In short: A decision record without a stated falsifier is a press release. The one field that turns it into judgment is "what would prove this wrong, by when". It is also the field that reveals, at the moment of decision, whether the people agreeing actually agree.
Key points
- A good falsifier has four required parts: measure, threshold, date and observer.
- Phrases like "if adoption is disappointing" are not falsifiers; they defer deciding what the decision meant.
- Two people who agree on a choice often write different falsifiers, exposing different bets.
- When a falsifier triggers, reverse, hold with a new recorded reason, or accept the loss; never quietly edit the threshold.
- If almost no falsifiers trigger, they are too safe and measure nothing.
A rationale is an argument, and arguments do not expire. Most decision templates ask for context, options, the choice and the rationale; every one of those can be written persuasively after the fact, and none of them can be checked.
A falsifier does. "If renewal rates in this segment fall below 85% within two quarters, this pricing change was wrong." It names a measure, a threshold and a date, and someone other than the author can observe it. It turns a decision into a prediction, and predictions can be graded.
Anatomy of a good falsifier
Four parts, all required:
- Measure: a quantity or an observable event, not a feeling.
- Threshold: the level at which you would say the decision was wrong.
- Date: when you will look.
- Observer: who checks, and from what source.
Weak falsifiers fail one of these. "If it doesn't work out" has no measure. "If adoption is disappointing" has no threshold. "If we don't see results" has no date. These are not falsifiers; they are permission to decide later what the decision meant.
Some examples across decision classes:
- Credit: "If leverage exceeds 5.5x at the next two reporting dates, the approval assumptions failed."
- Pricing: "If win rate in the mid-market band drops more than 8 points over the next 40 opportunities, the new price is wrong."
- Partnership: "If the partner has not sourced three qualified introductions by end of Q2, the channel thesis is wrong."
- Hiring: "If the new team lead has not run the weekly review independently by month four, the internal promotion was the wrong call."
The falsifier as an alignment test
Writing a falsifier does work at the moment of decision. It forces each person to state what they are actually betting on. Two people who agreed on the choice often write different falsifiers, because they agreed on what to do but not on why. One supports the price change because it will lift margin; the other because it will screen out bad-fit customers. Those are different bets with different failure signals.
That disagreement is worth more than the agreement. Surfacing it before the decision is cheaper than discovering it in the post-mortem.
What happens when one triggers
Record it as triggered. Then make one of three explicit choices: reverse the decision, hold it with a new and recorded reason, or accept the loss and close the record. What must never happen is a quiet edit of the threshold. Moving the goalposts is sometimes legitimate, but it is a new decision, and it should be recorded as a supersession with its own reason.
Calibrating the falsifiers themselves
Track the share of decisions whose falsifiers trigger. If it is close to zero, the falsifiers are too safe and are measuring nothing. If it is very high, either the decisions are too bold or the thresholds too tight. Somewhere in between, you are learning.
In practice
- Make measure, threshold, date and observer required on every material decision.
- Ask each decision-maker to write their own falsifier before discussion, then compare.
- Put the check date in the observer's calendar, not just the record.
- Report the share with valid falsifiers, the share reviewed on time, and the share triggered.
What would prove this wrong. Decisions recorded with falsifiers showing no better follow-through, learning or later outcomes than decisions recorded without them. Then the field is ritual.
Common questions
What is a falsifier in a decision record?
A statement of what would prove the decision wrong, by when. It names a measure, a threshold, a date and an observer, which turns a decision into a prediction that can be graded.
How do I write a good falsifier for a business decision?
Include a measure that is a quantity or observable event, a threshold at which the decision is wrong, a date when you will look, and who checks from what source. For example: if renewal rates in this segment fall below 85% within two quarters, this pricing change was wrong.
What should happen when a decision's falsifier is triggered?
Record it as triggered, then explicitly reverse the decision, hold it with a new recorded reason, or accept the loss and close the record. Changing the threshold is a new decision and should be recorded as a supersession.