A bet has a thesis. A zombie has a defender.
How do you tell a slow bet from a dead one?
I ended my last essay with an admission. I had argued that when building becomes free, nothing dies on its own, and that the fix is kill criteria written at inception: a number and a date agreed before you build, so that sunsetting is the default and keeping the thing alive is the decision that has to be defended. Then I said I did not know where the line sits between a zombie and a bet that just needs more time.
That is the part people wrote to me about. Nobody argued with kill criteria. Everybody had the same worry, usually with a specific product attached to it: our thing missed its number in year one and became the business in year three.
They are right, and it is the whole problem. So here is my attempt at the missing half.
The simple part
Kill criteria do not decide anything. They only tell you when to look.
A number agreed eighteen months ago can be wrong in three ordinary ways. The metric was badly chosen. The strategy changed underneath it. Or the world moved and the number is now measuring something nobody cares about. Treating the criterion as the verdict just replaces one bad habit, keeping everything alive, with another one, killing on a technicality.
What the criterion buys you is a scheduled, unavoidable moment where somebody has to say something out loud. The question you ask in that moment is what decides it.
The one I am going to use: what specifically would change the trajectory, by when, and who is doing it?
Call it the trajectory test. There are only three shapes of answer.
The first names a change, a mechanism and a date. "We are shipping the new onboarding flow in October, it removes two of the five steps, and we expect activation to roughly double." That is a thesis. It might be wrong, and that is fine. It is fundable because being wrong will be visible.
The second is time. "It needs more time." No mechanism, no date, no claim about what will be different.
The third is faith. "The market will catch up." "The team believes in it."
Only the first one is a bet. The other two are defenses, and a product kept alive by a defense is a zombie product.
The complicated part
The reason the test works is not that it is clever. It is that a thesis is falsifiable and a defense is not.
A thesis tells you what you should see and when you should see it, which means it books its own second meeting. It also puts a name against that meeting. Defenses never schedule anything, and that is exactly the property that lets them survive for years. Nobody is ever wrong on a specific Tuesday.
There is a version of this test that people already run informally, and it is worth naming the difference. Most portfolio reviews ask "is this working?" and the honest answer for anything in trouble is "not yet", which is unfalsifiable and therefore useless. The trajectory question does not ask for a verdict on the past. It asks for a claim about the future that the person making it will be held to. That is a much easier question to answer honestly and a much harder one to answer vaguely.
Now the part that makes this genuinely hard in my world. In payments, "the market will catch up" is sometimes literally true.
I have watched payment methods do almost nothing for years and then become the default in their country in about two years, because a central bank shipped a rail and everyone moved at once. Brazil is the obvious case. If your kill criteria had been pointed at a local rail in its first eighteen months, you would have killed it, correctly by the number and expensively in reality. At Yuno we orchestrate 1,000+ payment methods across 190+ countries, and a real part of that list looks dead right up until it doesn't.
That is not an exception to the test. It is the test doing its job. "The market will catch up" becomes a thesis the moment you can name the event that would make it true and the date by which you would expect the first sign of it. "When the central bank mandates instant settlement, we expect volume within two quarters" is falsifiable. "Eventually people will want this" is not. Same optimism, completely different accountability.
The other half of the argument is cost, and it is the half that operators underrate. Every live product carries maintenance, a support surface, an on-call story and a place in the team's head. That last one is the expensive one. Every roadmap conversation and every architecture decision has to route around the things you kept. The heaviest company loses, and weight accumulates from decisions that all felt reasonable individually.
What I'm not sure about
The obvious hole is that this test rewards people who talk well.
The most articulate person in the room can generate a thesis on demand, complete with a mechanism and a date, and I would fund it. Fluency is not evidence, but under time pressure it reads like evidence. My current patch is procedural: write the thesis down at the moment it is offered, in the owner's words, and read it out loud to the same people when the date arrives. If a thesis has to survive being quoted back, the incentive to invent one drops sharply.
I am also not sure where the floor is. For a two-week experiment, making someone write a falsifiable thesis probably costs more than just letting the thing run and quietly turning it off. There is a size below which this is bureaucracy. I do not know yet whether the threshold is cost, headcount, or the fact that a product has external users, and I suspect it is the last one.
And I do not know how it behaves when the defender is me. It is easy to run this on other people's products. The products I have personally championed have always had, in my own head, a very good reason to keep going for one more quarter, and I doubt I am unusual. The only honest answer I have is to let someone else ask me the question and to write my answer down where the team can read it.
There is also a failure mode where the test works too well. A room that learns the format starts producing tidy theses for everything, dates included, and the review turns into a compliance exercise where nothing is ever killed because everything is technically a bet. If that happens the fix is not a better question, it is a harder budget: a fixed number of active bets, so keeping one alive costs another one its slot.
We are going to use it in our next portfolio review. I will write about what it killed, what survived on a thesis and then failed anyway, and the cases where I overruled my own test.
Until then, the question I would actually like answered: what is the last thing you kept alive on a defense, and what would the thesis have had to say for you to keep funding it?