false
5	The cheapest safe way to look for that result
6	When you will look (a date or an event)

Part 2: The challenge passes

Line	Pass	Your answer	Evidence you checked
7	Wrong frame: what if the real question is different?
8	Another explanation: what else fits the facts?
9	Adversary: who gains if you believe line 2?
10	Dependent sources: how many truly separate sources back line 2?
11	What would flip it: the result that reverses the decision

Part 3: What you will do

Line	Prompt	Your answer
12	If the kill test fires, what do you do instead?
13	Your first move, and whether it can be undone
14	Who must sign before any step you cannot undo

A filled-in example (illustrative)

This example is invented. It is not a real business or a real result.

Situation. A two-person design studio bills by the hour. The owners think clients would rather pay a fixed monthly retainer, and they plan to switch all clients over next quarter.

Line	Answer
1	Move every client from hourly billing to a monthly retainer by January 1.
2	Most clients prefer a predictable monthly fee to an hourly bill.
3	Low. Basis: two clients mentioned it, both in passing.
4	Kill test: if fewer than three of the next six clients offered both options choose the retainer, the claim is dead.
5	Offer both options, side by side, on the next six proposals. No existing client is moved yet.
6	After the sixth proposal, or December 1, whichever comes first.
7	Wrong frame: maybe clients do not want predictability; they want smaller surprise bills. Test: ask the next three clients which problem the retainer solves for them. Result: pending.
8	Another explanation: the two clients who asked may have had one unusually large bill. Checked: both had a bill above their usual range last spring. Weakens the claim.
9	Adversary: a software vendor selling retainer billing tools published the article that started the idea. Its interest is plain. The claim should not rest on it.
10	Dependent sources: the two clients know each other and may have discussed it. Truly separate sources: possibly one.
11	What would flip it: if retainer clients use far more hours than the fee covers, the move loses money even if clients love it. Cheapest test: track hours against the fee for the first three retainer clients for two months.
12	If the kill test fires, keep hourly billing and offer the retainer as an option only.
13	Offer both options on new proposals; move no existing client.
14	Both owners sign before any existing contract is changed.

Look at what the worksheet did. The owners started with a plan to move everyone at once. They ended with a small, reversible test, a date to check it, and a clear rule for what happens if they are wrong. Line 10 alone was worth the page: their “two clients” might be one opinion.

Do it today

Pick the decision you are most confident about. Confidence is where kill tests help most.
Fill in lines 1 to 6. If you cannot write line 4, your claim cannot lose, which means it cannot teach you anything yet.
Spend ten minutes on lines 7 to 11. Write “no evidence yet” where that is the truth.
Choose the reversible step on line 13 and do it this week.

Kill tests sit on the map next to the claims they target; see What goes on a decision map. The method page covers the challenge passes.