Clarify who owns a decision
A decision with no named owner cannot be closed; it can only be reversed. Naming the decider before building turns a future veto into an earlier approval.
Procedure
Name the decision in one line: what must be chosen, and by when.
Sort the people involved into three roles: Decider (exactly one name), Consulted (input needed), Informed (told after).
If there is no single decider, that is the problem; resolve it upward before building. Two owners means none.
Confirm with the decider directly, not through a chain: "you own the call on the schema, is that right?"
Get consent decisions before irreversible steps, not after; a review that can only rubber-stamp is not a decision.
Record the decider's name and the date beside the decision, so a later reversal points to a named moment.
When the decider is unreachable past the deadline, fall back to the stated default from the ask, in writing.
Pitfalls
Building the whole feature, then discovering the architecture owner disagrees.
"The team decided" with nobody accountable, so nobody can be wrong and nobody can be right.
Consulting everyone and calling the consensus the decision; consultation is not decision rights.
Confusing the requester with the decider; the asker may only be the messenger.
Deciding silently to avoid a meeting, then defending it as a fait accompli.
Verification
grep -nE '^Decider:' decisions.md # one named decider per decision, with a date
Report the decision, the named decider, and the deadline they must decide by.