Two-way Door

Deciding carefully reads as a good habit. The problem appears when that care gets applied at the same weight to every decision.

That is what happened when Noa proposed changing the document list's default sort from alphabetical to recently edited. There was reasoning behind it and nobody objected, and the decision still took three weeks — two meetings, a round of gathering opinions, and an approval step. Reversing the setting would have taken ten minutes.

The same team's database choice went the other way. Because it was urgent, it got settled in two days, and when the time came to migrate a year later the migration alone ran two months. The three weeks and the two days had landed on exactly the wrong decisions.

The rule that came out of this gets described as doors. Some you can walk back out of; some close behind you. What divides them isn't how important a decision is but what reversing it costs, and the thing to judge is which kind of door you are standing in. Importance and reversibility are independent, which is why sorting by importance produces the wrong answer.

Measure the cost of reversal first, then attach a different process depending on the answer.

Three questions measure it.

1. How long does putting it back take? Flipping one setting is ten minutes; migrating a data structure is two months.
2. Has it left the building? A price announced to customers, a clause written into a contract, an API exposed publicly — none of those can be reversed among ourselves.
3. What accumulates? Decisions that accumulate data get more expensive to reverse over time. Something that takes a day today takes two weeks in six months, and nobody notices the shift while it happens.

Those three answers sort decisions into two groups.

The reversible group gets a light process. Whoever is closest decides on the spot and leaves one line saying what they decided and why. No meeting, no approval. The record exists not for approval but because reversing later requires knowing what there is to reverse.

The irreversible group gets a heavy process. Write down three or more alternatives, note the risk in each, have the affected people read it, and approve explicitly. This is the place for a document like an architecture decision record, which preserves the context and the alternatives.

There is one more move: using mechanisms that convert a one-way door into a two-way one. Wrap a feature so it can be switched on and off with a feature flag and the decision to ship a new screen moves into the reversible group. A feature flag is a switch that turns functionality on or off while the code stays deployed. Releasing to a small slice first with a canary release has the same property. The kind of door isn't fixed; preparation changes it. A team that builds those mechanisms early finds most of its decisions sitting on the light side of the line.

In practice most decisions land in the reversible group.

1. Screen copy and layout. Change it, and if the reaction is bad change it back.
2. Defaults. Sort order, whether notifications start on — values a setting can restore.
3. Internal tool choices. Moving the document tool or the planning tool a team uses doesn't cost much, even though the argument about it can consume a week.
4. Experiments. An A/B test is run on the assumption of reversal from the start — showing two versions to randomly split groups and comparing the results.

The irreversible group is smaller and heavier. The shape of a publicly exposed API, a data schema, a decision to raise prices in pricing, a brand name, hiring and firing. Prices are asymmetric between up and down: raise one and reverse it and the number returns, but what customers remember does not, which is why a price rise belongs in the heavy process.

Some cases sit in between. Technology choices are the obvious example. The framework chosen in a project's first week is close to a two-way door because there is barely any code yet; after a year of use it has become a one-way door. The same decision changes character over time, so which side it sits on gets re-examined rather than fixed.

Using this as a team works better with the list written down. Deciding in advance which categories of decision belong on which side avoids relitigating "does this need a meeting" every time. When a category arrives that isn't on the list, settle it once and add it, so the list gets more useful rather than more disputed.

"This one's reversible, so let's decide today and move on."

The cost of the process exceeds the cost of the decision.

"Schema changes are a one-way door. Let's write it up and get review."

High reversal cost, so attach the heavy process.

"Wrapped in a flag it becomes reversible, so let's just ship it."

Use a mechanism to change the decision's character.

Pagedeck keeps a one-page list of decision categories. When Grace first proposed it there was pushback — the worry being that letting people decide reversible things alone would let directions drift apart.

So one condition got attached. You can decide alone, but what you decided goes in the team channel as one line. That's a notification step, not an approval step. Mia reads through those each week, and anything that has drifted gets reversed then. The premise is that for decisions this cheap to reverse, reversing them later is not too late. A week of drift on a default sort order costs less than a week of waiting for approval.

The most common reading is that reversible means it doesn't matter how you decide. If being wrong can be undone, the quality of the judgment looks less important. But undoing requires knowing what was changed, and with no record the reversibility itself disappears. A team that made thirty small decisions over six months without recording any of them ends up unable to explain why the current screen looks the way it does. What gets lighter is the process; the record doesn't. One line is enough, and one line is not optional.

Reading a fast decision as an under-considered one is also frequent. More time spent sounds like better judgment. But three weeks does not raise the accuracy of a decision that takes ten minutes to reverse. What usually grows is repetition of the same discussion, while other decisions queue behind it. The point isn't saving time but moving the saved time toward the decisions that can't be undone, where three weeks of examination genuinely changes the outcome.