Nobody decides this well in the moment. By the time the question arrives you are the sunk cost, and every number has two readings. The version that works is a condition written down before the evidence exists — a number, a date, and somebody who can see it. We publish our own: 10 ways this business could end, each with the thing that would settle it, and 1 of them still carrying no test at all.
Why the decision cannot be made when it is due
Two things are true at once by the time you are asking. You know the most you have ever known about the market, and you are the least able to act on it.
The mechanism is not weakness. It is that the evidence and the attachment arrive together: every month of work adds information and adds a reason to discount the information. A bar set today gets argued down tomorrow, because tomorrow you will have a genuinely new fact to argue with.
So the decision has to be moved earlier, to the one moment when you are informed enough to set a bar and not yet invested enough to move it. That moment is now, whenever now is, and the artefact it produces is not a plan — it is a sentence somebody else could check.
What a published register looks like
Here is ours. Each row is a way the business ends, in the words our own gate used when we ran this idea through it, with what would settle the row and a live status:
The 10 rows of our own register, counted by status from the same list the public page renders. The useful column is the middle one: 7 rows have a named experiment running, which is the only status that produces an answer rather than a feeling.
The row that does the most work is the one still open, and it is worth quoting because a register that hides its worst entry is a brochure:
No durable distribution channel — A skeptical, online-native audience that is easy to reach and hard to convert — and the founder won't do sales calls.
Open means no test is running against it. Publishing that is uncomfortable and it is the whole point: a risk with no experiment attached is one you have decided to live with, and saying so out loud is different from not having noticed.
Turning a bar into something checkable
Most stop conditions fail because they are not conditions. They are moods with dates attached.
What founders write | What a stranger could check |
|---|---|
| If it is not working by the summer | If there are fewer than ten paying customers on 1 June |
| If nobody seems interested | If under two per cent of visitors start a trial across four hundred visits |
| If I stop enjoying it | If I skip three consecutive weeks of the scheduled work |
| If it costs too much | If the monthly bill passes what I earn from it for two months running |
The same intention written twice. The test for the right-hand column is mechanical: hand it to somebody who does not care about the project and ask whether they could tell you the answer without consulting you.
Revenue bars need one extra step, because a revenue number does not tell you whether the deadline was ever plausible. Convert it to a customer count first:
| Price | Customers for $1,000 a month | New ones per month to get there in 3 |
|---|---|---|
| $19 | 53 | 18 |
| $49 | 21 | 7 |
| $99 | 11 | 4 |
Counts from the same function behind the customers-needed calculator. The third column is the one that decides whether your deadline is a plan or a wish — at $19 it is 18 new customers every month, and at $99 it is 4.
What to do on the day it triggers
- Do not renegotiate it alone. If the condition was written for a reason, the argument to move it is being made by the person it was written to protect you from. Say it out loud to somebody before you act on it.
- Separate stopping the work from stopping the thing. Archiving a product is reversible. Cancelling the domain, deleting the data and telling the customers are not. Do the reversible half first and wait a week.
- Write the post-mortem while you still have the numbers. What you believed, what the bar was, what actually happened, and what you would need to see to try again. Ours is a public page; a text file works.
- Keep the idea where you can find it. We give killed ideas their own place — the graveyard — because a rejection is dated, and the thing that killed it can change.
What a stop condition does not do
It does not make the decision correct. A bar set in advance can be set wrong, and plenty of businesses that were shut down on schedule would have worked with one more quarter. There is no version of this that removes the risk of stopping too early — the claim is narrower: a bar set in advance is not distorted by the attachment that arrives afterwards.
It also does not settle whether the business could work for someone else. Our own register is full of rows about this founder's constraints — the channels available, the time, the refusal to do sales calls — and those are reasons to stop that say nothing about the market.
The short version
- The decision has to be made before the evidence arrives, because by then you are the sunk cost.
- A stop condition is a number, a date and a witness. If a stranger could not check it for you, it is a mood.
- Convert revenue bars into customer counts before you commit to them — the per-month column is what tells you whether the window was ever real.
- Publish the register if you can bear it. Ours has 10 rows and 1 of them still has no experiment running against it.