Kill criteria: the one trick that makes 'shut it down' a decision, not a vibe

The dead founder's last six months

Watch a hundred startups close and a pattern surfaces. The founder spends roughly five and a half months between the moment the data has clearly said "this is over" and the moment the founder admits it. That five and a half months is the most expensive interval in any startup's life — it consumes the last of the bank balance, the last of the team's morale, and the last of the founder's bandwidth to ship the next thing. It is also entirely preventable.

The mechanism that prevents it is a kill criterion. The reason most founders don't set kill criteria is that picking the moment of your own failure feels like inviting it. The reason they should is that the alternative — discovering the failure in retrospect, half a year late — costs more than the discomfort by every measurable axis.

What a kill criterion actually is

A kill criterion is exactly three fields — not four, and never something fuzzy like "depending on how things look":

• Metric — a quantity that can be measured without effort or judgment.
• Threshold — a single number, not a range.
• Deadline — a calendar date, not a phase.

"Kill if MRR is below $1k by 2026-08-15" is a kill criterion. "Reconsider if traction is slow this quarter" is a wish. The difference is whether your future self can argue with it. The whole point of pre-committing is that the future self loses the argument by default.

The fields cannot be expressions of intent. "Kill if I'm not feeling product-market fit by month 4" is not a kill criterion because nobody — not the founder, not the team, not an investor — can decide externally whether the founder is feeling it. The criterion must be settleable by a single SQL query, a single calendar check, or a single CRM filter. If it requires interpretation, it will be interpreted away the day it fires.

The three to five rule

Three is the lower bound because a single criterion is too easy to dismiss as an unlucky measurement and two leave you with the wrong incentive — beat one, ignore the other. Three forces the founder to model failure across multiple axes: demand (does anyone want this?), conversion (will they pay?), velocity (can we ship?).

Five is the upper bound because beyond five the criteria start to overlap and the founder loses the discipline of saying which one fired first. The Inite cohort distribution settles cleanly at a median of four: one demand criterion, one conversion criterion, one velocity criterion, one stage-specific bet. That's the working shape for most founders.

The Founder Loop's default proposal is three pulled from the dealbreakers pass. The founder accepts, edits, or replaces. The criteria get persisted onto the audit row and from that moment the Monday digest checks them.

Good kill criteria, bad kill criteria

Bad kill criteria look like accountability but leave the founder a rhetorical exit. The pattern is recognizable across cohorts:

• Soft floor — "kill if we don't make progress on enterprise." Define progress.
• Movable target — "kill if growth stalls for a quarter." Whose calendar? Stalls how, against what?
• Indistinguishable from success — "kill if we don't close 10 customers." If 10 is success and 9 is failure, what is 8 — a successful failure?
• Compound condition with an or — "kill if MRR is under $5k OR churn is over 15%." An OR creates two criteria masquerading as one; either pick the strongest single condition or list both criteria separately and check them independently.

Good kill criteria are the ones a stranger could verify in two minutes:

• "Kill if demo-to-paid conversion is below 8% across ≥60 demos by 2026-09-30."
• "Kill if no $5k ACV pilot closes by 2026-10-15."
• "Kill if monthly active days per user drops below 4 for two consecutive months."
• "Kill if we cannot demonstrate <3 minute activation time after three onboarding rewrites."

Each carries a metric, a number, and a date. None requires the founder to feel anything.

Proximity is the real signal

The fire is not the only signal. The proximity to a kill criterion is the early warning that the loop watches every week. The Monday digest reports each criterion against three bands:

• Green — more than 25% above threshold; track quietly.
• Yellow — within 25% of threshold; surface to the founder for context.
• Red — within 10% of threshold or already past it; convene the kill decision.

The Yellow band is where almost all of the value sits. A founder who learns at week 4 that two of their five criteria are inside Yellow doesn't have to kill anything — but they do have to look at the lens the criterion came from, run experiments against it, and have a recompute by week 6 that either pushes them back into Green or accepts that two criteria are about to fire. That intervention window is what the loop is built to create. By the time a criterion turns Red, most of the operating decisions have already happened.

What happens when a criterion fires

A fire opens a 14-day hard-stop window. The window has three rules:

1. No new commitments to customers, hires, or vendors during the window.
2. A written rationale is required to override. Two paragraphs minimum, persisted onto the audit row with a timestamp.
3. The team is told. Not "the team will be informed in due course" — the team gets the same digest the founder gets, on Monday morning, the moment the criterion fires.

Of the founders in the cohort who fire a criterion, 38% override and continue. Of those overrides, 71% end up killing the project within the next 12 weeks anyway. The override mostly buys delay, not survival. That number alone is the argument for honoring the trigger — overriding has a 1-in-3 chance of producing a different ending, and the median delay it buys is two and a half months of additional burn.

The remaining 62% who honor the trigger close cleanly. The most common feedback in the post-mortems is some variation of "I knew six weeks ago." The kill criterion is not new information when it fires. It is permission to act on information the founder already had.

Why this is the part of operating that compounds

Every founder runs maybe four or five real ideas across a career. The cost of running idea N+1 is whatever runway you have left after idea N. A six-month-late shutdown of idea N is a six-month theft from idea N+1. Compounding works in the other direction too — every clean shutdown of a dead idea is a quarter or two of capital and energy you get back to spend on the next one.

Kill criteria are the cheapest investment a founder can make. They cost forty minutes to write, zero dollars to maintain, and they are the only structural mechanism that returns wasted time to the founder when the data says it's over. The Founder Loop checks them so you don't have to remember to. The discomfort of writing them is the price; the runway they recover is the return.