One-way vs two-way doors

Match how hard you think to how hard it is to walk back — reversible calls deserve speed, permanent ones deserve rigor.

The idea

Most decisions are two-way doors: if you’re wrong, you walk back at small cost. Those deserve speed — decide at roughly 70% confidence and move, because waiting for certainty costs more than the mistake would.

A few are one-way doors: irreversible, expensive to undo. Those deserve rigor — gather more, raise your confidence, decide slowly.

The senior move is to notice which door you’re at — and, when it’s a , to ask whether a small bet can turn it into a series of cheap two-way ones.

set your confidence, then read the two doors

two-way door — reversible one-way door — irreversible

Your confidence 55%

Two-way door — value now

—

wrong costs a reversible −20

One-way door — value now

—

wrong costs an unrecoverable −200

Cost of waiting so far

—

delay to reach this confidence

Drag your confidence. Watch the one-way door’s downside — the wide hump far to the left — and notice how little the two-way door’s downside ever moves.

How it works

Put a number on it. The value of deciding now is your confidence times the upside, minus the chance you’re wrong times what being wrong costs. The only thing that changes between the two doors is that last term — the cost of being wrong.

value now = confidence x upside  −  (1 − confidence) x loss-if-wrong

Two-way door  (reversible; wrong costs a small −20 you can walk back)
  EV = 100p − 20(1−p) = 120p − 20
  break-even at p = 20/120 = 17%.   go fast, even at 55%.

One-way door  (irreversible; wrong costs −200 you cannot undo)
  EV = 100p − 200(1−p) = 300p − 200
  break-even at p = 200/300 = 67%.   at 55%, EV = −35.

Convert it: run a pilot -> three cheap two-way steps.
  worst case is capped at the pilot cost (−40, not −200),
  and you learn your way to ~85% before the irreversible commit.

Two numbers, one lesson: the reversible door pays off from 17% confidence, the irreversible one not until 67%. Same confidence, opposite advice — because reversibility, not certainty, sets how much rigor you owe the decision.

When to use it

Sort the decision first	Then
Reversible & cheap to undo (a copy tweak, a config, a reversible reorg).	Decide fast at ~70%. Perfect information isn’t worth the delay.
Irreversible & costly (a public commitment, a data migration, an acquisition, firing).	Slow down, raise confidence, invite dissent, write the memo.
Irreversible but you could test a slice.	Run a pilot — stage it into reversible steps so each is a two-way door.

Watch out for

Treating every door like a one-way door. Over-analyzing reversible calls is the more common failure — it looks careful but quietly burns speed and momentum. Most doors are two-way.
Treating a one-way door like a two-way door. The rarer, costlier mistake: shipping a data deletion or a public promise at 55% because “we’ll iterate.” Some doors don’t reopen.
Confusing hard-to-undo with impossible-to-undo. Many “irreversible” calls have a reversible slice. Ask what the smallest test is before you assume you must commit fully.
Waiting for certainty on a reversible call. The extra confidence you buy is worth ~1.2 per point; the delay often costs more. On a two-way door, waiting usually loses value.
Naming the door to win the argument. Calling something reversible just to move fast — or irreversible to stall — is a tell. Classify honestly by the real cost of walking it back.

Worked example

An interviewer asks: “You have to decide whether to migrate the whole company to a new database this quarter. How do you approach it?” A strong answer starts by classifying the door. “A full cut-over is a one-way door — rolling back a live migration is expensive and risky, so I wouldn’t commit at 55%. But I don’t have to commit fully. I’d convert it into two-way doors: migrate one low-traffic service behind a flag, run both databases in parallel, and keep a reversible switch. If the pilot holds for two weeks, my confidence is 85%, not 55% — and only then do I walk through the one irreversible door of decommissioning the old system.” You’ve shown judgment about reversibility, not just enthusiasm.

Check yourself

Your team is at 65% on renaming a set of internal API endpoints behind a compatibility shim. What’s the read?

A VP wants to announce a hard public launch date on Friday. Confidence the date holds: 60%. Best move?

One-way vs two-way doors: deciding at 60% information

The idea

How it works

When to use it

Watch out for

Worked example