Skip to content
Anchor
All posts

Teardown: the decision that got made four times

Nobody in any of those rooms was wrong. The record just never survived long enough to stop the next argument.

Noa Bergman5 min read

Guest checkout was cut in March, reinstated in June, cut again in September, and shipped in January. Four decisions, four rooms, and at no point did anybody behave unreasonably.

This is a composite, and its shape is the point. Every one of those meetings reached a defensible conclusion from what was in the room. What was in the room was different every time.

The four rooms

  • March — cut it. The auth rebuild was already late and guest checkout sat on top of it.
  • June — bring it back. Support had a quarter of tickets from people who would not make an account.
  • September — cut it again. A security review flagged a cart-persistence rule nobody could produce in writing.
  • January — ship it. Someone found the March thread, and the rule had been agreed all along.

Ten months. The version that shipped in January was close enough to the March one that the engineer who built it asked, in the review, whether this had been specced before.

Nothing was lost. It was unfindable.

Every input existed the whole time. The support figures were in a dashboard. The security constraint was a comment on a closed ticket. The original reasoning was in a Slack channel that got archived in a reorg. None of it was deleted. All of it was out of reach at the exact moment somebody needed it.

The opposite of a documented decision is not an undocumented one. It is the same decision, made again, by people who do not know they are repeating themselves.

The tell

Re-litigation does not announce itself. Nobody says “we are about to redo March”. It arrives as a reasonable question from a new person, and the room answers it honestly from scratch, because answering from scratch is faster than proving what was decided before. That trade is rational every single time, and it is how ten months go.

What would have stopped it

Not a better meeting. June was a good meeting. What would have stopped it is any one person being able to ask why this was cut in March and get the answer with the thread attached — during the meeting, not two days after it.

The cost was never the ten months. It was that four separate groups did the same careful thinking, and only the last one got to keep it.

TeardownDecisionsContext

Stop writing PRDs without understanding the product.

Anchor reads your interviews, tickets and docs first — then writes, citing every line.