mindjump
decisions

Reversible and irreversible decisions

Published November 6, 2026 · 3 min read

Some decisions are like doors you can walk back through: if the choice turns out badly, you undo it at little cost. Others lock behind you. The common mistake is to treat both kinds the same way — agonising over choices that are cheap to reverse, and rushing choices that are not.

The idea

Before deciding, ask one question: if this turns out to be wrong, what will it cost to undo, and how long will it take? Some people call the two kinds two-way doors and one-way doors. For a two-way door, the cheapest way to learn is usually to go through it, watch what happens and adjust. For a one-way door, there is no second attempt, so the time to think is before.

The most useful consequence is the third option: turning one-way doors into two-way doors. A pilot, a prototype, a trial period, renting before buying — each converts an irreversible commitment into a reversible experiment.

opens againTwo-way doorcheap and quick to undoDecide fastTry it, watch, adjustSet a date to reviewlocksOne-way doorcostly or impossible to undoSlow downInvert: how would it fail?Leave a margin
The speed of a decision should match the cost of undoing it.
Case study · Software and construction

Releasing in stages

Updating software for a million users used to be a one-way door. Once a faulty version was out, everyone had it, and the damage was done before anyone noticed. Large software teams now routinely make releases reversible by canarying, named after the canaries once taken into coal mines as early warnings. A new version goes first to a small share of users while the team watches error rates and complaints. If something is wrong, only that small group is affected and the release is rolled back; if not, it is widened step by step.

Suppose one release in ten has a serious bug, and a bug gets noticed once about 500 people have run into it. On average, per release:

1001,00010,000100,000Everyone at once100,000 · 1 day10%, then everyone10,000 · 2 days1%, then 10%, then everyone1,000 · 3 days100 users, then everyone1,000 · 2 daysusers on a faulty version, per release (log scale) · time to reach everyone
Expected users who receive a faulty version, per release, for one million users (10% of releases faulty; a bug is noticed after 500 users hit it).

Shipping to a tenth first cuts the damage tenfold, at the cost of one extra day. Starting at 1% cuts it another tenfold. The last row holds a surprise: a first stage of only 100 users does not help, because 100 users are not enough for anyone to notice the bug. A trial run only makes a decision reversible if it is big enough to tell you something.

Construction shows the opposite case. Concrete, once poured and set, cannot be adjusted. So the industry concentrates its checks before the pour: drawings are reviewed, reinforcement is inspected in place, and the concrete is tested on delivery. Where a door is truly one-way, the caution moves to the front.

Try it yourselfHow much does a careful release save?Compare four release plans for a million users.

How to use it

  1. Ask: if this is wrong, what does undoing it cost, and how long does it take?
  2. If undoing is cheap, decide quickly, and set a date to look at the result.
  3. If undoing is expensive or impossible, slow down: list how it could fail, and leave a margin.
  4. Look for a way to make the choice reversible: a pilot, a stage, a trial, a smaller first step.
  5. Check that the trial is big enough to show a problem, or it only feels safe.

Where it misleads

Many decisions look reversible and are not. A product can be withdrawn, but the lost trust of the people it disappointed stays; deleted data cannot be undeleted; a team asked to change direction three times may not follow a fourth. Doors also close with time: something easy to undo this week may be expensive to undo next year, once other things have been built on top of it.

The opposite failure is just as real. Treating every decision as a one-way door makes an organisation slow, and the cost of not deciding is easy to overlook because nothing visibly happens.

In one line: match the care you take to the cost of undoing, and look for ways to make undoing cheap.

Next modelInversionAsk how to guarantee failure, then make each of those paths hard to reach.
mindjump · small thinking tools, written down so I actually use them.